jpad journal

AND option

OR option



X.-X. Zhang1, Y. Tian1, Z.-T. Wang1, Y.-H. Ma1, L. Tan1, J.-T. Yu2

1. Department of Neurology, Qingdao Municipal Hospital, Qingdao University, China; 2. Department of Neurology and Institute of Neurology, Huashan Hospital, Shanghai Medical College, Fudan University, Shanghai, China.

Corresponding Author: Dr. Jin-Tai Yu, Department of Neurology, Huashan Hospital, Fudan University, No. 12 Wulumuqi Road, Shanghai, China; or Dr. Lan Tan, Department of Neurology, Qingdao Municipal Hospital, Qingdao University, No.5 Donghai Middle Road, Qingdao, China, E-mail addresses:
(J.T. Yu); (L. Tan), Tel: +86 21 52888160; Fax: +86 21 62483421.

J Prev Alz Dis 2021;
Published online April 9, 2021,



Mild Alzheimer’s disease is the leading cause of dementia, accounting for 50-70% of cases. Alzheimer’s disease is an irreversible neurodegenerative disease, which affects daily life activities and social functioning. As life expectancy increases and demographic ageing occurs, the global prevalence of Alzheimer’s disease is expected to continue to rise especially in developing countries, leading to a costly burden of disease. Alzheimer’s disease is a complex and multifactorial disorder that is determined by the interaction of genetic susceptibility and environmental factors across the life course. Epidemiological studies have identified potential modifiable risk and protective factors for Alzheimer’s disease prevention. Moreover, Alzheimer’s disease is considered to start decades earlier before clinical symptoms occur, thus interventions targeting several risk factors in non-demented elderly people even middle-aged population might prevent or delay Alzheimer’s disease onset. Here, we provide an overview of current epidemiological advances related to Alzheimer’s disease modifiable risk factors, highlighting the concept of early prevention.

Key words: Alzheimer’s disease, epidemiology, modifiable risk factors, prevention.



Alzheimer’s disease (AD), as the most prevalent cause of dementia, is defined by deterioration in cognition, function and behavior, which typically begins in memory loss about recent events (1). The decisive pathological features in AD patients’ brain tissues are raised levels of both amyloid-β (Aβ) composing of extracellular senile plaques and hyperphosphorylated tau (p-tau) aggregating intracellularly as neurofibrillary tangles (NFTs) (2). About 50 million people are living with dementia around the world, due to the aging population, the number of patients is predicted to triple by 2050, which increases the risk of disability, burden of illness and health care costs (3). Moreover, current treatment strategies only ameliorate symptoms and there is no effective cure for AD. However, AD has a long prodromal period during which early prevention appears to be particularly important to slow down the progression of AD. Therefore, epidemiological investigations are essential to identify risk and protective factors that strongly influence cognitive status. In fact, one-third of AD cases worldwide are attributable to underlying modifiable risk factors (4), which might modulate an individual’s risk of developing AD. In this review, we classified these factors into several categories including psychosocial factors, pre-existing diseases, lifestyles and others, exploring the potential effects on cognition to provide better implications for AD prevention.


Descriptive epidemiology of Alzheimer’s disease

The number of dementia patients is projected to reach 152 million by mid-century worldwide, with the greatest increase expected in low-and middle-income countries (3). According to 2020 Alzheimer’s disease facts and figures, the number of AD patients (≥ 65 years) might increase greatly from 5.8 million to 13.8 million by 2050 in America (5). The obviously increased AD prevalence was found in community-dwelling investigations of Japan and China over the last few decades (6, 7). Particularly, age-specific global prevalence in women was 1.17 times larger than in men and the age-standardized mortality rate of women was also higher than men, suggesting the longer lifespan was not the only determinant of the women dominance (8). In addition, death tolls with AD increased 146.2% from 2000 to 2018 and AD became the fifth-largest cause of death in American old people (5). Notably, caregivers would experience more mental stresses and negative emotional influences (5). Therefore, the social and family burden of caring for AD population will be huge and unsustainable.
Wellbeing is the aim of much of AD care. The AD patients would have perplexing problems and symptoms in many domains. And some epidemiological investigations have provided robust evidence that behavioral and environmental factors have key roles in disease pathogenesis and progression. Especially, pre-existing disease is more common in AD patients than others of the same age, it is essential to keep physically healthy to protect the cognition. In addition, many risk factors could contribute to the development of AD and also be regarded as the symptoms of AD simultaneously, the reverse causality might account for this. Therefore, the accurate diagnosis is so important for those individuals suffering from cognitive dysfunction. Though AD is indicated by Aβ and tau biomarkers, some cognitively normal individuals having only these biomarkers never develop AD (9), it means that the pre-symptomatic diagnosis is more difficult to obtain. Future challenges would include discovering less-invasive and more-sensitive biomarkers or methods that can also be used for early screening and diagnosis purposes. Anyway, evidence-based prevention strategies, in line with the potential link between modifiable risk factors and late-onset AD, need to be explored in future studies.


Putative modifiable risk factors and prevention for late-onset Alzheimer’s disease

Evidence from observational studies has accumulated during the past few years and shown several potentially modifiable risk factors (Figure 1), concerning AD prevention some potential feasible suggestions are provided (Figure 2).

Figure 1. Potential modifiable risk factors for Alzheimer’s disease

Risk factors mainly included pre-existing diseases, unhealthy lifestyles and environmental exposures, while some factors concerning psychosocial conditions as well as healthy lifestyles might protect against AD. In addition, some factors appeared to be risk factors as well as symptoms of AD, possibly due to the reverse causality, these factors were highlighted in bold. Abbreviation: BP = blood pressure, DASH = Dietary Approach to Stop Hypertension, MIND = Mediterranean-DASH diet Intervention for Neurodegeneration Delay, PUFA = polyunsaturated fatty acid, HDL- cholesterol = high-density lipoprotein cholesterol.

Figure 2. Implications for preventing Alzheimer’s disease and slowing its progression

It is imperative to increase the cognitive reserve mainly via enhancing education attainment and promoting social contact. Additionally, good conditions of body health and healthy lifestyles as well as reducing environmental exposures might be favorable to reduce the neuropathological damage for AD prevention.


Psychosocial factors

Some prospective cohorts have studied whether these psychosocial factors might affect the cognition specifically (Supplemental Table 1).

Educational attainment

Continuing adult education could benefit language processing and intellectual capacities (10), long-term education might also have favorable influences on total brain volume such as elevated cortical surface area and thickness in the prodromal stage of AD (11, 12). Mendelian randomization studies confirmed causal associations between education and reduced AD risk [ odds ratio (OR) 0.64, 95% confidence intervals (CI) 0.56-0.74] as well as delayed AD onset [hazard ratio (HR) 0.76, 95% CI 0.67-0.85] (12), which could be mediated by intelligence in part (13). These causal relationships should be delineated profoundly in future studies.

Cognitive activity and bilingualism

Active engagements in cognitive activity were likely to have 46% reduction of AD risk in the Swedish study (14). Specifically, reduced decline in memory ability and cognitive speed among those playing more analog games was reported in the Lothian cohort study (15), which could be explained in part by intensive connectivity between the hippocampus and superior frontal cortex (16). Bilingualism, as a constituent of cognitive reserve, could enhance neural efficiency by increased functional connectivity in the frontoparietal control network for executive control and the default mode network for behavior control (17), certainly, bilinguals had stronger executive and visual-spatial functions than monolinguals (18). Specifically, compared with those only speaking Cantonese or Mandarin, lifelong bilinguals manifested the first symptoms of AD markedly later (19).
Higher educational attainment could delay the onset of AD by building cognitive reserve and brain volume (11). Similarly, cognitive activity and bilingualism were also beneficial to preserve the healthy cognitive functioning (14, 15, 18). Therefore, it is needed to increase access to education especially language and promote cognitive activities in the general population to protect against AD.

Social engagement

Frequent social contact with friends was related to a modestly decrease in dementia risk, owing to the creation of cognitive reserve at the early-stage via regular heathy social engagement (20). Particularly, community cultural activities might also confer benefits to the whole cognitive function (21). A healthy couple relationship might exert greater protective effects on cognition, whereas the widowed, especially those APOE ε4 carriers, had higher risk of AD (OR 7.67, 95% CI 1.6-40.0) than those married people (22). Indeed, among cognitively intact older people, widowed individuals were at higher likelihood of suffering serious Aβ-related cognitive deterioration (23). More social engagements could exercise memory and language which might further increase cognitive reserve (20). Of note, the widowed or the singled seemed to have less communication with others, it is encouraging to establish healthy social relationships and engage more regular social activities.

Depression and stress

Depression was a significant risk for AD (OR 1.65, 95% CI 1.42-1.92) (24). The finding of Australian study further proposed that the dementia odds ratios of mild, moderate to severe depressive symptoms were 1.2 (95% CI 1.0-1.4), 1.7 (95% CI 1.4-2.2) and 2.1 (95% CI 1.4-3.2). In addition, treatment with the citalopram for more than 4 years was strongly linked to a 3-year delay in progression from mild cognitive impairment (MCI) to AD (25). Notably, citalopram could reduce cerebrospinal fluid (CSF) amyloid plaque in experimental transgenic AD mice and healthy volunteers (26), conferring great significance to clinical practice. Besides, greater microglial activation, a symbol of neurodegenerative inflammation, existed in depressive patients not receiving medications (27), thus, reasonable treatment of depression might mean significant to prevent from neurodegeneration. However, depression and AD might have common causes such as inflammation and neurodegeneration, or similar symptoms in the prodromal stage of AD (28), thus depressive symptoms appeared to be a prodromal marker of AD rather than a causal risk factor. Further, the causal association of depression with AD was not supported in a mendelian randomization study (12), therefore, this association should be explored more in detail in future studies.
Higher risk of AD incidence (HR 1.36, 95% CI 1.12-1.67) was found among those individuals with stress-related disorders (29). Chronic work-related stress was also an emerging risk factor for AD (30). Notably, oxidative stress and neuroinflammation could be induced by stress, which could further promote the production of amyloid plaques (31). Regulating stress and adjusting emotions would ameliorate detrimental psychological effects on cognition.

Pre-existing diseases

We concluded detailed discoveries of prospective studies concerning the relationships between all these pre-existing diseases and AD development (Supplemental Table 2).

Diabetes, hypertension and dyslipidemia

Evidence supporting the higher risk of affecting with AD among the diabetics [relative risks (RR) 1.43, 95% CI 1.25-1.62] was based on 24 longitudinal studies (32). Specifically, a steeper fall in perceptual speed and verbal abilities among the diabetics was also found in the Swedish study (33). Interestingly, the interaction of Aβ with islet amyloid polypeptide exacerbating AD pathogenesis might explain the molecular association of diabetes with AD (34).The diabetics without proper treatment had higher CSF p-tau level than those using antidiabetic drugs or euglycemic adults, suggesting tau pathology might be ameliorated with the antidiabetic drugs (35). Moreover, metformin use and treatment with anti-diabetic drugs might slow down the development of cognitive decline and reduce the risk of dementia (36, 37).
A remarkably increase in AD risk (HR 1.73, 95% CI 1.02-2.94) among those with midlife and late-life high systolic pressure has also been reported in the Framingham Offspring Study (38). However, decreased systolic pressure from mid-to late life also contributed to higher risk of AD (HR 2.12, 95% CI 1.12-4.00) (38). Individuals with large blood pressure variability exhibited a larger than 2-fold risk of AD (39), indicating excessive blood pressure variation might partially be responsible for AD deterioration. Fortunately, the AD risk was decreased (RR 0.78, 95% CI 0.66-0.91) because of the treatment of antihypertensives, longer use of these drugs might have significantly protective effects on cognition (40).
Blood lipids such as total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C) might play important parts in the process of AD. The Three-City Study reported a significant increase in AD risk among those with hypercholesterolemia, particularly, this association would be weakened after adjusting for APOE ε4 (41). In addition, the cholesterol was divided into two subtypes including HDL-C and non-HDL-C in the Adult Changes in Thought (ACT) Study, the finding revealed a U-shaped association of non-HDL-C level with risk of AD among 60-69 aged individuals, compared with 160 mg/dl non-HDL-C, the HR was 1.29 (95% CI 1.04-1.61) at 120 mg/dl and 1.16 (95% CI 1.01-1.33) at 210 mg/dl (42). Notably, hypercholesterolemia could compromise the integrity of the blood-brain barrier, increase Aβ deposition, as well as cause neuroinflammation, all of which could exacerbate the development of AD (43). In addition, increased LDL-C levels were associated with higher AD risk (41), while HDL-C was likely to be a protective factor against AD (44). There existed no significant association between AD risk and TG level (41). Encouragingly, a meta-analysis indicated that uses of stain drugs were likely to lower the risk of AD (RR 0.86, 95% CI 0.80-0.92) among those with hyperlipidemia in prospective studies, but these beneficial effects were not stronger in randomized controlled trials (40).
Young adults had better stay away from diabetes, hypertension, hypercholesterolemia via healthier lifestyles. Patients could manage these diseases via regular uses of pharmaceuticals. Taking regular antidiabetics was associated with less tau dysfunction delaying the progression of AD (35). The antihypertensive drugs and stain drugs might be favorable to lower the risk of AD among those patients (40, 45). More studies are required to investigate whether particular medications would effectively lower the risk of developing AD.


Consistent evidences have proposed a higher risk of developing AD among middle-aged patients with raised body mass index (BMI) whereas overweight or obesity in late-life might have protective effects on cognition (46, 47). Particularly the HR for AD was 0.89 (95% CI 0.81-0.98) among those with high late-life BMI, in contrast, there was 20% increased risk for AD (95% CI 1.09-1.33) among those with greater loss of BMI from midlife to late-life (46). Consistently, the obese elderly tended to have less Aβ load and larger hippocampus volume (47). The underlying mechanism might be partially attributed to the increased leptin produced mainly by adipocytes, leptin could robustly facilitate the neurogenesis of hippocampus in AD mouse models (48). Of note, we should emphasize the role of age in evaluating the association of obesity with AD. Young adults should maintain or lose weight under the healthy range (18.5-24.9 kg/m2) while older individuals should avoid greater weight loss otherwise attention should be focused on this situation.

Cardiovascular diseases

The HR of AD following atrial fibrillation was 1.31 (95% CI 1.20-1.43) in the Korean study (49). In addition, mild reductions in cardiac index were also related to a marked increase in AD risk (HR 2.87, 95% CI 1.21-6.80) over a long period of 7.7 years (50). Cerebral hypoperfusion, caused by lower cardiac output in atrial fibrillation patients, could explain the deteriorations of cortical atrophy and neuropathological features (51). Encouragingly, individuals with atrial fibrillation undergoing oral anticoagulants were at a lower risk of developing AD (HR 0.61, 95% CI 0.54-0.68) (49). Interestingly, compared with medical therapy, the risk of AD was decreased among atrial fibrillation patients treated with catheter ablation (HR 0.77, 95% CI 0.61-0.97) (52).
About 59% elevated risk of AD among stroke patients was well documented in a meta-analysis incorporating six studies (53).Consequently, effective strategies to reduce the extent of brain injury after stroke may help delay or prevent the progression of AD. A markedly increase in AD risk among individuals with multiple cerebral microbleeds has been reported in the Rotterdam Study (HR 2.10, 95% CI 1.21-3.64) (54). Meanwhile, severe cerebral atherosclerosis (OR 1.33, 95% CI 1.11-1.58) and arteriolosclerosis (OR 1.20, 95% CI 1.04-1.40) were also regarded as strong risk factors for AD (55). Neurovascular unit disorder, due to cerebrovascular diseases, could cause decreased cerebral blood flow, induce blood-brain barrier disruptions as well as selective brain atrophies, all of which could result in direct damage to neurons and trigger Aβ accumulation indirectly (56). Thus, better understanding of these relationships may acquire cognitive benefits from appropriate treatment.
Early screening and intervention of vascular risks as well as maintaining good cardiovascular conditions should become the top priority for AD prevention. Pharmaceuticals uses like oral anticoagulants and catheter ablation would significantly lower the risk of AD among atrial fibrillation patients (49, 52).

Traumatic brain injury

Participants exposed to traumatic brain injury (TBI) had increased risk of developing AD in the Danish study (HR 1.16, 95% CI 1.12-1.22) and in a Swedish study (OR 1.58, 95% CI 1.49- 1.69), which could be strengthened by some detailed discoveries of TBI such as severe and multiple TBIs, the first few months since the trauma occurrence, younger individuals with the injury as well as TBI involving the skull or spine (57, 58). Particularly, athletes experiencing many years of head injury were more susceptible to die from AD according to their death certificates (59). Furthermore, among the American military veterans, women having TBI were at higher risk of developing dementia (60). Encouragingly, stain medication especially the rosuvastatin, owing to the potential neuroprotective benefits, might reduce the dementia risk among those individuals having a concussion (45). Thus, more studies are required to focus on the risk of TBI individuals to have AD and formulate therapeutic strategies to mitigate the risk and impact of AD. Considering the deleterious effects of TBI on cognition, the public had better take measures to protect the head properly from injuries when engaging in dangerous activities or work.


A meta-analysis incorporating 5 prospective studies found a linear dose-response relationship between blood homocysteine (Hcy) levels and risk of AD (HR 1.15, 95% CI 1.04-1.26, per 5 µmol/L increment) (61). However, in some findings of longitudinal studies such dose-dependent relationship only existed in the range of high serum Hcy concentrations (about ≥10 µmol/l) (62). These inconsistent exposure-response associations should be further assessed in large-scale prospective studies. In addition, the elevation of Aβ deposition and tau hyperphosphorylation could be modulated by high Hcy levels via γ-secretase pathway and cdk5 kinase in mouse models (63). Additionally, hyperhomocysteinemia could be alleviated via folic acid supplementation (64), improving total homocysteine metabolism may also represent a viable strategy for AD prevention.

Hearing loss and oral diseases

And in a case-control study the OR was 1.39 for AD (95% CI 1.05-1.84) following hearing loss (65). Owing to the awful listening conditions, individuals might have difficulty in understanding speech and experience communication disorders even social isolation, contributing to reduced cognitive stimulation from the acoustic environment (66), which could aggravate cognitive impairment mediated by accelerated brain atrophy. Indeed, individuals with midlife hearing impairment tended to have prominent temporal lobe volume loss (67). Encouragingly, hearing aids and cochlear implants could mitigate some worse listening status, slowing down the rate of cognitive decline (68), early screening and correction of hearing loss might hold significant influence on AD prevention.
Oral diseases particularly tooth loss and chronic periodontitis were great concerns for cognitive dysfunction, probably mediated by local and systemic inflammatory responses (69), tooth loss was a strong risk factor for AD in the Hisayama Study (69). Specifically, individuals with chronic periodontitis were more susceptible to AD (HR 1.707, 95% CI 1.152-2.528) (70), periodontitis might induce peripheral inflammation through byperiodontal pathological bacteria directly or proinflammatory cytokines indirectly (71). Much attention should be attached to oral care especially in developing counties to prevent AD.


We identified plenty of prospective studies regarding lifestyles related to the risk of cognitive decline and AD (Supplemental Table 3, Supplemental Table 4).

Physical activity

Higher participation in daily physical activity was related to about half decreased risk of AD in the Hisayama Study (72). In addition, cognitive benefits of regular resistance exercise and choreographic intervention could delay the neurodegenerative process especially in those domains related to the conversion to dementia (73, 74). Besides, aerobic exercise was also beneficial to some cognitive domains including executive function and oral fluency (75), possibly due to the protective effects on hippocampus volume and neuronal health (76). After one-year aerobic exercise training, improvements of cardiorespiratory function particularly cerebral perfusion and memory ability were found in a prospective study (77). The neurotrophic effects of active exercise are needed to be further investigated, especially its type, intensity, duration and timing might have greater implications on AD prevention. However, in the Whitehall II study physical activities appeared to have no neuroprotective effects on cognitive functioning, individuals in the preclinical stage of AD tended to have lower physical activity levels than the healthy elderly (78), in other words, reverse causality might account for the relationship between active exercise and reduced risk of AD. It still remains controversial whether reverse causality could explain the favorable effects of physical activity on cognition.

Sleep disturbances

Sleep disturbances (insomnia and sleep disordered breathing) were risk factors for AD, and those with sleep disorders were more likely to experience more accumulation of neurotoxic substances as a result of the decreased metabolite clearance ability (79). There was a 66% (95% CI 1.03-2.68) increased AD risk in participants having severe obstructive sleep apnea (≥30 vs.<5 apnea-hypopnea events/hour) (80). Additionally, individuals with longer sleep length (> 9 hours) showed a greater than 2-fold risk of AD (81), the RR of AD following habitual shorter sleep duration was 1.25 (95% CI 0.88-1.76) (80). Moreover, both short duration (≤6 hours) and long duration (≥8 hours) all had detrimental effects on cognitive function, it further offered a V-shaped association of daily sleep duration with cognitive decline and subsequent risk for dementia (82). Notably, greater amyloid deposition was more common among those who had insufficient or excessive nocturnal sleep time (83). And there existed a causal relationship between sleep length and elevated cortical thickness in a mendelian randomization study (12). There may also exist a bidirectional relationship between sleep dysregulation and AD pathology, sleep disorders could stimulate the accumulation of Aβ and tau, meanwhile the enhanced aggregation of Aβ and tau may exacerbate the progression of sleep disturbances (84). More long-term longitudinal studies are needed to further explore the potential role of sleep dysregulation as a biomarker of AD and the potential bidirectional relationship. High quality sleep is extremely important in maintaining cognition, when having sleep problems patients should consult a doctor and receive therapy in time.


Smoking increased the risk of AD (RR 1.40, 95% CI 1.13-1.73), which was also prominent among non-APOE ε4 carriers in a meta-analysis including 37 longitudinal studies (85). Of note, smoking initiation was associated with lower cortical thickness (12). Smoking-related cerebral oxidative stress might facilitate the production of amyloid or tau pathology (86). In contrast, never-smokers had 18% risk reduction of AD than continual smokers (87), indicating that early smoking cessation would confer greater benefits on cognition.

Alcohol consumption

The effects of alcohol intake on cognitive function may remain controversial in many epidemiological findings. The Nord-Trøndelag Health study supported a 47% elevated AD risk (95% CI 1.00-2.16) among frequent alcohol drinkers (≥5 times/two weeks) vs. infrequently drinkers (1-4 times) (88). This relationship was thought to be J-shaped, suitable alcohol intake (<12.0 g/day) could protect against dementia (89, 90), similar with the relationship between coffee intake and AD. It is not advisable to take excessive drinking in daily life, as for the heavy alcohol drinkers, extracellular cold-inducible RNA-binding protein (eCIRP) might mediate tau phosphorylation, leading to the progression of alcohol-induced AD (91). After matured hop bitter acid supplementation from beer, improvement of cognitive status was confirmed in a randomized trial (92). Particularly, alcohol from wine appeared to be stronger inversely related to the risk of dementia (89, 90), polyphenolic and antioxidant contents in wine showed greater protective effects against neurodegeneration (93). More primary studies should be warranted to clarify the underlying mechanism explaining the AD risk related to alcohol intake. The only established causal relationship between earlier AD onset and alcohol consumption was found in a mendelian randomization study (94), which indicated that moderate alcohol intake might be harmful to the brain health but not beneficial. This study also stressed that potential confounding factors should be seriously taken into consideration, especially the survival bias (94), therefore future studies should delineate this relationship thoroughly and precisely. Coffee and tea A J-shaped association of coffee intake with AD was proposed, low (1-2 cups/day) but not high (>3 cups/day) coffee intake was related to a 18% reduction of AD risk (RR 0.82 , 95% CI 0.71-0.94, vs. <1 cup/day) (95), which was more pronounced among women (96). However, this gender characteristic was different from tea intake. The protective effects of green tea on cognition were more prominent in men (97, 98). Some neuroprotective components may exist in the coffee or green tea drinking like caffeine and L-theanine (99), anti-amyloid effects of green tea might protect against AD mainly including inhibition of Aβ aggregation and reduction of Aβ-induced oxidative stress (100). Moderate coffee consumption and green tea drinking should be encouraged to the public. There was reduced level of CSF total-tau protein among frequent green tea consumers, probably owing to improved abnormal tau metabolism (98). However, the black tea and oolong tea did not show cognitive benefits in the overall elderly Han study population (97).


Three dietary patterns including the Mediterranean diet (HR 0.46, 95% CI 0.26-0.79), the DASH (Dietary Approaches to Stop Hypertension) diet (HR 0.61, 95% CI 0.38-0.97) and the MIND (Mediterranean-DASH Intervention for Neurodegenerative Delay) diet (HR 0.47, 95% CI 0.26-0.76), were all inversely associated with the risk of AD (101), the protective effects of these dietary patterns on cognition might be attributed to anti-oxidant, anti-inflammatory, and anti-diabetic effects and enough mono-/poly-unsaturated fats (102). More interestingly, the Three-city study proposed the concept of novel diet pattern, a more diverse diet including vegetables, fresh fruits and seafood, might be particularly beneficial to the cognitive function (103). Particularly, ketogenic diets could reduce the AD risk via altering gut mycobiome (104), while following higher glycemic load intake the HR increased by 27% for AD (105). Additionally, a meta-analysis incorporating 21 cohort studies proposed that higher intake of fish (RR 0.93, 95% CI 0.90-0.95) and marine-derived dietary docosahexaenoic acid (RR 0.63, 95% CI 0.51-0.76) could protect against the risk of developing AD (106). Besides, habitual intake of seafood was associated with lower burden of AD brain pathology among APOE ε4 carriers, which was not affected by higher brain levels of mercury (107).
Severe 25-Hydroxyvitamin D [25(OH)D] deficiency (<25 nmol/L) was related to elevated risk of AD (HR 2.22, 95% CI 1.02-4.83), as was 25(OH)D inadequacy (25-50 nmol/L) (HR 1.69, 95% CI 1.06-2.69), using the level of more than 50 nmol/L as the refence category (108). Persons in early adulthood taking more B vitamins including niacin, folate, vitamin B-6, and vitamin B-12 were likely to acquire better cognitive performance in late life (109). In addition, the potential cognitive benefits of vitamin C were more remarkable among women APOE 4-carriers while higher blood vitamin E level might hold such benefits among APOE 4-negative men (110). Higher flavonoids intake from daily food might reduce the risk of AD incidence by 38% (95% CI 0.39-0.98), which was independent of other lifestyle factors and cardiovascular related diseases (111).Therefore, further studies are required to delineate the potential biologic explanations, and more mendelian randomization studies are needed to elucidate whether these associations of micronutrients intake with the cognition are causal. It is necessary to take certain vitamins and other micronutrients from daily diets properly.



Current or former nonsteroidal anti-inflammatory drugs (NSAIDs) exposure was associated with a 19% reduction in AD risk (95% CI 0.70-0.94) (112). However, in a randomized placebo-controlled trial, no robust evidence supported that aspirin could effectively lower the risk of AD (113), aligned with the meta-analysis conducted by Veronese et al (114). In this meta-analysis, low-dose aspirin did not seem to improve the cognitive function (114). Therefore, it still remains controversial whether NSAIDs would protect against AD.
The adverse impact of anticholinergic medications on cognition was underscored. Over a long period of 4 years, excess use of anticholinergics was considered as a risk for AD (HR 1.63, 95% CI 1.24-2.14) (115). Minimizing anticholinergic use over time might be important to preserve the cognition.


We identified several longitudinal cohorts studying some environment factors associated with the risk of cognitive decline and AD (Supplemental Table 3). Misfolding and abnormal aggregation of p-tau and Aβ were found in the brainstem of children and young adults exposed to Mexico City’s air pollution (116). In a study from Northern Sweden, there was a 38% elevated risk for AD among those exposed to residential traffic-related air pollution (95% CI 0.87-2.19) when comparing the highest with the lowest quartile of NOx (≥26 vs. ≤9 μg/m3) (117). Indeed, Particulate Matter2.5 (PM2.5) relevant with gray matter atrophy indicated higher risk of AD (HR 1.24, 95% CI 1.14-1.34) in older women (118). The adverse effects of air pollutants on cognition might be amplified by cardiovascular diseases including heart failure and ischemic heart disease (120). Crucially, long-term air pollution might accelerate the progression of neurodegeneration via vascular disease, Aβ deposition and neuroinflammation (116, 120). Therefore, reducing the exposures of air pollution seems particularly important for AD prevention. More studies are required to investigate the effects of environmental exposures on AD and the potential mechanisms underlying these relationships.



Many longitudinal studies have identified various risk and protective factors for AD, including some that could be targeted to reduce risk of AD or delay the onset of AD, suitable preventions might help slow down the progress of AD. More policies targeting education popularization and social or cognitive activities promotion should be put forward among the public. Managing the pre-existing disease reasonably and maintaining daily healthy lifestyles would protect against AD. Additionally, environment protection especially targeting air pollutants would be of great importance to AD prevention. If possible, more studies should focus on individuals at high risk of AD or in the prodromal stage of AD, among whom daily preventions and neuroprotective interventions are likely to exert greater favorable effects.


Acknowledgements: This study was supported by grants from the National Key R&D Program of China (2018YFC1314700), Shanghai Municipal Science and Technology Major Project (No.2018SHZDZX01) and ZHANGJIANG LAB, Tianqiao and Chrissy Chen Institute, and the State Key Laboratory of Neurobiology and Frontiers Center for Brain Science of Ministry of Education, Fudan University.

Conflicting interests: None.

Open Access: This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits use, duplication, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.



1. Scheltens P, Blennow K, Breteler MM, et al. Alzheimer’s disease. Lancet (London, England) 2016;388: 505-517.
2. Winblad B, Amouyel P, Andrieu S, et al. Defeating Alzheimer’s disease and other dementias: a priority for European science and society. The Lancet. Neurology 2016;15: 455-532.
3. C. P World Alzheimer report 2018. London: Alzheimer’s Disease International, 2018.
4. Norton S, Matthews FE, Barnes DE, Yaffe K, Brayne C Potential for primary prevention of Alzheimer’s disease: an analysis of population-based data. The Lancet. Neurology 2014;13: 788-794.
5. 2020 Alzheimer’s disease facts and figures. Alzheimer’s & Dementia 2020;16: 391-460.
6. Ohara T, Hata J, Yoshida D, et al. Trends in dementia prevalence, incidence, and survival rate in a Japanese community. Neurology 2017;88: 1925-1932.
7. Chan KY, Wang W, Wu JJ, et al. Epidemiology of Alzheimer’s disease and other forms of dementia in China, 1990-2010: a systematic review and analysis. Lancet (London, England) 2013;381: 2016-2023.
8. Global, regional, and national burden of Alzheimer’s disease and other dementias, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. The Lancet. Neurology 2019;18: 88-106.
9. Livingston G, Huntley J, Sommerlad A, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet (London, England) 2020;396: 413-446.
10. Matyas N, Keser Aschenberger F, Wagner G, et al. Continuing education for the prevention of mild cognitive impairment and Alzheimer’s-type dementia: a systematic review and overview of systematic reviews. BMJ open 2019;9: e027719.
11. Wada M, Noda Y, Shinagawa S, et al. Effect of Education on Alzheimer’s Disease-Related Neuroimaging Biomarkers in Healthy Controls, and Participants with Mild Cognitive Impairment and Alzheimer’s Disease: A Cross-Sectional Study. Journal of Alzheimer’s disease : JAD 2018;63: 861-869.
12. Andrews SJ, Fulton-Howard B, O’Reilly P, Marcora E, Goate AM Causal Associations Between Modifiable Risk Factors and the Alzheimer’s Phenome. Annals of neurology 2020.
13. Anderson EL, Howe LD, Wade KH, et al. Education, intelligence and Alzheimer’s disease: evidence from a multivariable two-sample Mendelian randomization study. International journal of epidemiology 2020.
14. Najar J, Östling S, Gudmundsson P, et al. Cognitive and physical activity and dementia: A 44-year longitudinal population study of women. Neurology 2019;92: e1322-e1330.
15. Altschul DM, Deary IJ Playing Analog Games Is Associated With Reduced Declines in Cognitive Function: A 68-Year Longitudinal Cohort Study. The journals of gerontology. Series B, Psychological sciences and social sciences 2020;75: 474-482.
16. Suo C, Singh MF, Gates N, et al. Therapeutically relevant structural and functional mechanisms triggered by physical and cognitive exercise. Molecular psychiatry 2016;21: 1633-1642.
17. Grady CL, Luk G, Craik FIM, Bialystok E Brain network activity in monolingual and bilingual older adults. Neuropsychologia 2015;66: 170-181.
18. Estanga A, Ecay-Torres M, Ibañez A, et al. Beneficial effect of bilingualism on Alzheimer’s disease CSF biomarkers and cognition. Neurobiology of aging 2017;50: 144-151.
19. Zheng Y, Wu Q, Su F, et al. The Protective Effect of Cantonese/Mandarin Bilingualism on the Onset of Alzheimer Disease. Dementia and geriatric cognitive disorders 2018;45: 210-219.
20. Sommerlad A, Sabia S, Singh-Manoux A, Lewis G, Livingston G Association of social contact with dementia and cognition: 28-year follow-up of the Whitehall II cohort study. PLoS medicine 2019;16: e1002862.
21. Fancourt D, Steptoe A, Cadar D Community engagement and dementia risk: time-to-event analyses from a national cohort study. Journal of epidemiology and community health 2020;74: 71-77.
22. Håkansson K, Rovio S, Helkala E-L, et al. Association between mid-life marital status and cognitive function in later life: population based cohort study. BMJ (Clinical research ed.) 2009;339: b2462.
23. Biddle KD, Jacobs HIL, d’Oleire Uquillas F, et al. Associations of Widowhood and beta-Amyloid With Cognitive Decline in Cognitively Unimpaired Older Adults. JAMA network open 2020;3: e200121.
24. Diniz BS, Butters MA, Albert SM, Dew MA, Reynolds CF Late-life depression and risk of vascular dementia and Alzheimer’s disease: systematic review and meta-analysis of community-based cohort studies. The British journal of psychiatry : the journal of mental science 2013;202: 329-335.
25. Bartels C, Wagner M, Wolfsgruber S, Ehrenreich H, Schneider A Impact of SSRI Therapy on Risk of Conversion From Mild Cognitive Impairment to Alzheimer’s Dementia in Individuals With Previous Depression. The American journal of psychiatry 2018;175: 232-241.
26. Sheline YI, West T, Yarasheski K, et al. An antidepressant decreases CSF Aβ production in healthy individuals and in transgenic AD mice. Science translational medicine 2014;6: 236re234.
27. Setiawan E, Attwells S, Wilson AA, et al. Association of translocator protein total distribution volume with duration of untreated major depressive disorder: a cross-sectional study. The lancet. Psychiatry 2018;5: 339-347.
28. Singh-Manoux A, Dugravot A, Fournier A, et al. Trajectories of Depressive Symptoms Before Diagnosis of Dementia: A 28-Year Follow-up Study. JAMA psychiatry 2017;74: 712-718.
29. Song H, Sieurin J, Wirdefeldt K, et al. Association of Stress-Related Disorders With Subsequent Neurodegenerative Diseases. JAMA neurology 2020.
30. Sindi S, Hagman G, Hakansson K, et al. Midlife Work-Related Stress Increases Dementia Risk in Later Life: The CAIDE 30-Year Study. The journals of gerontology. Series B, Psychological sciences and social sciences 2017;72: 1044-1053.
31. Greenberg MS, Tanev K, Marin M-F, Pitman RK Stress, PTSD, and dementia. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2014;10: S155-S165.
32. Xue M, Xu W, Ou YN, et al. Diabetes mellitus and risks of cognitive impairment and dementia: A systematic review and meta-analysis of 144 prospective studies. Ageing research reviews 2019;55: 100944.
33. Marseglia A, Dahl Aslan AK, Fratiglioni L, et al. Cognitive Trajectories of Older Adults With Prediabetes and Diabetes: A Population-Based Cohort Study. The journals of gerontology. Series A, Biological sciences and medical sciences 2018;73: 400-406.
34. Moreno-Gonzalez I, Edwards Iii G, Salvadores N, et al. Molecular interaction between type 2 diabetes and Alzheimer’s disease through cross-seeding of protein misfolding. Molecular psychiatry 2017;22: 1327-1334.
35. McIntosh EC, Nation DA Importance of Treatment Status in Links Between Type 2 Diabetes and Alzheimer’s Disease. Diabetes care 2019;42: 972-979.
36. Samaras K, Makkar S, Crawford JD, et al. Metformin Use Is Associated With Slowed Cognitive Decline and Reduced Incident Dementia in Older Adults With Type 2 Diabetes: The Sydney Memory and Ageing Study. Diabetes care 2020.
37. Wium-Andersen IK, Osler M, Jørgensen MB, Rungby J, Wium-Andersen MK Antidiabetic medication and risk of dementia in patients with type 2 diabetes: a nested case-control study. Eur J Endocrinol 2019;181: 499-507.
38. McGrath ER, Beiser AS, DeCarli C, et al. Blood pressure from mid- to late life and risk of incident dementia. Neurology 2017;89: 2447-2454.
39. Oishi E, Ohara T, Sakata S, et al. Day-to-Day Blood Pressure Variability and Risk of Dementia in a General Japanese Elderly Population: The Hisayama Study. Circulation 2017;136: 516-525.
40. Larsson SC, Markus HS Does Treating Vascular Risk Factors Prevent Dementia and Alzheimer’s Disease? A Systematic Review and Meta-Analysis. Journal of Alzheimer’s disease : JAD 2018;64: 657-668.
41. Schilling S, Tzourio C, Soumare A, et al. Differential associations of plasma lipids with incident dementia and dementia subtypes in the 3C Study: A longitudinal, population-based prospective cohort study. PLoS medicine 2017;14: e1002265.
42. Marcum ZA, Walker R, Bobb JF, et al. Serum Cholesterol and Incident Alzheimer’s Disease: Findings from the Adult Changes in Thought Study. Journal of the American Geriatrics Society 2018;66: 2344-2352.
43. Chew H, Solomon VA, Fonteh AN Involvement of Lipids in Alzheimer’s Disease Pathology and Potential Therapies. Front Physiol 2020;11: 598.
44. Svensson T, Sawada N, Mimura M, et al. The association between midlife serum high-density lipoprotein and mild cognitive impairment and dementia after 19 years of follow-up. Translational psychiatry 2019;9: 26.
45. Redelmeier DA, Manzoor F, Thiruchelvam D Association Between Statin Use and Risk of Dementia After a Concussion. JAMA neurology 2019.
46. Tolppanen AM, Ngandu T, Kareholt I, et al. Midlife and late-life body mass index and late-life dementia: results from a prospective population-based cohort. Journal of Alzheimer’s disease : JAD 2014;38: 201-209.
47. Sun Z, Wang Z-T, Sun F-R, et al. Late-life obesity is a protective factor for prodromal Alzheimer’s disease: a longitudinal study. Aging 2020;12: 2005-2017.
48. Pérez-González R, Antequera D, Vargas T, et al. Leptin induces proliferation of neuronal progenitors and neuroprotection in a mouse model of Alzheimer’s disease. Journal of Alzheimer’s disease : JAD 2011;24 Suppl 2: 17-25.
49. Kim D, Yang P-S, Yu HT, et al. Risk of dementia in stroke-free patients diagnosed with atrial fibrillation: data from a population-based cohort. European heart journal 2019;40: 2313-2323.
50. Jefferson AL, Beiser AS, Himali JJ, et al. Low cardiac index is associated with incident dementia and Alzheimer disease: the Framingham Heart Study. Circulation 2015;131: 1333-1339.
51. Austin BP, Nair VA, Meier TB, et al. Effects of hypoperfusion in Alzheimer’s disease. Journal of Alzheimer’s disease : JAD 2011;26 Suppl 3: 123-133.
52. Kim D, Yang P-S, Sung J-H, et al. Less dementia after catheter ablation for atrial fibrillation: a nationwide cohort study. European heart journal 2020.
53. Zhou J, Yu JT, Wang HF, et al. Association between stroke and Alzheimer’s disease: systematic review and meta-analysis. Journal of Alzheimer’s disease : JAD 2015;43: 479-489.
54. Akoudad S, Wolters FJ, Viswanathan A, et al. Association of Cerebral Microbleeds With Cognitive Decline and Dementia. JAMA neurology 2016;73: 934-943.
55. Arvanitakis Z, Capuano AW, Leurgans SE, Bennett DA, Schneider JA Relation of cerebral vessel disease to Alzheimer’s disease dementia and cognitive function in elderly people: a cross-sectional study. The Lancet. Neurology 2016;15: 934-943.
56. Zlokovic BV Neurovascular pathways to neurodegeneration in Alzheimer’s disease and other disorders. Nature reviews. Neuroscience 2011;12: 723-738.
57. Fann JR, Ribe AR, Pedersen HS, et al. Long-term risk of dementia among people with traumatic brain injury in Denmark: a population-based observational cohort study. The lancet. Psychiatry 2018;5: 424-431.
58. Nordström A, Nordström P Traumatic brain injury and the risk of dementia diagnosis: A nationwide cohort study. PLoS medicine 2018;15: e1002496.
59. Mackay DF, Russell ER, Stewart K, et al. Neurodegenerative Disease Mortality among Former Professional Soccer Players. The New England journal of medicine 2019;381: 1801-1808.
60. Yaffe K, Lwi SJ, Hoang TD, et al. Military-related risk factors in female veterans and risk of dementia. Neurology 2019;92: e205-e211.
61. Zhou F, Chen S Hyperhomocysteinemia and risk of incident cognitive outcomes: An updated dose-response meta-analysis of prospective cohort studies. Ageing research reviews 2019;51: 55-66.
62. Chen S, Honda T, Ohara T, et al. Serum homocysteine and risk of dementia in Japan. Journal of neurology, neurosurgery, and psychiatry 2020;91: 540-546.
63. Li J-G, Chu J, Barrero C, Merali S, Praticò D Homocysteine exacerbates β-amyloid pathology, tau pathology, and cognitive deficit in a mouse model of Alzheimer disease with plaques and tangles. Annals of neurology 2014;75: 851-863.
64. Kaye AD, Jeha GM, Pham AD, et al. Folic Acid Supplementation in Patients with Elevated Homocysteine Levels. Adv Ther 2020;37: 4149-4164.
65. Hung S-C, Liao K-F, Muo C-H, et al. Hearing Loss is Associated With Risk of Alzheimer’s Disease: A Case-Control Study in Older People. Journal of epidemiology 2015;25: 517-521.
66. Griffiths TD, Lad M, Kumar S, et al. How Can Hearing Loss Cause Dementia? Neuron 2020.
67. Armstrong NM, An Y, Doshi J, et al. Association of Midlife Hearing Impairment With Late-Life Temporal Lobe Volume Loss. JAMA Otolaryngol Head Neck Surg 2019.
68. Maharani A, Dawes P, Nazroo J, Tampubolon G, Pendleton N Longitudinal Relationship Between Hearing Aid Use and Cognitive Function in Older Americans. Journal of the American Geriatrics Society 2018;66: 1130-1136.
69. Takeuchi K, Ohara T, Furuta M, et al. Tooth Loss and Risk of Dementia in the Community: the Hisayama Study. Journal of the American Geriatrics Society 2017;65: e95-e100.
70. Chen C-K, Wu Y-T, Chang Y-C Association between chronic periodontitis and the risk of Alzheimer’s disease: a retrospective, population-based, matched-cohort study. Alzheimer’s research & therapy 2017;9: 56.
71. Pritchard AB, Crean S, Olsen I, Singhrao SK Periodontitis, Microbiomes and their Role in Alzheimer’s Disease. Frontiers in aging neuroscience 2017;9: 336.
72. Kishimoto H, Ohara T, Hata J, et al. The long-term association between physical activity and risk of dementia in the community: the Hisayama Study. European journal of epidemiology 2016;31: 267-274.
73. Broadhouse KM, Singh MF, Suo C, et al. Hippocampal plasticity underpins long-term cognitive gains from resistance exercise in MCI. NeuroImage. Clinical 2020;25: 102182.
74. Bisbe M, Fuente-Vidal A, López E, et al. Comparative Cognitive Effects of Choreographed Exercise and Multimodal Physical Therapy in Older Adults with Amnestic Mild Cognitive Impairment: Randomized Clinical Trial. Journal of Alzheimer’s disease : JAD 2020;73: 769-783.
75. Guadagni V, Drogos LL, Tyndall AV, et al. Aerobic exercise improves cognition and cerebrovascular regulation in older adults. Neurology 2020;94: e2245-e2257.
76. Firth J, Stubbs B, Vancampfort D, et al. Effect of aerobic exercise on hippocampal volume in humans: A systematic review and meta-analysis. Neuroimage 2018;166: 230-238.
77. Thomas BP, Tarumi T, Sheng M, et al. Brain Perfusion Change in Patients with Mild Cognitive Impairment After 12 Months of Aerobic Exercise Training. Journal of Alzheimer’s disease : JAD 2020;75: 617-631.
78. Sabia S, Dugravot A, Dartigues JF, et al. Physical activity, cognitive decline, and risk of dementia: 28 year follow-up of Whitehall II cohort study. BMJ (Clinical research ed.) 2017;357: j2709.
79. Shi L, Chen S-J, Ma M-Y, et al. Sleep disturbances increase the risk of dementia: A systematic review and meta-analysis. Sleep medicine reviews 2018;40.
80. Lutsey PL, Misialek JR, Mosley TH, et al. Sleep characteristics and risk of dementia and Alzheimer’s disease: The Atherosclerosis Risk in Communities Study. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2018;14: 157-166.
81. Westwood AJ, Beiser A, Jain N, et al. Prolonged sleep duration as a marker of early neurodegeneration predicting incident dementia. Neurology 2017;88: 1172-1179.
82. Chen J-C, Espeland MA, Brunner RL, et al. Sleep duration, cognitive decline, and dementia risk in older women. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2016;12: 21-33.
83. Xu W, Tan L, Su B-J, et al. Sleep characteristics and cerebrospinal fluid biomarkers of Alzheimer’s disease pathology in cognitively intact older adults: The CABLE study. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2020.
84. Wang C, Holtzman DM Bidirectional relationship between sleep and Alzheimer’s disease: role of amyloid, tau, and other factors. Neuropsychopharmacology 2020;45: 104-120.
85. Zhong G, Wang Y, Zhang Y, Guo JJ, Zhao Y Smoking is associated with an increased risk of dementia: a meta-analysis of prospective cohort studies with investigation of potential effect modifiers. PloS one 2015;10: e0118333.
86. Durazzo TC, Mattsson N, Weiner MW Smoking and increased Alzheimer’s disease risk: a review of potential mechanisms. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2014;10: S122-S145.
87. Choi D, Choi S, Park SM Effect of smoking cessation on the risk of dementia: a longitudinal study. Ann Clin Transl Neurol 2018;5: 1192-1199.
88. Langballe EM, Ask H, Holmen J, et al. Alcohol consumption and risk of dementia up to 27 years later in a large, population-based sample: the HUNT study, Norway. European journal of epidemiology 2015;30: 1049-1056.
89. Xu W, Wang H, Wan Y, et al. Alcohol consumption and dementia risk: a dose-response meta-analysis of prospective studies. European journal of epidemiology 2017;32: 31-42.
90. Handing EP, Andel R, Kadlecova P, Gatz M, Pedersen NL Midlife Alcohol Consumption and Risk of Dementia Over 43 Years of Follow-Up: A Population-Based Study From the Swedish Twin Registry. The journals of gerontology. Series A, Biological sciences and medical sciences 2015;70: 1248-1254.
91. Sharma A, Brenner M, Wang P Potential Role of Extracellular CIRP in Alcohol-Induced Alzheimer’s Disease. Molecular neurobiology 2020;57: 5000-5010.
92. Fukuda T, Ohnuma T, Obara K, et al. Supplementation with Matured Hop Bitter Acids Improves Cognitive Performance and Mood State in Healthy Older Adults with Subjective Cognitive Decline. Journal of Alzheimer’s disease : JAD 2020.
93. Reale M, Costantini E, Jagarlapoodi S, et al. Relationship of Wine Consumption with Alzheimer’s Disease. Nutrients 2020;12.
94. Andrews SJ, Goate A, Anstey KJ Association between alcohol consumption and Alzheimer’s disease: A Mendelian randomization study. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2020;16: 345-353.
95. Wu L, Sun D, He Y Coffee intake and the incident risk of cognitive disorders: A dose-response meta-analysis of nine prospective cohort studies. Clinical nutrition (Edinburgh, Scotland) 2017;36: 730-736.
96. Sugiyama K, Tomata Y, Kaiho Y, et al. Association between Coffee Consumption and Incident Risk of Disabling Dementia in Elderly Japanese: The Ohsaki Cohort 2006 Study. Journal of Alzheimer’s disease : JAD 2016;50: 491-500.
97. Xu H, Wang Y, Yuan Y, et al. Gender differences in the protective effects of green tea against amnestic mild cognitive impairment in the elderly Han population. Neuropsychiatric disease and treatment 2018;14: 1795-1801.
98. Ma Y-H, Wu J-H, Xu W, et al. Associations of Green Tea Consumption and Cerebrospinal Fluid Biomarkers of Alzheimer’s Disease Pathology in Cognitively Intact Older Adults: The CABLE Study. Journal of Alzheimer’s disease : JAD 2020;77: 411-421.
99. Mancini E, Beglinger C, Drewe J, et al. Green tea effects on cognition, mood and human brain function: A systematic review. Phytomedicine : international journal of phytotherapy and phytopharmacology 2017;34: 26-37.
100. Polito CA, Cai ZY, Shi YL, et al. Association of Tea Consumption with Risk of Alzheimer’s Disease and Anti-Beta-Amyloid Effects of Tea. Nutrients 2018;10.
101. Morris MC, Tangney CC, Wang Y, et al. MIND diet associated with reduced incidence of Alzheimer’s disease. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2015;11: 1007-1014.
102. Chen X, Maguire B, Brodaty H, O’Leary F Dietary Patterns and Cognitive Health in Older Adults: A Systematic Review. Journal of Alzheimer’s disease : JAD 2019;67: 583-619.
103. Samieri C, Rajendra Sonawane A, Lefèvre-Arbogast S, et al. Using network science tools to identify novel diet patterns in prodromal dementia. Neurology 2020: 10.1212/WNL.0000000000009399.
104. Nagpal R, Neth BJ, Wang S, et al. Gut mycobiome and its interaction with diet, gut bacteria and alzheimer’s disease markers in subjects with mild cognitive impairment: A pilot study. EBioMedicine 2020;59: 102950.
105. Gentreau M, Chuy V, Féart C, et al. Refined carbohydrate-rich diet is associated with long-term risk of dementia and Alzheimer’s disease in apolipoprotein E ε4 allele carriers. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2020.
106. Zhang Y, Chen J, Qiu J, et al. Intakes of fish and polyunsaturated fatty acids and mild-to-severe cognitive impairment risks: a dose-response meta-analysis of 21 cohort studies. The American journal of clinical nutrition 2016;103: 330-340.
107. Morris MC, Brockman J, Schneider JA, et al. Association of Seafood Consumption, Brain Mercury Level, and APOE epsilon4 Status With Brain Neuropathology in Older Adults. Jama 2016;315: 489-497.
108. Littlejohns TJ, Henley WE, Lang IA, et al. Vitamin D and the risk of dementia and Alzheimer disease. Neurology 2014;83: 920-928.
109. Qin B, Xun P, Jacobs DR, et al. Intake of niacin, folate, vitamin B-6, and vitamin B-12 through young adulthood and cognitive function in midlife: the Coronary Artery Risk Development in Young Adults (CARDIA) study. The American journal of clinical nutrition 2017;106: 1032-1040.
110. Noguchi-Shinohara M, Abe C, Yuki-Nozaki S, et al. Higher Blood Vitamin C Levels are Associated with Reduction of Apolipoprotein E E4-related Risks of Cognitive Decline in Women: The Nakajima Study. Journal of Alzheimer’s disease : JAD 2018;63: 1289-1297.
111. Holland TM, Agarwal P, Wang Y, et al. Dietary flavonols and risk of Alzheimer dementia. Neurology 2020.
112. Zhang C, Wang Y, Wang D, Zhang J, Zhang F NSAID Exposure and Risk of Alzheimer’s Disease: An Updated Meta-Analysis From Cohort Studies. Frontiers in aging neuroscience 2018;10: 83.
113. Ryan J, Storey E, Murray AM, et al. Randomized placebo-controlled trial of the effects of aspirin on dementia and cognitive decline. Neurology 2020.
114. Veronese N, Stubbs B, Maggi S, et al. Low-Dose Aspirin Use and Cognitive Function in Older Age: A Systematic Review and Meta-analysis. Journal of the American Geriatrics Society 2017;65: 1763-1768.
115. Gray SL, Anderson ML, Dublin S, et al. Cumulative use of strong anticholinergics and incident dementia: a prospective cohort study. JAMA Intern Med 2015;175: 401-407.
116. Calderón-Garcidueñas L, González-Maciel A, Reynoso-Robles R, et al. Quadruple abnormal protein aggregates in brainstem pathology and exogenous metal-rich magnetic nanoparticles. The substantia nigrae is a very early target in young urbanites and the gastrointestinal tract likely a key brainstem portal. Environ Res 2020: 110139.
117. Oudin A, Forsberg B, Adolfsson AN, et al. Traffic-Related Air Pollution and Dementia Incidence in Northern Sweden: A Longitudinal Study. Environmental health perspectives 2016;124: 306-312.
118. Younan D, Wang X, Casanova R, et al. PM(2.5) associated with gray matter atrophy reflecting increased Alzheimers risk in older women. Neurology 2020.
119. Cacciottolo M, Wang X, Driscoll I, et al. Particulate air pollutants, APOE alleles and their contributions to cognitive impairment in older women and to amyloidogenesis in experimental models. Translational psychiatry 2017;7: e1022.
120. Grande G, Ljungman PLS, Eneroth K, Bellander T, Rizzuto D Association Between Cardiovascular Disease and Long-term Exposure to Air Pollution With the Risk of Dementia. JAMA neurology 2020.



D. Kellar1, S.N. Lockhart1, P. Aisen2, R. Raman2, R.A. Rissman2,3, J. Brewer2,3, S. Craft1


1. Department of Internal Medicine–Geriatrics, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA; 2. Alzheimer’s Therapeutic Research Institute, University of Southern California, San Diego, USA; 3. Department of Neurosciences, University of California, San Diego, La Jolla, USA

Corresponding Author: Suzanne Craft, PhD, Department of Internal Medicine–Geriatrics, Wake Forest School of Medicine, One Medical Center Boulevard, Winston-Salem, NC 27157,
J Prev Alz Dis 2021;
Published online April 7, 2021,



Background: Intranasally administered insulin has shown promise in both rodent and human studies in Alzheimer’s disease; however, both effects and mechanisms require elucidation.
Objective: We assessed the effects of intranasally administered insulin on white matter health and its association with cognition and cerebral spinal fluid biomarker profiles in adults with mild cognitive impairment or Alzheimer’s disease in secondary analyses from a prior phase 2 clinical trial (NCT01767909).
Design: A randomized (1:1) double-blind clinical trial.
Setting: Twelve sites across the United States.
Participants: Adults with mild cognitive impairment or Alzheimer’s disease.
Intervention: Participants received either twice daily placebo or insulin (20 IU Humulin R U-100 b.i.d.) intranasally for 12 months. Seventy-eight participants were screened, of whom 49 (32 men) were enrolled.
Measurements: Changes from baseline in global and regional white matter hyperintensity volume and gray matter volume were analyzed and related to changes in cerebral spinal fluid biomarkers, Alzheimer’s Disease Assessment Scale-Cognition, Clinical Disease Rating-Sum of Boxes, Alzheimer’s Disease Cooperative Study–Activities of Daily Living Scale, and a memory composite.
Results: The insulin-treated group demonstrated significantly reduced changes in white matter hyperintensity volume in deep and frontal regions after 12 months, with a similar trend for global volume. White matter hyperintensity volume progression correlated with worsened Alzheimer’s disease cerebral spinal fluid biomarker profile and cognitive function; however, patterns of correlations differed by treatment group.
Conclusion: Intranasal insulin treatment for 12 months reduced white matter hyperintensity volume progression and supports insulin’s potential as a therapeutic option for Alzheimer’s disease.

Key words: Alzheimer’s disease, clinical trial, intranasal insulin, white matter, CSF.



Alzheimer’s disease (AD) is the leading cause of dementia and, as there are currently no disease modifying treatments, its prevalence is expected to increase in response to an aging population (1). AD is characterized by aggregation of amyloid beta (Aβ) plaques and tau neurofibrillary tangles (NFT). Clinical trials have attempted to reduce accumulation of these proteins in the brain and prevent further cognitive decline; however, although some amyloid antibody trials have successfully reduced plaque load, none have been successful to date in halting the progression of AD symptoms (2, 3). Positron Emission Tomography (PET) measures AD pathology by quantifying the load of Aβ and NFT in the brain, and a number of studies have demonstrated relationships between PET measures and concentrations of these proteins in cerebral spinal fluid (CSF) (4, 5). CSF Aβ42 decreases as amyloid accumulation in brain increases, suggesting it is being sequestered in the brain parenchyma, while hyperphosphorylated tau (p-tau) in CSF increases with increased propagation of NFT (6). Utilizing ratios of Aβ42 to Aβ40, p-tau, and total tau (t-tau) has been found to further improve specificity and sensitivity to identify AD dementia (7, 8). There remains a clear need for a pharmacological intervention to prevent or slow AD progression.
As AD progresses, gray matter volume is reduced. While this is also true of non-pathological aging, adults with AD exhibit a far greater reduction in overall gray matter associated with a predictable pattern (9). The hippocampus and entorhinal cortex are perhaps the most easily detected regions affected by AD (10); however, an AD-defined meta region including the hippocampus and other regions such as entorhinal, inferior temporal, middle temporal, inferior parietal, fusiform, and precuneus has recently been utilized in order to increase sensitivity (11). Cortical thickness has also been postulated to be a more reliable marker of AD progression as it is less affected by total intracranial volume which can vary greatly between patients (11). While progression of the disease and corresponding gray matter reduction have been well established, its utility as a clinical trial endpoint needs to be validated. Interventions that have successfully removed amyloid or even allayed cognitive decline have also been associated with reduced gray matter volume (12, 13). These findings illustrate a complex relationship between brain health and gray matter volumetrics which is not fully understood.
White matter integrity can be reflected by the presence of white matter hyperintensities (WMH) detected with magnetic resonance imaging (MRI); however, the relationship of WMHs and AD progression has not been characterized in detail. WMHs are detected using fluid-attenuated inversion recovery (FLAIR) MRI, are presumed to indicate cerebrovascular pathology (14), and are associated with gliosis, demyelination, axonal loss, and arteriosclerosis (15). It is postulated that WMHs reflect a number of factors including hypoxia, amyloid angiopathy, blood brain barrier damage, degeneration, hypoperfusion, and inflammation (15). WMH volume (WMHV) increases with age and some studies have found WMHV to be independent of Aβ burden, leading to the proposal that WMHs should be considered a co-pathology that do not directly contribute to AD (16, 17). Other studies have found correlations between WMHs and cortical tau load in AD (18). It has been proposed that cerebrovascular pathology represented by WMHs precedes and therefore could initiate Aβ aggregation (19-21). Conversely, other investigators claim that Aβ induces vascular damage through neuroinflammation, formation of reactive oxygen species, and oxidative stress (22, 23). It is possible that vascular and AD-specific pathology form a vicious cycle giving rise to these differing viewpoints, or that the precise time course of the association differs for subgroups of patients with AD.
Identification of the nature of the association of WMHs and AD is hindered by lack of a commonly-accepted standardized approach to measurement of WMHs (24). Methods evaluating WMHs range from semi-quantitative visual reads using one of 3 established rating scales (Manolio, Fazekas and Schmidt, Scheltens) to fully automated lesion segmentation (25, 26). Representation of the data is also not consistent, with some studies reporting global ratings or volumes and others focusing on spatial patterns (27, 28). There are also many ways to spatially segment WMHs such as periventricular versus deep (29); however, deep can also be further split in to juxtacortical and non-juxtacortical[30]. Studies in which WMHs are segmented in classical lobular fashion have reported that AD is associated with temporal WMHs (31), or with global and parietal/occipital volumes (32). Volumes can also be displayed as raw volumes (33), log transformed values (33), or percent change ratios (34, 35). As the field progresses automated techniques generating quantitative spatially accurate information may prove the best way to track WMH progression in AD. It is clear that WMHs are associated with poor cognitive outcomes and preventing progression is a clinically relevant marker.
A promising area of research in the treatment and prevention of AD focuses on metabolism, inflammation and, in particular, the role of insulin in the central nervous system. With respect to metabolism, although insulin does not appear to impact global transport of glucose into the brain, it has been shown to increase glucose uptake via the glucose transporter GLUT4 in selected regions such as the hippocampus (36). Further, insulin increases glycogen storage in astrocytes, thereby providing an alternate energy source during glucose deprivation or intense neuronal activity (37). Insulin has long been implicated in AD and several reviews have highlighted both its importance and therapeutic potential (38-40). In short, insulin has been demonstrated to modulate both Aβ and pathological tau formation, and improves neuronal health, dendritic spine proliferation, and white matter integrity. Insulin can be administered intranasally where it is detectible in perivascular spaces with PET imaging (41) and in the CSF in less than 30 minutes (42). A promising pilot trial documented improvements in delayed memory recall, preserved Alzheimer’s Disease Assessment Scale (ADAS-Cog) scores, and functional abilities assessed by the Alzheimer’s Disease Cooperative Study–Activities of Daily Living Scale (ADCS-ADL) after 4 months of treatment with intranasal insulin compared to placebo (43). A recent large 18-month phase II clinical trial of INI treatment in AD and MCI patients found differing patterns of results depending on the device used to administer the insulin (13). For the device used by the primary intent-to-treat cohort, no significant differences in rates of decline measured by the ADAS-Cog13, Clinical Disease Rating-Sum of Boxes (CDR-SOB), ADCS-ADL, or CSF Aβ and tau were observed between placebo and insulin groups. In a secondary cohort, a different device showed better performance on ADAS-Cog13 in the insulin-treated group compared to placebo at 6 months with a similar trend at 12 months. In open-label analyses, the early-start secondary device cohort treated with insulin performed better on the ADAS-Cog13 and ADL-MCI at 18 months than the delayed start secondary group. The insulin-treated group using this device also demonstrated an improvement in CSF Aβ42/Aβ40 and Aβ42/t-tau ratios at 12 months. This study highlights the need for additional investigation to definitively determine the potential for intranasal insulin as a therapeutic for AD.
In the present study, we assessed the effects of INI on white matter health in the secondary cohort of participants using the device associated with improved cognition and AD biomarker profiles. There are several mechanisms through which insulin could act directly to improve white matter health and prevent WMH progression (44). Reduced insulin levels or activity impair oligodendrocyte myelin survival and maintenance, and increase ceramides and decreases sulfatides, leading to oxidative stress, inflammation, and lipid peroxidation. These factors all contribute to myelin damage and subsequent WMHs. Insulin resistance impairs vascular responsiveness, causing luminal narrowing and fibrosis, which cause decreased blood flow and blood brain barrier damage. These effects lead to ischemia and inflammation and promote the formation of WMHs. As mentioned previously, insulin reduces Aβ and p-tau levels in the brain, both of which can cause inflammation, neuronal and glial damage, and vascular impairment (45). These distinct pathways could all culminate in the formation and progression of WMHs, thus poising insulin at a convergence point in several potential cascades, and raising the possibility that providing insulin to the brain to overcome deficient insulin availability or activity may have therapeutic benefit in AD.
Based on this evidence, we examined the effect of 12 months of INI treatment vs. placebo on change in WMHs. We also examined the relationships among changes in WMHs, cognition, and AD CSF biomarkers.



The study was overseen by the Alzheimer’s Therapeutic Research Institute (P. Aisen, Director) together with the Principal Investigator (S. Craft). Eligibility and recruitment for this study have been described previously (13). Two devices were used in the parent study; however, only one device demonstrated cognitive benefits or changes in AD CSF biomarkers across the 18-month long study. For this reason, we evaluated only the group using the device that showed potential beneficial effects. Briefly, participants with AD (n=31) or amnestic MCI (n=18) were recruited from 12 sites. Participants received baseline testing including CDR, MMSE, ADAS-Cog13, a lumbar puncture, and an MRI, then were randomized on a 1:1 basis to receive either 20 IU intranasal insulin (n=24) or placebo (n=25) twice daily for 12 months. After 12 months the cognitive battery was readministered, a lumbar puncture was performed, and another MRI was obtained. There were a total of 40 participants (insulin n=20; placebo n=20) with MRI data that passed quality control measures at baseline and month 12.
T1 and Fluid Attenuated Inversion Recovery images were collected with 1.5 or 3T MRI. T1 weighed images were processed using FreeSurfer 6.0.0 to produce participant specific gray matter volume, thickness, and area. FLAIR images were segmented by the lesion growth algorithm[46] as implemented in the LST toolbox version 3.0.0 ( for SPM. The algorithm first segments the T1 images into the three main tissue classes (CSF, GM and WM). This information is then combined with the coregistered FLAIR intensities in order to calculate lesion belief maps. By thresholding these maps with a pre-chosen initial threshold (κ= 0.3) an initial binary lesion map is obtained which is subsequently grown along voxels that appear hyperintense in the FLAIR image. The result is a lesion probability map. The lesion probability maps were then warped to MNI space and lobular volume was extracted using Mayo Clinic Adult Lifespan Template (47). A temporal-parietal volume meta-ROI was created to examine volume and was defined as bilateral entorhinal, inferior temporal, middle temporal, inferior parietal, fusiform, and precuneus (11). Cortical thickness was similarly defined by Jack (48) as the surface area weighted thickness of the entorhinal, inferior temporal, middle temporal, and fusiform.
CSF was collected in the morning after an overnight fast and was immediately placed on dry ice and was shipped overnight to the central biomarker laboratory. AD biomarkers Aβ42, Aβ40, total tau, and tau phosphorylated at threonine 181 were quantified with the Meso Scale Discovery platform (Meso Scale Diagnostics). Blood was collected for APOE genotyping using established protocols.
Cross sectional analysis was performed to assess group differences at baseline in age, cognitive status, sex, baseline surface weighted cortical thickness, AD signature region volume, total WMHV, and regional WMHV at baseline using general linear models or chi squared tests when appropriate. Change variables for gray matter and WMHV were defined as percentage change from baseline as previously described (35). General linear modeling was performed in SAS v 9.4 with covariates age, ApoE4 status, study site, and sex included in all initial models. Baseline volumes were also included for GMV and WMHV models, and total intracranial volume was included in models analyzing data in native space. Non-contributing covariates (p>0.15) were dropped from the model. No adjustments were made for multiple comparisons; rather, results are reported as mean estimates and corresponding 95% confidence intervals. In exploratory analyses, we examined whether individual treatment groups showed reliable change in WMHV over time with within-group LSMEANS t-tests. Change variables for WMHV, cognitive scores, and CSF values were subjected to Pearson’s r correlations to determine inter-relationships.




For the parent study secondary cohort that utilized the device associated with cognitive benefit, 78 participants were screened, of whom 49 (32 men [65.3%]) were enrolled. Twenty-four were randomized to the insulin arm and 25 were randomized to the placebo arm (Figure 1). Of those 49, 40 participants (insulin n=20, placebo n=20) had usable MRI data at both time points and were analyzed for this study. There were no demographic or other notable clinical differences between participants with usable and unusable data. There were also no differences in demographic characteristics between arms at baseline (Table 1).

MRI Results

The temporal-parietal meta-ROI decreased in volume over time as did the surface weighted cortical thickness (both ps<.001, Figure 2). There was no interaction between treatment arm and rate of decline of gray matter or cortical thickness (Supplementary Table 1).

Figure 1. CONSORT diagram


Figure 2. Changes in gray matter A) volume and B) surface weighted thickness in the temporal-parietal meta-ROI

There were no significant differences between treatment group and placebo. Error bars represent 95% confidence intervals.


An interaction between treatment arm and global WMHV was observed such that the insulin-treated arm tended to have less global WMH volume increase over the 12 month intervention compared to the placebo group (insulin lsmeans [95% CI]=18.98 [-1.38,39.33] and placebo 42.21 [21.70,62.72], p=0.064, Figure 3). Given this trend, exploratory analyses were conducted for comparisons of individual ROIs between treatment arms. Insulin significantly reduced WMHV change over the 12 month intervention in both the frontal lobe and deep white matter compared to placebo (frontal insulin lsmeans [95% CI]=15.14 [-3.84,34.12] and placebo=39.18 [20.05,58.30], p=0.042; deep WM insulin lsmeans [95% CI]=56.94 [-20.20,134.08] and placebo=161.37 [81.68, 241.05], p=0.042, Figure 3). Change in WMHV was less in the insulin arm than the placebo arm for all other regions, although these comparisons did not reach statistical significance.

Figure 3. White Matter Hyperintensity Volume (WMHV) as percent change from baseline both globally and regionally split by MCALT (excluding cerebellum and midbrain regions)

There were significant differences between the degree of change for insulin and placebo groups in the deep white matter and frontal regions with a similar trend for global change (+ p<0.10, * p<0.05). The placebo group showed significantly increased change from baseline WMHV in all regions, whereas the insulin group showed significant change only in temporal lobe with a trend for global change (# <0.10, ## p<0.05). Error bars represent 95% confidence intervals.


When we examined whether individual treatment groups showed reliable change in WMHV over time with within-group LSMEANS t-tests, the placebo group showed significantly increased WMHVs across all regions (all ps<0.05, Figure 3; raw means for baseline and month 12 for all regions are presented in Supplementary Table 2), whereas WMHV was unchanged following insulin treatment in the deep white matter, corpus callosum, occipital, parietal, and frontal regions (all p>0.1, Figure 3). Temporal WMHV increased slightly over the 12 month intervention with insulin treatment with a similar trend in global WMHV (temporal p=0.033; global p=0.066), although to a lesser degree than with placebo.

Correlation between MRI and Cognitive Outcomes

For the combined cohort including both insulin and placebo groups, increased global WMHV correlated with lowered memory composite scores (r=-0.38, p=0.024, Figure 4A) and similar trends were observed for the ADAS-Cog13 and CDR-SOB (r=0.297, p=0.062; r=0.278, p=0.081, figure 4A). Regional analysis revealed a significant correlation between the memory composite score and parietal and occipital WMHVs and a trend correlation for the corpus callosum (r=-0.536, p=0.001; r=-0.405, p=0.015; r=0.31, p=0.069, Figure 4A). Increased temporal WMHV was associated with worsened (higher) scores for the ADAS-Cog13 (r=0.313, p=0.049, Figure 4A). A similar trend correlation was observed between frontal WMHV and ADCS-ADL scores (r=-0.267, p=0.095, Figure 4A).

Figure 4. Changes in global and regional White Matter Hyperintensity Volume (WMHV) correlate with changes in ADAS-cog, CDR-SB, ADCS-ADL, and a memory composite

Analyses were performed for both insulin and placebo groups combined (A) and for the insulin treatment arm (B) and placebo (C) groups independently. Light colors represent correlations with lower p values (ps range from <0.001 to 0.10 from light to dark). Exemplar scatterplots are shown that demonstrate relationships between change in frontal WMHVs (which differed between insulin and placebo groups) and change in (D) CDR-SB, (E) ADCS-ADL, and (F) memory composite scores.


When analyzed by treatment group, the insulin group showed a significant correlation between change in ADAS-Cog13 scores and WMHVs change in the corpus callosum, and trending correlations for the deep white matter and temporal regions (r=0.459, p=0.041; r=0.409, p=0.073; r=0.368, p=0.092, Figure 4B). Increased (worsened) CDR-SOB scores correlated significantly with increased frontal and global WMHV with trends noted for deep white matter, temporal, and the corpus callosum regions (r=0.535, p=0.014; r=0.537, p=0.014; r=0.439, p=0.052; r=0.426, p=0.059; r=0.409, p=0.061, Figure 4B). Smaller increases in frontal WMHV also correlated with smaller increases in ADCS-ADL scores (r=-0.475, p=0.034). In the placebo group, declines in memory composite scores only correlated with increased WMHV in global, parietal, and occipital regions with trends for frontal, corpus callosum, and temporal regions (r=-0.639, p=0.005; r=-0.773, p=0.001; r=-0.585, p=0.013; r=-0.48, p=0.05; r=0.435, p=0.08; r=-0.425, p=0.088, Figure 4C).

Correlation between MRI and CSF outcomes

For the combined cohort including both insulin and placebo groups, global WMHV increase correlated with a decrease in CSF Aβ42 and with a similar trend for the Aβ42/Aβ40 ratio (r=-0.375, p=0.028; r=-0.328, p=0.058, Figure 5A). Frontal WMHV increases also correlated with decreases in both Aβ42 and Aβ42/Aβ40 ratio (r=-0.355, p=0.039; r=-0.44, p=0.009, Figure 5A). Increases in both corpus callosum and deep white matter WMHV correlated with decreases in Aβ42/tau ratio (r=-0.397, p=0.001; r=-0.734, p=0.001, figure 5A).

Figure 5. Changes in global and regional White Matter Hyperintensity Volume (WMHV) correlate with changes in CSF AD biomarkers

Analyses were performed for both insulin and placebo groups combined (A) and for the insulin treatment arm (B) and placebo (C) groups independently. Light colors represent correlations with lower p values (ps range from <0.001 to 0.10 from light to dark). Exemplar scatterplots are shown that demonstrate relationships between change in frontal WMHVs (which differed between insulin and placebo groups) and change in CSF (D) Aβ42, (E) Aβ42/Aβ40 ratio, and (F) Aβ42/T-tau ratio.


When analyzed by treatment group, there was a significant correlation between increased parietal WMHV and decreased Aβ40 in the insulin group (r=-0.525, p=0.036, figure 5B). There were trending relationships between increased global WMHV and decreased Aβ42/Aβ40 ratio, increased frontal WMHV and decreased Aβ42/Aβ40 ratio and Aβ42/t-tau ratio changes, and increased corpus callosum and decreased Aβ42 levels (r=-0.443, p=0.085; r=-0.467, p=0.067; r=-0.477, p=0.061; r=-0.446, p=0.083, figure 5B). In the placebo group, increases in both corpus callosum and deep white matter WMHV correlated with decreases in Aβ42/t-tau ratio (r=-0.733, p=0.001; r=-0.724, p=0.001, figure 5C). Temporal WHV change positively correlated with ptau-181/t-tau ratio (r=0.481, p=0.043, figure 5C). There was a trend for both global and frontal WMHV change to negatively correlate with Aβ42 change (r=-0.431, p=0.073; r=-0.408, p=0.092, figure 5C).



The present study found that increased WMHV correlated with greater declines in cognition and worsening of CSF AD biomarker profiles, and that INI treatment for 12 months reduced WMHV progression in key brain regions. White matter hyperintensities represent cerebral small vessel disease and white matter damage resulting from degraded myelin, and have been shown to increase with aging. Some studies have suggested that regional, and not global WMHV may best predict, or correlate, with AD progression (49). While there are numerous ways to segregate WMHs the most promising research has suggested a division between frontal and posterior regions in differentiating normal versus pathological aging, with lobular divisions further increasing the regional specificity (32, 50). We observed that INI treatment for 12 months slowed the progression of WMHs globally and in deep WM and frontal regions compared to placebo assignment. This finding supports previous studies linking frontal WMHs and pathological aging and suggests that INI may reduce AD-related WMH progression in these key areas. Exploratory analysis also demonstrated that for all other regions, with the exception of the temporal lobe, progression of WMHs were stabilized following insulin treatment. These findings may be considered clinically significant as growing research has shown that WMHVs correlate with numerous pathological outcomes such as ependymal loss, cerebral ischemia, and demyelination (51). Insulin could be acting through a number of pathways to preserve white matter health (44). Insulin resistance impairs oligodendrocyte myelin maintenance and survival, while also leading to reduced vascular smooth muscle responsiveness. Increasing insulin availability could ameliorate these deficiencies thereby reducing WMH progression.
Previous research has shown that higher baseline WMHV predicts worse cognitive outcomes measured by the ADAS-Cog (52) and CDR (53). We demonstrated in the combined cohort that over 12 months, longitudinal increases in WMHV correlate with decreases in cognition measured by the ADAS-Cog, CDR-SOB, ADCS-ADL, and a memory composite. We also found that correlations differed depending on treatment group. The placebo group showed correlations only between WMHV and memory composite scores, while the insulin treated group showed associations between WMHs and both the ADAS-Cog13 and CDR-SOB, as well as the ADCS-ADL. Of note, in the parent study, insulin treatment was shown to benefit performance on the ADAS-Cog13, CDR, and ADCS-ADL for the secondary cohort (13). The present findings raise the possibility that insulin’s effects on these measures are mediated in part by factors associated with WMHV. These findings suggest that preventing or even delaying progression of white matter damage could prevent global cognitive and functional decline.
Increased CSF Aβ42 levels and decreased hyperphosphorylated tau concentrations have been proposed as markers demonstrating improvement in AD pathology (54, 55). Studies have shown that greater Aβ load, indicated by lower CSF Aβ42 concentration and increased Aβ PET, correlates with greater WMH burden (56-59); however, the findings regarding p-tau and total tau are less consistent, with some studies reporting correlations (18, 60) while other do not (59, 61, 62). In the parent study, insulin treatment was associated with improved (increased) Aβ42/Aβ 40 and Aβ42/tau ratios (13) for the secondary cohort treated with the device which was associated with cognitive benefit. The present study adds to these findings by demonstrating that WMHV changes inversely correlate with changes in Aβ40, Aβ42, Aβ42/Aβ40 ratio, and Aβ42/t-tau ratio; however, these associations differ by region and CSF metric. The relationship between WMHV and tau and Aβ may change throughout the course of the disease as Aβ pathology is stabilized and tau pathology increases; however, this hypothesis is still in contention (59). Insulin affects the regulation of both Aβ and hyperphosphorylated tau (63). Both Aβ40 (64) and Aβ42 (65) cause inflammation via production of reactive oxygen species. Aβ42 aggregates and starts seeding points which grow to plaques leading to death in several cell types including oligodendrocytes (66, 67), while Aβ40 is more prone to form deposits on vessels walls where it damages pericytes (68, 69). Hyperphosphorylated tau also leads to neurovascular dysfunction resulting in reduced energy supply due to aberrant vessel dilation (70). Insulin may act indirectly to stop WMH progression by preventing damage caused by hyperphosphorylated tau, Aβ40, and Aβ42 to both the blood brain barrier and myelin producing oligodendrocytes. Further research needs to be done to elucidate patterns behind regional WMH load and CSF AD biomarkers of Aβ and tau.
Numerous studies have sought to identify the temporal and spatial patterns of gray matter volume loss in typical AD progression (71). While there is not a consensus across all studies, several vulnerable regions have been highlighted (11, 72). Measures of gray matter volume and thickness were unchanged by 12 months of INI administration. It is possible that our region of interest, while widely accepted, failed to capture subtle differences between treatment groups. Other regions of the brain that were excluded from our analysis may be more sensitive to changes specific to this intervention. Previous interventions have shown decreased rates of atrophy in patients with MCI treated with other non-insulin interventions; however, these findings were over a 2-year period (73). It is possible that our intervention was not long enough to detect subtle changes in gray matter volume. Cognitive improvement has been observed without halting gray matter atrophy. A phase II clinical trial testing daily administration of resveratrol showed greater reduction in brain volume in the treatment group compared to placebo after one year of treatment, but the resveratrol-treated group also showed less decline on the ADCS-ADL suggesting interventions can still be beneficial independent of apparent reduction of gray matter volume (74).
Our study had several limitations. Our small sample size may have contributed to our inability to detect preservation of gray matter volume and cortical thickness by INI treatment, and may also have reduced our ability to detect relationships between WMHVs and other measures. Our cohort was predominantly white (93%) and male (62%) and thus results may not generalize to a more diverse population. A number of participants did not complete the trial or had usable data at either baseline or follow up. These missing datapoints could result in a completer bias; however, this is unlikely as those with incomplete data did not differ demographically or in any baseline measures from completers. It is also possible that a longer intervention may be needed to observe a divergence between groups on our measures. These results need to be replicated in a larger, longer study in order to determine the effects of insulin on the brain WM health and the mechanistic pathways underlying these effects.
In conclusion, we found that treating MCI and AD patients with 12 months of INI significantly reduced WMH progression without affecting gray matter volume or cortical thickness, and that increases in WMHV correlated with both worsening in AD CSF biomarker profile and cognitive/functional measures. These findings support insulin’s potential as a therapeutic option for AD; however, more research needs to be conducted to elucidate mechanism through which insulin may impact white matter integrity.


Funding and acknowledgments: This study was supported by the National Institute on Aging (NIA RF1AG041845). Eli Lilly provided diluent placebo for the blinded phase of the trial, and insulin for the open label portion at no cost; they had no input into the design, conduct or interpretation of trial results or in the preparation of this manuscript. The authors would like to thank the site personnel, and the following Site Prinicipal Investigators for their perseverance throughout the trial: Deniz Erten-Lyons, David Knopman, Joseph Kass, Rachelle Doody, Hillel Grossman, Neelum Aggarwal, Esther Oh, Gregory Jicha, Anton Porsteinsson, Aimee Pierce, Gaby Thai, Ruth Mulnard, Allan Levey, Jeffrey Burns, Neill Graff-Radford, Jared Brosch, Martin Farlow, Christopher van Dyck, Marek-Marsel Mesulam, Ian Grant, Raymond Scott Turner, Scott McGinnis, Thomas Obisesan, Alan Lerner, Allison Perrin, Laura Baker, Elaine Peskind, Henry Querfurth, Brian Ott, Ralph Richter, Jacobo Mintzer, Marwan Sabbagh, Jiong Shi, Daniel Press, Shauna Yuan, David Carr, Nupar Ghoshal, Amanda Smith, James Galvin, and Kyle Womack. We are particularly grateful to the trial participants for their dedication.

Conflict of interest: The authors have no conflict of interest to report.
Ethical Standards: The study was approved by the Institutional Review Boards of all participating institutions. All participants provided written informed consent to participate in the study.

Open Access: This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits use, duplication, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.


Supplementary Material 1

Supplementary Material 2


1. Association, A.s., 2019 Alzheimer’s disease facts and figures. Alzheimer’s & Dementia, 2019. 15(3): p. 321-387.
2. Sevigny, J., P. Chiao, T. Bussiere, et al., The antibody aducanumab reduces Abeta plaques in Alzheimer’s disease. Nature, 2016. 537(7618): p. 50-6.
3. Klein, G., P. Delmar, G.A. Kerchner, et al., Thirty-Six-Month Amyloid Positron Emission Tomography Results Show Continued Reduction in Amyloid Burden with Subcutaneous Gantenerumab. J Prev Alzheimers Dis, 2021. 8(1): p. 3-6.
4. La Joie, R., A. Bejanin, A.M. Fagan, et al., Associations between [(18)F]AV1451 tau PET and CSF measures of tau pathology in a clinical sample. Neurology, 2018. 90(4): p. e282-e290.
5. Schipke, C.G., N. Koglin, S. Bullich, et al., Correlation of florbetaben PET imaging and the amyloid peptide Ass42 in cerebrospinal fluid. Psychiatry Res Neuroimaging, 2017. 265: p. 98-101.
6. Olsson, B., R. Lautner, U. Andreasson, et al., CSF and blood biomarkers for the diagnosis of Alzheimer’s disease: a systematic review and meta-analysis. Lancet Neurol, 2016. 15(7): p. 673-684.
7. Hansson, O., S. Lehmann, M. Otto, H. Zetterberg, and P. Lewczuk, Advantages and disadvantages of the use of the CSF Amyloid beta (Abeta) 42/40 ratio in the diagnosis of Alzheimer’s Disease. Alzheimers Res Ther, 2019. 11(1): p. 34.
8. Blennow, K. and H. Hampel, CSF markers for incipient Alzheimer’s disease. Lancet Neurol, 2003. 2(10): p. 605-13.
9. Driscoll, I., C. Davatzikos, Y. An, et al., Longitudinal pattern of regional brain volume change differentiates normal aging from MCI. Neurology, 2009. 72(22): p. 1906-13.
10. Jack, C.R., Jr., F. Barkhof, M.A. Bernstein, et al., Steps to standardization and validation of hippocampal volumetry as a biomarker in clinical trials and diagnostic criterion for Alzheimer’s disease. Alzheimers Dement, 2011. 7(4): p. 474-485 e4.
11. Schwarz, C.G., J.L. Gunter, H.J. Wiste, et al., A large-scale comparison of cortical thickness and volume methods for measuring Alzheimer’s disease severity. Neuroimage Clin, 2016. 11: p. 802-812.
12. Sur, C., J. Kost, D. Scott, et al., BACE inhibition causes rapid, regional, and non-progressive volume reduction in Alzheimer’s disease brain. Brain, 2020.
13. Craft, S., R. Raman, T.W. Chow, et al., Safety, Efficacy, and Feasibility of Intranasal Insulin for the Treatment of Mild Cognitive Impairment and Alzheimer Disease Dementia: A Randomized Clinical Trial. JAMA Neurol, 2020. 77(9): p. 1099-1109.
14. Raman, M.R., K. Kantarci, M.E. Murray, C.R. Jack, Jr., and P. Vemuri, Imaging markers of cerebrovascular pathologies: Pathophysiology, clinical presentation, and risk factors. Alzheimers Dement (Amst), 2016. 5: p. 5-14.
15. Abraham, H.M., L. Wolfson, N. Moscufo, C.R. Guttmann, R.F. Kaplan, and W.B. White, Cardiovascular risk factors and small vessel disease of the brain: Blood pressure, white matter lesions, and functional decline in older persons. J Cereb Blood Flow Metab, 2016. 36(1): p. 132-42.
16. Roseborough, A., J. Ramirez, S.E. Black, and J.D. Edwards, Associations between amyloid beta and white matter hyperintensities: A systematic review. Alzheimers Dement, 2017. 13(10): p. 1154-1167.
17. Villeneuve, S. and W.J. Jagust, Imaging Vascular Disease and Amyloid in the Aging Brain: Implications for Treatment. J Prev Alzheimers Dis, 2015. 2(1): p. 64-70.
18. McAleese, K.E., M. Firbank, M. Dey, et al., Cortical tau load is associated with white matter hyperintensities. Acta Neuropathol Commun, 2015. 3: p. 60.
19. Thal, D.R., The pre-capillary segment of the blood-brain barrier and its relation to perivascular drainage in Alzheimer’s disease and small vessel disease. ScientificWorldJournal, 2009. 9: p. 557-63.
20. Iturria-Medina, Y., R.C. Sotero, P.J. Toussaint, J.M. Mateos-Perez, A.C. Evans, and I. Alzheimer’s Disease Neuroimaging, Early role of vascular dysregulation on late-onset Alzheimer’s disease based on multifactorial data-driven analysis. Nat Commun, 2016. 7: p. 11934.
21. Honjo, K., S.E. Black, and N.P. Verhoeff, Alzheimer’s disease, cerebrovascular disease, and the beta-amyloid cascade. Can J Neurol Sci, 2012. 39(6): p. 712-28.
22. Snyder, H.M., R.A. Corriveau, S. Craft, et al., Vascular contributions to cognitive impairment and dementia including Alzheimer’s disease. Alzheimers Dement, 2015. 11(6): p. 710-7.
23. Yamada, M., Predicting cerebral amyloid angiopathy-related intracerebral hemorrhages and other cerebrovascular disorders in Alzheimer’s disease. Front Neurol, 2012. 3: p. 64.
24. Wardlaw, J.M., E.E. Smith, G.J. Biessels, et al., Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration. The Lancet Neurology, 2013. 12(8): p. 822-838.
25. van den Heuvel, D.M., V.H. ten Dam, A.J. de Craen, et al., Measuring longitudinal white matter changes: comparison of a visual rating scale with a volumetric measurement. AJNR Am J Neuroradiol, 2006. 27(4): p. 875-8.
26. Kapeller, P., R. Barber, R.J. Vermeulen, et al., Visual rating of age-related white matter changes on magnetic resonance imaging: scale comparison, interrater agreement, and correlations with quantitative measurements. Stroke, 2003. 34(2): p. 441-5.
27. Ramirez, J., A.A. McNeely, C.J.M. Scott, M. Masellis, S.E. Black, and I. Alzheimer’s Disease Neuroimaging, White matter hyperintensity burden in elderly cohort studies: The Sunnybrook Dementia Study, Alzheimer’s Disease Neuroimaging Initiative, and Three-City Study. Alzheimers Dement, 2016. 12(2): p. 203-210.
28. Rizvi, B., A. Narkhede, B.S. Last, et al., The effect of white matter hyperintensities on cognition is mediated by cortical atrophy. Neurobiol Aging, 2018. 64: p. 25-32.
29. DeCarli, C., E. Fletcher, V. Ramey, D. Harvey, and W.J. Jagust, Anatomical mapping of white matter hyperintensities (WMH): exploring the relationships between periventricular WMH, deep WMH, and total WMH burden. Stroke, 2005. 36(1): p. 50-5.
30. Lindemer, E.R., D.N. Greve, B. Fischl, J.C. Augustinack, D.H. Salat, and I. Alzheimer’s Disease Neuroimaging, Differential Regional Distribution of Juxtacortical White Matter Signal Abnormalities in Aging and Alzheimer’s Disease. J Alzheimers Dis, 2017. 57(1): p. 293-303.
31. Guzman, V.A., O.T. Carmichael, C. Schwarz, et al., White matter hyperintensities and amyloid are independently associated with entorhinal cortex volume among individuals with mild cognitive impairment. Alzheimers Dement, 2013. 9(5 Suppl): p. S124-31.
32. Lee, S., F. Viqar, M.E. Zimmerman, et al., White matter hyperintensities are a core feature of Alzheimer’s disease: Evidence from the dominantly inherited Alzheimer network. Ann Neurol, 2016. 79(6): p. 929-39.
33. Garnier-Crussard, A., S. Bougacha, M. Wirth, et al., White matter hyperintensities across the adult lifespan: relation to age, Abeta load, and cognition. Alzheimers Res Ther, 2020. 12(1): p. 127.
34. Birdsill, A.C., R.L. Koscik, E.M. Jonaitis, et al., Regional white matter hyperintensities: aging, Alzheimer’s disease risk, and cognitive function. Neurobiol Aging, 2014. 35(4): p. 769-76.
35. Taylor, W.D., D.C. Steffens, J.R. MacFall, et al., White matter hyperintensity progression and late-life depression outcomes. Arch Gen Psychiatry, 2003. 60(11): p. 1090-6.
36. McNay, E.C. and J. Pearson-Leary, GluT4: A central player in hippocampal memory and brain insulin resistance. Exp Neurol, 2020. 323: p. 113076.
37. Muhic, M., N. Vardjan, H.H. Chowdhury, R. Zorec, and M. Kreft, Insulin and Insulin-like Growth Factor 1 (IGF-1) Modulate Cytoplasmic Glucose and Glycogen Levels but Not Glucose Transport across the Membrane in Astrocytes. J Biol Chem, 2015. 290(17): p. 11167-76.
38. de la Monte, S.M., Insulin resistance and Alzheimer’s disease. BMB Rep, 2009. 42(8): p. 475-81.
39. Arnold, S.E., Z. Arvanitakis, S.L. Macauley-Rambach, et al., Brain insulin resistance in type 2 diabetes and Alzheimer disease: concepts and conundrums. Nat Rev Neurol, 2018. 14(3): p. 168-181.
40. Suzanne, M., Insulin resistance and neurodegeneration: progress towards the development of new therapeutics for Alzheimer’s disease. Drugs, 2017. 77(1): p. 47-65.
41. Lochhead, J.J., D.J. Wolak, M.E. Pizzo, and R.G. Thorne, Rapid transport within cerebral perivascular spaces underlies widespread tracer distribution in the brain after intranasal administration. J Cereb Blood Flow Metab, 2015. 35(3): p. 371-81.
42. Born, J., T. Lange, W. Kern, G.P. McGregor, U. Bickel, and H.L. Fehm, Sniffing neuropeptides: a transnasal approach to the human brain. Nat Neurosci, 2002. 5(6): p. 514-6.
43. Craft, S., L.D. Baker, T.J. Montine, et al., Intranasal insulin therapy for Alzheimer disease and amnestic mild cognitive impairment: a pilot clinical trial. Arch Neurol, 2012. 69(1): p. 29-38.
44. de la Monte, S.M. and P. Grammas, Insulin Resistance and Oligodendrocyte/Microvascular Endothelial Cell Dysfunction as Mediators of White Matter Degeneration in Alzheimer’s Disease, in Alzheimer’s Disease, T. Wisniewski, Editor. 2019: Brisbane (AU).
45. Ashrafian, H., E.H. Zadeh, and R.H. Khan, Review on Alzheimer’s disease: Inhibition of amyloid beta and tau tangle formation. Int J Biol Macromol, 2020. 167: p. 382-394.
46. Schmidt, P., C. Gaser, M. Arsic, et al., An automated tool for detection of FLAIR-hyperintense white-matter lesions in Multiple Sclerosis. Neuroimage, 2012. 59(4): p. 3774-83.
47. Schwarz, C., J. Gunter, C. Ward, et al. The mayo clinic adult lifespan template (MCALT): better quantification across the lifespan. in Alzheimer’s Association International Conference.[Google Scholar]. 2017.
48. Jack, C.R., Jr., H.J. Wiste, T.M. Therneau, et al., Associations of Amyloid, Tau, and Neurodegeneration Biomarker Profiles With Rates of Memory Decline Among Individuals Without Dementia. JAMA, 2019. 321(23): p. 2316-2325.
49. Brickman, A.M., F.A. Provenzano, J. Muraskin, et al., Regional white matter hyperintensity volume, not hippocampal atrophy, predicts incident Alzheimer disease in the community. Arch Neurol, 2012. 69(12): p. 1621-7.
50. Tubi, M.A., F.W. Feingold, D. Kothapalli, et al., White matter hyperintensities and their relationship to cognition: Effects of segmentation algorithm. Neuroimage, 2020. 206: p. 116327.
51. Gouw, A.A., A. Seewann, W.M. van der Flier, et al., Heterogeneity of small vessel disease: a systematic review of MRI and histopathology correlations. J Neurol Neurosurg Psychiatry, 2011. 82(2): p. 126-35.
52. Carmichael, O., C. Schwarz, D. Drucker, et al., Longitudinal changes in white matter disease and cognition in the first year of the Alzheimer disease neuroimaging initiative. Arch Neurol, 2010. 67(11): p. 1370-8.
53. Heo, J.H., S.T. Lee, C. Kon, H.J. Park, J.Y. Shim, and M. Kim, White matter hyperintensities and cognitive dysfunction in Alzheimer disease. J Geriatr Psychiatry Neurol, 2009. 22(3): p. 207-12.
54. Mitchell, A.J., CSF phosphorylated tau in the diagnosis and prognosis of mild cognitive impairment and Alzheimer’s disease: a meta-analysis of 51 studies. J Neurol Neurosurg Psychiatry, 2009. 80(9): p. 966-75.
55. Hampel, H., A. Goernitz, and K. Buerger, Advances in the development of biomarkers for Alzheimer’s disease: from CSF total tau and Abeta(1-42) proteins to phosphorylated tau protein. Brain Res Bull, 2003. 61(3): p. 243-53.
56. Kandel, B.M., B.B. Avants, J.C. Gee, et al., White matter hyperintensities are more highly associated with preclinical Alzheimer’s disease than imaging and cognitive markers of neurodegeneration. Alzheimers Dement (Amst), 2016. 4: p. 18-27.
57. Marnane, M. and G.Y. Hsiung, Could Better Phenotyping Small Vessel Disease Provide New Insights into Alzheimer Disease and Improve Clinical Trial Outcomes? Curr Alzheimer Res, 2016. 13(7): p. 750-63.
58. Pietroboni, A.M., T. Carandini, A. Colombi, et al., Amyloid PET as a marker of normal-appearing white matter early damage in multiple sclerosis: correlation with CSF beta-amyloid levels and brain volumes. Eur J Nucl Med Mol Imaging, 2019. 46(2): p. 280-287.
59. Walsh, P., C.H. Sudre, C.M. Fiford, et al., CSF amyloid is a consistent predictor of white matter hyperintensities across the disease course from aging to Alzheimer’s disease. Neurobiol Aging, 2020. 91: p. 5-14.
60. Tosto, G., M.E. Zimmerman, J.L. Hamilton, O.T. Carmichael, A.M. Brickman, and I. Alzheimer’s Disease Neuroimaging, The effect of white matter hyperintensities on neurodegeneration in mild cognitive impairment. Alzheimers Dement, 2015. 11(12): p. 1510-1519.
61. Kester, M.I., J.D. Goos, C.E. Teunissen, et al., Associations between cerebral small-vessel disease and Alzheimer disease pathology as measured by cerebrospinal fluid biomarkers. JAMA Neurol, 2014. 71(7): p. 855-62.
62. Osborn, K.E., D. Liu, L.R. Samuels, et al., Cerebrospinal fluid beta-amyloid42 and neurofilament light relate to white matter hyperintensities. Neurobiol Aging, 2018. 68: p. 18-25.
63. Kellar, D. and S. Craft, Brain insulin resistance in Alzheimer’s disease and related disorders: mechanisms and therapeutic approaches. Lancet Neurol, 2020. 19(9): p. 758-766.
64. Parks, J.K., T.S. Smith, P.A. Trimmer, J.P. Bennett, Jr., and W.D. Parker, Jr., Neurotoxic Abeta peptides increase oxidative stress in vivo through NMDA-receptor and nitric-oxide-synthase mechanisms, and inhibit complex IV activity and induce a mitochondrial permeability transition in vitro. J Neurochem, 2001. 76(4): p. 1050-6.
65. Kim, H.C., K. Yamada, A. Nitta, et al., Immunocytochemical evidence that amyloid beta (1-42) impairs endogenous antioxidant systems in vivo. Neuroscience, 2003. 119(2): p. 399-419.
66. Xu, J., S. Chen, S.H. Ahmed, et al., Amyloid-beta peptides are cytotoxic to oligodendrocytes. J Neurosci, 2001. 21(1): p. RC118.
67. Pak, K., S.L. Chan, and M.P. Mattson, Presenilin-1 mutation sensitizes oligodendrocytes to glutamate and amyloid toxicities, and exacerbates white matter damage and memory impairment in mice. Neuromolecular Med, 2003. 3(1): p. 53-64.
68. Verbeek, M.M., D.J. Ruiter, and R.M. de Waal, The role of amyloid in the pathogenesis of Alzheimer’s disease. Biol Chem, 1997. 378(9): p. 937-50.
69. Schultz, N., K. Brannstrom, E. Byman, et al., Amyloid-beta 1-40 is associated with alterations in NG2+ pericyte population ex vivo and in vitro. Aging Cell, 2018. 17(3): p. e12728.
70. Park, L., K. Hochrainer, Y. Hattori, et al., Tau induces PSD95-neuronal NOS uncoupling and neurovascular dysfunction independent of neurodegeneration. Nat Neurosci, 2020. 23(9): p. 1079-1089.
71. Frisoni, G.B., N.C. Fox, C.R. Jack, Jr., P. Scheltens, and P.M. Thompson, The clinical use of structural MRI in Alzheimer disease. Nat Rev Neurol, 2010. 6(2): p. 67-77.
72. Wirth, M., A. Bejanin, R. La Joie, et al., Regional patterns of gray matter volume, hypometabolism, and beta-amyloid in groups at risk of Alzheimer’s disease. Neurobiol Aging, 2018. 63: p. 140-151.
73. Smith, A.D., S.M. Smith, C.A. de Jager, et al., Homocysteine-lowering by B vitamins slows the rate of accelerated brain atrophy in mild cognitive impairment: a randomized controlled trial. PLoS One, 2010. 5(9): p. e12244.
74. Turner, R.S., R.G. Thomas, S. Craft, et al., A randomized, double-blind, placebo-controlled trial of resveratrol for Alzheimer disease. Neurology, 2015. 85(16): p. 1383-91.



E. Tsoy1, S. Zygouris2,3,4, K.L. Possin1,4

1. Department of Neurology, Memory and Aging Center, University of California San Francisco, San Francisco, California, USA; 2. School of Medicine, Aristotle University of Thessaloniki, Thessaloniki, Greece; 3. Network Aging Research, Heidelberg University, Heidelberg, Germany; 4. Global Brain Health Institute, University of California San Francisco, San Francisco, California, USA; Trinity College Dublin, The University of Dublin, Dublin 2, Ireland

Corresponding Author: Katherine L. Possin, PhD, Associate Professor in Residence, Department of Neurology, University of California San Francisco, Memory and Aging Center, Box 1207, 675 Nelson Rising Lane, Suite 190, San Francisco, CA 94158, Tel: 415-476-1889, E-mail:
J Prev Alz Dis 2021;
Published online March 24, 2021,



Early diagnosis of cognitive disorders in older adults is a major healthcare priority with benefits to patients, families, and health systems. Rapid advances in digital technology offer potential for developing innovative diagnostic pathways to support early diagnosis. Brief self-administered computerized cognitive tools in particular hold promise for clinical implementation by minimizing demands on staff time. In this study, we conducted a systematic review of self-administered computerized cognitive assessment measures designed for the detection of cognitive impairment in older adults. Studies were identified via a systematic search of published peer-reviewed literature across major scientific databases. All studies reporting on psychometric validation of brief (≤30 minutes) self-administered computerized measures for detection of MCI and all-cause dementia in older adults were included. Seventeen studies reporting on 10 cognitive tools met inclusion criteria and were subjected to systematic review. There was substantial variability in characteristics of validation samples and reliability and validity estimates. Only 2 measures evaluated feasibility and usability in the intended clinical settings. Similar to past reviews, we found variability across measures with regard to psychometric rigor and potential for widescale applicability in clinical settings. Despite the promise that self-administered cognitive tests hold for clinical implementation, important gaps in scientific rigor in development, validation, and feasibility studies of these measures remain. Developments in technology and biomarker studies provide potential avenues for future directions on the use of digital technology in clinical care.

Key words: Computerized cognitive assessment, early detection, mild cognitive impairment, dementia, psychometrics.



Dementia remains a widely underdiagnosed condition, both in Western countries (1-5) and globally (6). In light of projected increases in prevalence and burden (7) of dementia, innovative solutions in diagnosis and clinical care of dementia will be critical to alleviate the impact of these changes on public healthcare systems. Most experts agree that underdiagnosed dementia is a major gap in care because early detection of cognitive decline in older adults with cognitive symptoms (i.e., patient’s concerns, informant concerns, etc.) is beneficial for both patients and their caregivers (8-11). Targeted evaluation of cognitive impairment can facilitate early detection of cognitive disorders, which in turn can promote patient safety and wellbeing through more informed medication management, implementation of comprehensive care plans, introduction of lifestyle modifications, improved management of symptoms, and the opportunity to participate in clinical trials (8-11). Additionally, earlier diagnosis of dementia may reduce healthcare costs by decreasing long-term care expenditures with projected economic benefits to the affected individuals and to the public health care systems (12-14).
Primary care providers are in a position to first detect cognitive decline because of their established relationships with their patients. In the United States, cognitive impairment detection in primary care is supported by the Medicare Annual Wellness Visit and a new billing code for cognitive assessment and care planning (11). However, based on a survey of 1,000 PCPs and 1,954 older adults conducted by the Alzheimer’s Association, nearly all PCPs (94%) recognized that routine cognitive assessments were important but only 16% of the older adults said that they received routine cognitive assessments (7). Moreover, the survey found that when cognitive assessments are performed in primary care, paper and pencil assessments are almost always used (7). Thus, the most commonly used instruments for detection of cognitive impairment by the PCPs were the Mini Mental State Examination (MMSE; 80%), the Clock Drawing Test (64%), and the Mini-Cog (52%) (7). These conventional paper-based evaluation tools are accurate at detecting dementia but have poorer sensitivity to milder forms of cognitive impairment (15). An additional and important limitation of these conventional tools is that administration, scoring, interpretation, and documentation require substantial clinical staff time. Indeed, among the commonly reported reasons to not conduct the screening were insufficient time during visits and lack of confidence in performing the evaluation (7, 16).
Computerized tools present both exciting potential advantages and significant challenges for improving the detection of cognitive impairment. Older adults endorse both eagerness to use technology (17) and actual usage of technology in their daily lives (18), particularly touchscreen devices, which allow for direct interaction and have lower motor demands and relative ease of use (17, 19). Technology-based assessments offer enhanced precision of measurement and scoring, instant automated scoring and interpretation, standardized administration, enhanced stimulus presentation, availability of multiple alternate forms to minimize practice effects, and potential for adaptive capabilities and more sophisticated algorithms (20-23). Additionally, computerized tools have been discussed as being more cost-effective, particularly with respect to materials and supplies.
Computerized measures also present with challenges related to examinee variables (familiarity with technology, attitude and anxiety towards technology) (24) and technological issues (variability in hardware and software characteristics, data and privacy issues, data charges and internet access) (20, 22, 23). Additionally, a number of past reviews highlighted the lack of adequately established psychometric standards, limited or unfamiliar response modality, and poorly designed user interface (22, 25). A number of studies also demonstrated that computerized measures failed to demonstrate equivalence between the examinee’s experience of computer versus traditional test administration (e.g., participants performed worse on electronic version of the Montreal Cognitive Assessment [MoCA] compared to paper MoCA) (26).
Although most brief cognitive assessments designed for primary care use are examiner-administered, self-administered instruments have the potential to minimize costs and practice barriers related to training and staffing costs (15, 27), and to support social distancing requirements during the COVID-19 pandemic. Patients could complete a self-administered brief cognitive assessment at the clinic prior to or following a provider appointment, or at home on their own device. Additional benefits include minimization of examiner effects and greater accessibility of the tool for patients in remote locations (23). If the self-administered test is translated into different languages, accessibility may be increased for patients who do not speak English because interpreter may not be required. Finally, there is some evidence that absence of an examiner may reduce observer-related stress and increase respondents’ openness during administration, although these findings were reported only in studies on unsupervised surveys (27). While self-administered assessments share the same challenges with examiner-administered computerized assessments, additional challenges include lack of monitoring to ensure response validity (compliance, effort, motivation), lack of support should the patient need help with a task or runs into technological issues, and loss of qualitative data available from a conventional in-person evaluation (25). Additionally, a bring-your-own-device paradigm may present additional challenges related to potential technological differences to ensure consistent stimuli presentation and reaction time measurement, such as screen size and resolution, operating system, central processing unit capacity, etc. (23).
In this study, we conducted a systematic review of studies on self-administered computerized assessments designed to detect mild cognitive impairment (MCI) and dementia in older adults, discuss benefits and weaknesses related to their use, and provide practical recommendations and considerations regarding implementation of these measures into clinical practice.




A systematic search of published literature was conducted from February 1, 2020 to April 20, 2020. Databases searched included PubMed, EMBASE, Web of Science, and PsycINFO. Example search items used were “computerized” or “tablet,” and “cognitive assessment” or “cognitive screen,” and “older adult” or “geriatric,” and “dementia” or “cognitive impairment.” Specific search strings for each database are included in Supplementary Methods. Additional search filters included 1) original peer-reviewed articles (not book chapters, abstracts or conference papers, unpublished dissertations, or review studies), 2) studies published on or after January 1, 2000, and 3) studies published in English language.

Inclusion and exclusion criteria

Inclusion criteria were: 1) studies including a control sample and a clinical (MCI or all-cause dementia diagnosed based on published consensus criteria; 28-34) sample of older adults (ages 50 years and above); 2) studies reporting on a brief (administration time of 30 minutes or less) computerized cognitive assessment tool; and 3) studies reporting on psychometric characteristics of the measure, including reliability and validity indices.
Exclusion criteria were: 1) studies reporting on the use of the cognitive assessment tool in individuals with medical conditions other than MCI or dementia (e.g., schizophrenia, multiple sclerosis, etc.); 2) studies reporting on computerized instruments that were not cognitive assessment tools (e.g., informant surveys, functional questionnaires, etc.); 3) studies reporting on computerized instruments that required a dedicated hardware platform for use (e.g., virtual reality sets, hardware kiosks, etc.) due to potential barriers of implementing these modalities in clinical settings; 4) studies reporting on computerized cognitive measures that were not validated in English; 5) studies reporting on computerized cognitive measures that assess a single cognitive domain; and 6) studies reporting on instruments that required an examiner to administer the tests.

Data extraction

To minimize selection bias, two authors (ET and SZ) independently conducted database searches using same search terms and reviewed titles and abstracts for inclusion criteria. Same two authors then conducted a full-text review of screened-in studies for exclusion criteria. Data extracted from the final set of studies included in the review were: 1) platform; 2) level of supervision required; 3) administration time; 4) characteristics of the validation samples; and 5) psychometric indices. We extracted additional data on commercial availability of the tool, requirements for devices, automated reporting of results, available languages, and number of associated publications based on review of bibliography and dedicated websites of the measures when available. A brief e-mail survey (Supplementary Methods) was also sent out to test developers to collect additional information. Any disagreements were resolved via consensus discussions with a third reviewer (KLP).

Quality assessment

The quality of the studies was assessed using a scale specifically designed for this study in order to capture important considerations and factors for self-administered cognitive assessments (Table 1). Development of the criteria included in the scale was based on prior works on computerized cognitive tools (20). Specifically, we assessed the measures based on the following criteria: comprehensive assessment of the core cognitive domains, size of the validation sample and use of standard diagnostic criteria for identifying participants with cognitive impairment, reliability and validity indices, degree to which an examiner is involved in the testing process in relation to fully automated procedures, current availability for clinical use including any requirements for purchase of a dedicated device, availability of offered tests in multiple languages for participants/patients whose first language is other than English, presence or absence of feasibility studies in the intended settings, issues related to data security and compliance with regulations, and comprehensiveness of the delivery of test results (Table 1). Any ambiguity or disagreements were resolved via a consensus agreement with the third author (KLP).



The search identified 11,617 citations which, after removal of duplicates, resulted in 9,986 unique records. Seventy studies, which were selected from the initial screening process, were further assessed for eligibility based on full-text review. The overall trends in peer-reviewed published studies on computerized tools which were included in the full-text review are presented in Figure 1. In total, 17 studies reporting on 10 self-administered computerized tools were included in the review (for PRISMA flowchart, see Supplementary Results).

Figure 1. Number of published peer-review studies included in the full-text review from January 1, 2000 to April 20, 2020

Abbreviations: SA, self-administered. Blue bars represent the number of studies that met inclusion criteria but were not self-administered, while orange bars represent the number of studies that met inclusion criteria and were self-administered.

Table 1. Quality assessment scale


The measures included in this review were Computer Assessment of Memory and Cognitive Impairment (CAMCI; 35), Computer-Administered Neuropsychological Screen for Mild Cognitive Impairment (CANS-MCI; 36,37), Computerized Cognitive Screening (CCS; 38), CNS Vital Signs (CNSVS; 39,40), Computerized Self Test (COGSelfTest; 41), CogState (CogState; 42-44), CogState Brief Battery (CogState BB; 45-48), Cognitive Testing on Computer (C-TOC; 49), digitally translated Self-Administered Gerocognitive Examination (eSAGE; 50), and an untitled test developed by Kluger et al. (51). Results of the quality assessment ratings of the included tools are reported in Table 2.

Table 2. Quality assessment ratings of included measures

Abbreviations: CAMCI, Computer Assessment of Memory and Cognitive Impairment; CANS-MCI, Computer-Administered Neuropsychological Screen for Mild Cognitive Impairment; CCS, Computerized Cognitive Screening; CNSVS, CNS Vital Signs; COGSelfTest, Computerized Self Test; CogState BB, CogState Brief Battery; C-TOC, Cognitive Testing on Computer; eSAGE, digitally translated Self-Administered; Gerocognitive Examination.


Tool characteristics

Detailed characteristics of the measures included in the review are reported in Table 3. Of the 10 tools included in the review, 3 were available only on a personal computer (PC) platform (CNSVS, COSSelfTest, untitled test), 2 only on a tablet (CAMCI, CCS), and 4 on both platforms (CANS-MCI, CogState, CogState BB, C-TOC, eSAGE). CogState measures were available on an unrestricted set of devices, while most other tools had some restrictions, such as requiring touchscreen capabilities (CANS-MCI, CCS, eSAGE), keyboard input (CNSVS, COGSelfTest, untitled test), or a specific set of devices (CAMCI, C-TOC). Regarding level of supervision, CAMCI and CANS-MCI were designed to be administered in medical and research settings and test developers do not recommend at-home remote testing. Additionally, C-TOC and eSAGE require a trained examiner for scoring. Administration times varied across measures averaging at approximately 15-20 minutes across tools. Commercial availability was a common characteristic with 8/10 measures, except for C-TOC and untitled test, available for purchase. Finally, more than half (6/10) of the measures had fewer than 5 peer-reviewed published studies on the use of the measure across any age groups or clinical populations, while the remaining 4 tools (CAMCI, CNSVS, CogState, CogState BB) were researched more widely with at least 10 peer-reviewed published studies.

Table 3. Summary of the features of included measures

Abbreviations: CAMCI, Computer Assessment of Memory and Cognitive Impairment; CANS-MCI, Computer-Administered Neuropsychological Screen for Mild Cognitive Impairment; CCS, Computerized Cognitive Screening; CNSVS, CNS Vital Signs; COGSelfTest, Computerized Self Test; CogState BB, CogState Brief Battery; C-TOC, Cognitive Testing on Computer; eSAGE, digitally translated Self-Administered Gerocognitive Examination; ND, no data; PC, personal computer; *Includes peer-reviewed published journal articles across all age groups and clinical populations.


Validation samples

Characteristics of the validation samples are presented in Table 4. There was substantial variability in sample sizes across studies, and only 2/10 instruments (CAMCI, CogState BB) were validated in large cohorts with at least 50 or more participants in each diagnostic group. Studies on more than half of the instruments (6/10) used published criteria to classify participants into diagnostic groups, while 3 measures were validated in samples classified by scores on standard cognitive testing (CAMCI, CNSVS, CogState), and the study on the untitled test did not provide sufficient details regarding diagnostic criteria used. Additionally, studies were varied with regard to selection of cognitively normal participants ranging from recruitment of spouses of individuals with MCI and dementia as controls to comprehensive assessment of control subjects. With regard to demographic characteristics of validation samples, the vast majority of participants were non-Hispanic White with 12 or more years of education (educational attainment was not reported for CNSVS and untitled test).

Table 4. Summary of the psychometric properties of included measures

Abbreviations: AD, Alzheimer’s disease; CAMCI, Computer Assessment of Memory and Cognitive Impairment; CANS-MCI, Computer-Administered Neuropsychological Screen for Mild Cognitive Impairment; CCS, Computerized Cognitive Screening; CIND, cognitively impaired no dementia; CN, cognitively normal; CNSVS, CNS Vital Signs; COGSelfTest, Computerized Self Test; CogState BB, CogState Brief Battery; C-TOC, Cognitive Testing on Computer; DEM, dementia; EF, executive functions; eSAGE, digitally translated Self-Administered Gerocognitive Examination; h, hours; m, months; MCI, mild cognitive impairment; ND, no data; PC, personal computer; w, weeks; *Reported only in the dementia group


Psychometric properties

Test-retest reliability was reported on 6/10 tools (Table 4), and there was substantial variability in the reported indices both across and within individual measures by constituent subtests. Reported reliability coefficients of 5/10 measures (CANS-MCI, CNSVS, COGSelfTest, CogState, CogState BB) were consistently within ranges of moderate to high stability based on standard psychometric criteria (52). The range of time intervals for test-retest reliability studies also varied ranging from 2 hours to 12 months. Internal consistency estimates were reported on 3/10 tools (CANS-MCI, CCS, COGSelfTest) with coefficients ranging from .43-.97.
Table 4 presents concurrent validity estimates with either paper-and-pencil brief cognitive assessments or conventional neuropsychological tests which differed by site and study design for 9/10 measures (not reported for CAMCI). Similar to reliability findings, these indices varied significantly across and within measures with most tools demonstrating moderate degree of concurrent validity with standard tests. Across studies, highest concurrent validity estimates were mostly reported with standard brief global assessments (e.g., MMSE), while domain-specific concurrent estimates were more varied. Criterion validity estimates were reported for all measures including discriminant analyses results for 9/10 tools and mean group differences on 1 tool (Table 4). Studies on 6/10 instruments (CAMCI, CANS-MCI, CNSVS, CogState, CogState BB, eSAGE) reported on discriminant analyses between control and MCI groups with sensitivity indices ranging from .41 to .90 and corresponding specificity estimates of .64-.94. As expected, the indices for distinguishing cognitively normal and dementia groups reported on 3/10 tools (CCS, CNSVS, CogState BB) were slightly higher with a sensitivity range of .53-.94 and a specificity range of .50-.94. Finally, discrimination between cognitively normal and impaired (MCI and dementia combined) groups was reported for 3/10 instruments (COGSelfTest, eSAGE, untitled test) with the sensitivity range of .71-.99 and the specificity range of .72-.95. The remaining study on C-TOC reported on criterion validity in the form of mean group differences suggesting significantly lower performance on tasks of episodic memory, executive functions, and speed measures in cognitively impaired no dementia group compared to controls.

Delivery of results and available languages

Automated delivery of results is a feature of 6/10 tools (CAMCI, CANS-MCI, CNSVS, CogState, CogState BB, eSAGE), all of which include interpretation of results based on normative adjustments and differ in presentation of the results (Table 3). Automated reports on 2 measures (eSAGE, CANS-MCI) are reviewed by a trained professional prior to generation and the delivery of the results is thus not immediate. Additionally, CANS-MCI reports include recommendations to physicians on next steps and potential areas for intervention, and eSAGE has functionality to generate patient-facing reports. User-friendliness of the automated reports for non-specialist physicians was empirically examined only for CAMCI (53).
Data on available languages was reported on 6/10 tools (Table 3). Out of these, CAMCI, C-TOC, and untitled test are available only in English, eSAGE is available in English and Spanish, CANS-MCI is available in 4 languages, and CNSVS, CogState, and CogState BB are available in more than 40 languages. Additionally, availability of peer-reviewed studies on validation of non-English versions of these tools for detection of cognitive disorders in older adults varied substantially with the majority of published studies conducted in English-speaking samples.



In this systematic review, we evaluated 10 brief self-administered computerized cognitive assessment measures designed to detect cognitive disorders in older adults. Similar to past reviews of computerized cognitive tools (20, 21), we found significant variability across measures with regard to characteristics and design of the tools, sizes of validation samples, availability in different languages, and psychometric qualities, all of which are crucial considerations for potential widescale implementation of these measures in clinical care. Specifically, we found that few of the reviewed measures were validated in sufficiently large samples (CAMCI, CogState BB) and are available in multiple languages (CANS-MCI, CNSVS, CogState, CogState BB). Test-retest reliability, which is critical for self-administered tools aiming to monitor cognitive functions, was reported only on 60% of the tools, and internal consistency measures were reported on even fewer measures. While almost all reviewed measures reported data on concurrent validity, the estimates for several individual domain subtests within some tools were low. These findings are concerning, particularly when considering the need for a battery to distinguish among different types of MCI and dementia and inform differential diagnoses in non-specialty settings (16). On the other hand, we found that most measures required minimal involvement of an examiner in test administration and scoring of results and were available as standalone applications on several device types (e.g., PC, tablet computer, etc.). These features are important benefits of self-administered computerized tools, particularly if additional built-in functionality for integration of results into electronic medical records (EMR) systems is developed (54). In general, despite the promise that self-administered cognitive tests hold for clinical applications, important gaps in scientific rigor in development, validation, and feasibility studies of these measures remain. Below we discuss critical areas of need for future development and validation of self-administered cognitive measures that would facilitate their potential for widescale clinical implementation.
One of the most critical gaps identified in the current review is the size and demographic constitution of the validation samples. In particular, several studies included in this review included fairly small (<50 participants in each diagnostic group) validation samples, and the majority of validations samples were comprised of White, highly educated individuals. Because we did not identify systematic reporting of the power analyses for detecting main effects in the reviewed studies, we applied a generous estimate of 50 participants per group as part of our criteria. However, given the variability in statistical approaches used in these studies, reporting of robust power estimations would not only support the overall results but also ensure transparency, comparability, and generalizability of results across cohorts.
Another important finding of this review is the scarcity of feasibility and implementation studies of self-administered instruments in care settings. In contrast to highly standardized research settings, self-administration of cognitive assessments in the real world may be subject to interruptions and practical limitations such as time and space, which could be detrimental not only to feasibility but also to the validity of results (22). Some domains, such as orientation, may not be applicable for self-administration altogether, as it would be difficult to ensure the fidelity of responses on such tasks in absence of examiner. Given these considerations, research on development and validation of self-administered computerized measures must be supported by well-designed feasibility and implementation studies, which will critically inform the clinical utility of these measures in intended settings. Specifically, feasibility and implementation studies have the potential to identify facilitators and barriers to clinical applications, inform development of optimal diagnostic and care pathways, and, based on the insights from 2 measures (CAMCI, 53,55 and CogState BB, 56) studied in clinical settings, are critical for informing targeted solutions for individual practices.
The automated delivery of results is key to the clinical utility of computerized tools. To facilitate integration of self-administered tests in non-specialty settings, they should have easy-to-interpret, safe automated report delivery, which would ideally inform the provider on follow-up care and diagnostic considerations based on evidence-based practice guidelines (54). Out of the measures reviewed, only CANS-MCI features an automated report that provides such recommendations to physicians. Moreover, a study on CAMCI with primary care physicians (53) suggested that providers expressed a need for training in interpretation of the report, which highlights the need for refinement of automated reporting and empirical studies on non-specialty providers’ attitudes and perceptions of cognitive testing results.
With regard to patient-level characteristics, there are number of critical considerations, particularly given the dearth of normative or validation data in older adults who are racially/ethnically diverse and have low educational attainment. Importantly, one of the prior studies on CogState BB reported that older adults with lower education were less likely to meet the integrity criteria on 3/5 subtests of the battery (47). This is a major issue given that one of the most promising potentials of self-administered cognitive assessments is supporting services in remote areas and populations less likely to seek specialty evaluations. Moreover, numerous studies suggest that older adults in the U.S. who report Hispanic ethnicity, non-White race, or low education are at a higher risk for neurodegenerative diseases (7) and experience significant disparities in healthcare access and delivery (57). Well-validated self-administered assessments may help substantially reduce these disparities given their potential to deliver tests in different languages (23) but only if they undergo rigorous scientific and cross-cultural validation development. In addition to language and education variables, it is important to validate computerized tools across socioeconomic groups, as past evidence suggests that older adults with lower socioeconomic status reported lower levels of intention to use computerized cognitive testing (24). Finally, successful clinical implementation of even the most well-validated tools would likely require continuous efforts for education and outreach to patients belonging to underrepresented groups as well as their medical providers and families.
Another important variable to consider for self-administration of computerized cognitive measures is the impact of familiarity with technology on test results. While some studies (CCS; 38) reported no differences in test scores between older adults with and without technology experience, these variables do appear to play a significant role through interactions with age (CogState BB; 48) and diagnostic status (C-TOC; 49). Moreover, comparisons between content-equivalent paper and electronic version measures revealed that older adults with no technology experience performed worse on the electronic version of the measure compared to those with digital proficiency (eSAGE; 50). Finally, while most studies examine the associations between familiarity with technology and computerized cognitive testing results on a group-based level, systematic research on the impact of these variables for individual patients is necessary to support utility of self-administered assessments in clinical practice.
Regarding technical considerations, the practice parameters on optimal development and validation of computerized cognitive tools, including issues related to end-user agreements, privacy, data security and reporting (23, 25), are highly relevant to self-administered paradigms. Of particular relevance are challenges related to the use of bring your own device (BYOD) model and dependence on broadband connection, which pose a threat for timing and measurement error and may thus lead to inaccurate interpretation of results. Built-in integrity measures designed to address this challenge are features of some self-administered tools (e.g., CogState; 47,48), but are not widely available across reviewed instruments. Moreover, because of rapidly evolving hardware and operating systems across both PC and tablet platforms, computerized assessments require continuous quality assurance testing and software maintenance investments, and these challenges are greater when many devices (i.e., BYOD models) are supported. Additionally, availability of a measure on multiple devices also requires supporting research to establish the equivalence of normative and psychometric data across different platforms and input parameters, such as touchscreen vs. keyboard response, screen size, etc. Finally, past studies highlighted concerns regarding underreporting of privacy and security safeguards and their limitations on currently available computerized measures (22), and test developers should strive to explicitly disclose any potential consequences of data loss or breaches, particularly for individual patients in clinical settings. As such, collaborative efforts among researchers, funding bodies, industry, policy regulators, and consumers are necessary to develop robust, sustainable platforms supporting optimal levels of security, privacy, confidentiality, and potential functionality of data sharing across sites in order to promote and maintain successful implementation of computerized tools into everyday clinical practice while meeting programming cost demands.
This study has a number of limitations. First, while all attempts were made to conduct a comprehensive search of available literature, our results were limited to studies available in databases searched. Second, due to variability in study design and test statistics, quantitative summary of the findings was not possible. Finally, our review was limited to inclusion of studies that reported on instruments available at least in English language, and a number of promising self-administered computerized cognitive measures validated in non-English cohorts were not considered.
At the same time, a major strength of this study is the scope of reviewed characteristics of the included measures, including not only psychometric qualities but also functional and technological features critical for clinical implementation considerations. Additionally, our review is conducted at a point in time when the need for self-administered cognitive assessments has never been so dire in both clinical and research settings. In light of rapidly developing technologies for identifying disease biomarkers, future studies should examine the associations of a variety of self-administered cognitive assessments with biomarkers of neurodegenerative diseases, particularly given promising existing studies within this line of research (CogState; 58 and Computerized Cognitive Composite [C3]; 59). Additionally, future studies on self-administered cognitive measures in clinical settings should explore optimal implementation paradigms and provider behavior patterns which would be valuable for informing public healthcare policy and efforts to support earlier diagnosis of cognitive disorders in older adults.



In conclusion, this systematic review identified 10 self-administered brief computerized cognitive measures which have a potential for future clinical implementation. Continuous collaborative efforts of different stakeholders are necessary to address the gaps in scientific rigor of development, validation, and implementation studies of these measures.


Funding: This study was supported by the National Institute of Neurological Disorders and Stroke [UG3 NS105557] and the Global Brain Health Institute. The sponsors had no role in the design and conduct of the study; in the collection, analysis, and interpretation of data; in the preparation of the manuscript; or in the review or approval of the manuscript.

Declarations of interest: Elena Tsoy declares no conflicts of interest. Stelios Zygouris declares no conflicts of interest. Katherine L. Possin has received research funding from the NIH, Quest Diagnostics, the Global Brain Health Institute, the Merck Foundation, and the Rainwater Charitable Foundation, consulting fees from ClearView Healthcare Partners and Vanguard, and a speaking fee from Swedish Medical Center.



1. Chodosh J, Petitti DB, Elliott M, et al. Physician recognition of cognitive impairment: evaluating the need for improvement. J Am Geriatr Soc. 2004;52(7):1051-1059. doi:10.1111/j.1532-5415.2004.52301.x
2. Connolly A, Gaehl E, Martin H, Morris J, Purandare N. Underdiagnosis of dementia in primary care: variations in the observed prevalence and comparisons to the expected prevalence. Aging Ment Health. 2011;15(8):978-984. doi:10.1080/13607863.2011.596805
3. Savva GM, Arthur A. Who has undiagnosed dementia? A cross-sectional analysis of participants of the Aging, Demographics and Memory Study. Age Ageing. 2015;44(4):642-647. doi:10.1093/ageing/afv020
4. Bradford A, Kunik ME, Schulz P, Williams SP, Singh H. Missed and delayed diagnosis of dementia in primary care: prevalence and contributing factors. Alzheimer Dis Assoc Disord. 2009;23(4):306-314. doi:10.1097/WAD.0b013e3181a6bebc
5. Amjad H, Roth DL, Sheehan OC, Lyketsos CG, Wolff JL, Samus QM. Underdiagnosis of dementia: an observational study of patterns in diagnosis and awareness in US older adults. J Gen Intern Med. 2018;33(7):1131-1138. doi:10.1007/s11606-018-4377-y
6. Lang L, Clifford A, Wei L, et al. Prevalence and determinants of undetected dementia in the community: a systematic literature review and a meta-analysis. BMJ Open. 2017;7(2):e011146. doi:10.1136/bmjopen-2016-011146
7. Alzheimer’s Association, 2019. Facts and Figures. Accessed June 30, 2020.
8. U.S. Department of Health & Human Services, 2019. National plan to address.Alzheimer’s disease: 2019 update. Accessed August 12, 2020.
9. Schneider J, Jeon S, Gladman JT, Corriveau RA, 2019. ADRD summit 2019 report to the National Advisory Neurological Disorders and Stroke Council. Accessed August 12, 2020.
10. U.S. Centers for Disease and Control Prevention, 2019. Advancing early detection: A healthy brain initiative issue map. Accessed August 12, 2020.
11. Cordell CB, Borson S, Boustani M, et al. Alzheimer’s Association recommendations for operationalizing the detection of cognitive impairment during the Medicare Annual Wellness Visit in a primary care setting. Alzheimers Dement. 2013;9(2):141-150. doi:10.1016/j.jalz.2012.09.011
12. Alzheimer’s Association, 2018. Facts and Figures. Accessed June 30, 2020.
13. Barnett JH, Lewis L, Blackwell AD, Taylor M. Early intervention in Alzheimer’s disease: a health economic study of the effects of diagnostic timing. BMC Neurol. 2014;14:101. doi:10.1186/1471-2377-14-101
14. Weimer DL, Sager MA. Early identification and treatment of Alzheimer’s disease: social and fiscal outcomes. Alzheimers Dement. 2009;5(3):215-226. doi:10.1016/j.jalz.2009.01.028
15. Scott J, Mayo AM. Instruments for detection and screening of cognitive impairment for older adults in primary care settings: a review. Geriatr Nurs. 2018;39(3):323-329. doi:10.1016/j.gerinurse.2017.11.001.
16. Bernstein A, Rogers KM, Possin KL, et al. Dementia assessment and management in primary care settings: a survey of current provider practices in the United States. BMC Health Serv Res. 2019;19(1):919. doi:10.1186/s12913-019-4603-2
17. Vaportzis E, Clausen MG, Gow AJ. Older adults perceptions of technology and barriers to interacting with tablet computers: a focus group study. Front Psychol. 2017;8:1687. doi:10.3389/fpsyg.2017.01687
18. Pew Research Center, 2018. Mobile fact sheet. Accessed June 30, 2020.
19. Canini M, Battista P, Della Rosa PA, et al. Computerized neuropsychological assessment in aging: testing efficacy and clinical ecology of different interfaces. Comput Math Methods Med. 2014;2014:804723. doi:10.1155/2014/804723
20. Wild K, Howieson D, Webbe F, Seelye A, Kaye J. Status of computerized cognitive testing in aging: a systematic review. Alzheimers Dement. 2008;4(6):428-437. doi:10.1016/j.jalz.2008.07.003
21. Zygouris S, Tsolaki M. Computerized cognitive testing for older adults: a review. Am J Alzheimers Dis Other Demen. 2015;30(1):13-28. doi:10.1177/1533317514522852
22. Gates NJ, Kochan NA. Computerized and on-line neuropsychological testing for late-life cognition and neurocognitive disorders: are we there yet? Curr Opin Psychiatry. 2015;28(2):165-172. doi:10.1097/YCO.0000000000000141
23. Parsons TD, McMahan T, Kane R. Practice parameters facilitating adoption of advanced technologies for enhancing neuropsychological assessment paradigms. Clin Neuropsychol. 2018;32(1):16-41. doi:10.1080/13854046.2017.1337932
24. Werner P, Korczyn AD. Willingness to use computerized systems for the diagnosis of dementia: testing a theoretical model in an Israeli sample. Alzheimer Dis Assoc Disord. 2012;26(2):171-178. doi:10.1097/WAD.0b013e318222323e
25. Bauer RM, Iverson GL, Cernich AN, Binder LM, Ruff RM, Naugle RI. Computerized neuropsychological assessment devices: joint position paper of the American Academy of Clinical Neuropsychology and the National Academy of Neuropsychology. Clin Neuropsychol. 2012;26(2):177-196. doi:10.1080/13854046.2012.663001
26. Wallace SE, Donoso Brown EV, Simpson RC, et al. A Comparison of electronic and paper versions of the Montreal Cognitive Assessment. Alzheimer Dis Assoc Disord. 2019;33(3):272-278. doi:10.1097/WAD.0000000000000333
27. Feenstra HE, Vermeulen IE, Murre JM, Schagen SB. Online cognition: factors facilitating reliable online neuropsychological test results. Clin Neuropsychol. 2017;31(1):59-84. doi:10.1080/13854046.2016.1190405
28. Petersen RC, Smith GE, Waring SC, Ivnik RJ, Tangalos EG, Kokmen E. Mild cognitive impairment: clinical characterization and outcome. Arch Neurol. 1999;56(3):303-308. doi:10.1001/archneur.56.3.303
29. Petersen RC, Doody R, Kurz A, et al. Current concepts in mild cognitive impairment. Arch Neurol. 2001;58(12):1985-1992. doi:10.1001/archneur.58.12.1985
30. Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Med. 2004;256(3):183-194. doi:10.1111/j.1365-2796.2004.01388.x
31. Winblad B, Palmer K, Kivipelto M, et al. Mild cognitive impairment – beyond controversies, towards a consensus: report of the International Working Group on Mild Cognitive Impairment. J Intern Med. 2004;256(3):240-246. doi:10.1111/j.1365-2796.2004.01380.x
32. American Psychiatric Association, 2000. Diagnostic and Statistical Manual of Mental Disorders IV Text Revision (DSM-IV-TR). Washington, DC: American Psychiatric Association.
33. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer’s disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease. Neurology. 1984;34(7):939-944. doi:10.1212/wnl.34.7.939
34. McKhann GM, Knopman DS, Chertkow H, et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):263-269. doi:10.1016/j.jalz.2011.03.005
35. Saxton J, Morrow L, Eschman A, Archer G, Luther J, Zuccolotto A. Computer assessment of mild cognitive impairment. Postgrad Med. 2009;121(2):177-185. doi:10.3810/pgm.2009.03.1990
36. Tornatore JB, Hill E, Laboff JA, McGann ME. Self-administered screening for mild cognitive impairment: initial validation of a computerized test battery. J Neuropsychiatry Clin Neurosci. 2005;17(1):98-105. doi:10.1176/jnp.17.1.98
37. Ahmed S, de Jager C, Wilcock G. A comparison of screening tools for the assessment of mild cognitive impairment: preliminary findings. Neurocase. 2012;18(4):336-351. doi:10.1080/13554794.2011.608365
38. Scanlon L, O’Shea E, O’Caoimh R, Timmons S. Usability and validity of a battery of computerised cognitive screening tests for detecting cognitive impairment. Gerontology. 2016;62(2):247-252. doi:10.1159/000433432
39. Gualtieri CT, Johnson LG. Neurocognitive testing supports a broader concept of mild cognitive impairment. Am J Alzheimers Dis Other Demen. 2005;20(6):359-366. doi:10.1177/153331750502000607
40. Gualtieri CT, Johnson LG. Reliability and validity of a computerized neurocognitive test battery, CNS Vital Signs. Arch Clin Neuropsychol. 2006;21(7):623-643. doi:10.1016/j.acn.2006.05.007
41. Dougherty JH Jr, Cannon RL, Nicholas CR, et al. The computerized self test (CST): an interactive, internet accessible cognitive screening test for dementia. J Alzheimers Dis. 2010;20(1):185-195. doi:10.3233/JAD-2010-1354
42. de Jager CA, Schrijnemaekers AC, Honey TE, Budge MM. Detection of MCI in the clinic: evaluation of the sensitivity and specificity of a computerised test battery, the Hopkins Verbal Learning Test and the MMSE. Age Ageing. 2009;38(4):455-460. doi:10.1093/ageing/afp068
43. Hammers D, Spurgeon E, Ryan K, et al. Reliability of repeated cognitive assessment of dementia using a brief computerized battery. Am J Alzheimers Dis Other Demen. 2011;26(4):326-333. doi:10.1177/1533317511411907
44. Lim YY, Jaeger J, Harrington K, et al. Three-month stability of the CogState brief battery in healthy older adults, mild cognitive impairment, and Alzheimer’s disease: results from the Australian Imaging, Biomarkers, and Lifestyle-rate of change substudy (AIBL-ROCS). Arch Clin Neuropsychol. 2013;28(4):320-330. doi:10.1093/arclin/act021
45. Hammers D, Spurgeon E, Ryan K, et al. Validity of a brief computerized cognitive screening test in dementia. J Geriatr Psychiatry Neurol. 2012;25(2):89-99. doi:10.1177/0891988712447894
46. Maruff P, Lim YY, Darby D, et al. Clinical utility of the CogState brief battery in identifying cognitive impairment in mild cognitive impairment and Alzheimer’s disease. BMC Psychol. 2013;1(1):30. doi:10.1186/2050-7283-1-30
47. Mielke MM, Machulda MM, Hagen CE, et al. Performance of the CogState computerized battery in the Mayo Clinic Study on Aging. Alzheimers Dement. 2015;11(11):1367-1376. doi:10.1016/j.jalz.2015.01.008
48. Fredrickson J, Maruff P, Woodward M, et al. Evaluation of the usability of a brief computerized cognitive screening test in older people for epidemiological studies. Neuroepidemiology. 2010;34(2):65-75. doi:10.1159/000264823
49. Jacova C, McGrenere J, Lee HS, et al. C-TOC (Cognitive Testing on Computer): investigating the usability and validity of a novel self-administered cognitive assessment tool in aging and early dementia. Alzheimer Dis Assoc Disord. 2015;29(3):213-221. doi:10.1097/WAD.0000000000000055
50. Scharre DW, Chang SI, Nagaraja HN, Vrettos NE, Bornstein RA. Digitally translated Self-Administered Gerocognitive Examination (eSAGE): relationship with its validated paper version, neuropsychological evaluations, and clinical assessments. Alzheimers Res Ther. 2017;9(1):44. doi:10.1186/s13195-017-0269-3
51. Kluger BM, Saunders LV, Hou W, et al. A brief computerized self-screen for dementia. J Clin Exp Neuropsychol. 2009;31(2):234-244. doi:10.1080/13803390802317559
52. Cohen J, 1988. Statistical power analysis for the behavioral sciences, second edition. Hillside, NJ: Lawrence Erlbaum Associates.
53. Millett G, Naglie G, Upshur R, Jaakkimainen L, Charles J, Tierney MC. Computerized cognitive testing in primary care: a qualitative study. Alzheimer Dis Assoc Disord. 2018;32(2):114-119. doi:10.1097/WAD.0000000000000219
54. Sabbagh MN, Boada M, Borson S, et al. Early detection of mild cognitive impairment (MCI) in primary care. J Prev Alzheimers Dis. 2020;7(3):165-170. doi:10.14283/jpad.2020.21
55. Tierney MC, Naglie G, Upshur R, Moineddin R, Charles J, Jaakkimainen RL. Feasibility and validity of the self-administered computerized assessment of mild cognitive impairment with older primary care patients. Alzheimer Dis Assoc Disord. 2014;28(4):311-319. doi:10.1097/WAD.0000000000000036
56. Adler K, Apple S, Friedlander A, et al. Computerized cognitive performance assessments in the Brooklyn Cognitive Impairments in Health Disparities Pilot Study. Alzheimers Dement. 2019;15(11):1420-1426. doi:10.1016/j.jalz.2019.07.004
57. Babulal GM, Quiroz YT, Albensi BC, et al. Perspectives on ethnic and racial disparities in Alzheimer’s disease and related dementias: update and areas of immediate need. Alzheimers Dement. 2019;15(2):292-312. doi:10.1016/j.jalz.2018.09.009
58. Mielke MM, Machulda MM, Hagen CE, et al. Influence of amyloid and APOE on cognitive performance in a late middle-aged cohort. Alzheimers Dement. 2016;12(3):281-291. doi:10.1016/j.jalz.2015.09.010
59. Papp KV, Rentz DM, Maruff P, et al. The Computerized Cognitive Composite (C3) in an Alzheimer’s Disease Secondary Prevention Trial. J Prev Alzheimers Dis. 2021;8(1):59-67. doi:10.14283/jpad.2020.38.



S. Zhou1, K. Wang2

1. Department of Respiratory and Critical Care Medicine, the Second People’s Hospital of Yibin/West China Yibin Hospital, Sichuan University, Yibin, Sichuan, China; 2. Department of Neurosurgery, Harvard Medical School and Brigham and Women’s Hospital, Boston, MA, USA, ORCID ID is 0000-0002-6958-7677

Corresponding Author: Kanran Wang, MD, Department of Neurosurgery, Harvard Medical School and Brigham and Women’s Hospital, 75 Francis Street, Boston, MA, USA, 02115,
J Prev Alz Dis 2021;
Published online March 24, 2021,



Background: This study aimed to investigate the associations between secondhand smoke exposure and dementia, Alzheimer’s disease (AD) and stroke.
Methods: This prospective study analyzed Framingham Offspring (FHS-OS) cohort participants with parents in the original Framingham Heart Study (FHS) cohort with known smoking status during offspring childhood. Surveillance for incident events, including dementia and stroke, among offspring participants exposed to parental smoking up to the age of 18 years commenced at examination 9 through 2014 and continued for approximately 30 years.
Results: At baseline, a total of 1683 (56.2%) subjects were not exposed to any secondhand smoke, whereas 670 (22.4%) subjects were exposed to 0-1 packs (20 cigarettes)/day, and 640 (21.4%) were exposed to over 1 pack/day. On follow-up (median: 31 years), 2993 patients developed dementia, including 103 with AD dementia and 315 with stroke. After adjusting for a wide range of established risk factors, participants with the highest exposure to secondhand smoke exhibited increased risks of all dementia, AD dementia and stroke compared with individuals with no exposure [HR 2.86 (2.00-4.09) for dementia; HR 3.13 (1.80-5.42) for AD dementia; HR 1.89 (1.37-2.61) for stroke]. The results remained comparable in the subgroup for individuals with median exposure to secondhand smoke.
Conclusion: Exposure to secondhand smoke may be associated with increased risks of dementia, AD dementia and stroke.

Key words: Secondhand smoke, Dementia, Alzheimer disease, stroke, cohort study.



Dementia is a common neurodegenerative disease, and it is estimated that the number of cases will reach approximately 81.1 million by 2040 with a complicated etiology underpinned by genetic and environmental components (1-3). In the past decade, the prevention and exploration of risk factors for dementia has been increasingly receiving attention in the field (4-6).

The smoking-dementia relationship has been investigated by many studies, and cigarette smoking remains one of the most important modifiable risk factors for incident dementia and AD (7, 8). Moreover, exposure to secondhand smoke has also been shown to alter the risk of cardiovascular and metabolic diseases (9-11). However, only a few reports have been published on the effects of secondhand smoke exposure on dementia and AD. A national cross-sectional study reported that exposure to secondhand smoke may be associated with increased odds of cognitive impairment, which relied on self-reported cigarette smoke exposure; therefore, recall bias may have played a role (12). Both stroke and dementia share common risk factors and etiologies (13), but the longitudinal relationship between exposure to secondhand smoke and the risk for stroke was also insufficiently verified. Thus, the potential effects of long-term and accurately evaluated exposure to secondhand smoke on the risk for dementia and stroke remain largely unexplored.
We therefore sought to leverage the multigenerational Framingham Heart Study to test the association between exposure to secondhand smoke and long-term risk (i.e., over 30 years) of dementia and AD and stroke with a detailed review of all medical records and precise assessment of secondhand smoke.



Study Design

We used data from the Framingham Heart Study (FHS), which commenced in 1948 with the enrollment of 5,209 original cohort participants in the Framingham, Massachusetts and Framingham Heart Study offspring (FHS-OS) cohort, which was established in 1970 and includes 5124 individuals who were the offspring of the FHS original cohort and their spouses (14, 15). These cohorts as well as their design and methods were described in greater detail elsewhere (16). Briefly, since their recruitment, participants from the FHS cohort have had serial examinations every 2-4 years and the FHS-OS cohort every 4-8 years, including standardized interviews, physician examinations and laboratory tests. For the present investigation, we included participants of the FHS-OS cohort with at least 1 parent in the FHS original cohort with a known smoking status at any point until his or her offspring reached the age of 18 years. The most recent examination period for both the FHS and FHS-OS cohorts concluded in 2014. This study complied with the Declaration of Helsinki; written informed consent was obtained from all study participants. This current study utilizing the Framingham Heart Study datasets was approved by the National Heart, Lung, and Blood Institute (NHLBI) of National Institutes of Health (NIH).

Smoking Assessment

Smoking was defined in both the FHS and FHS-OS cohorts as participants reporting smoking >1 cigarette daily during the year prior to their study examination. For those participants who reported smoking, the number of mean packs of cigarettes smoked per day was calculated based on the daily number of cigarettes (1 pack representing 20 cigarettes). Parental smoke exposure for the FHS-OS cohort was assessed and defined as the presence of parental smoking (either parent) of greater than 0 mean packs/day at any point in an examination period when his or her offspring participant was between 0 and 18 years of age. To account for a possible dose-response smoking relationship from variable exposure between individual parents, parental smoke exposure was also defined as the summation of the number of cigarettes smoked daily by both the mother and father, which was further categorized into a three-level variable: no exposure, 0-1 pack/day and >1 pack/day. Baseline offspring smoking status was defined as smoking of >0 mean packs per day in Exam 1 of the FHS-OS cohort. Details were further described in the FHS-OS protocol (

Ascertainment of Dementia and AD

Participants in the FHS-OS were under ongoing continuous surveillance for the onset of cognitive impairment and clinical dementia. We related childhood secondhand smoke exposure to the long-term risk of dementia and AD dementia. A diagnosis of dementia was made according to the Diagnostic and Statistical Manual of Mental Disorders, 4th edition. A diagnosis of Alzheimer’s disease (AD) dementia was based on the criteria of the National Institute of Neurological and Communicative Disorders and Stroke and the AD and Related Disorders Association for definite, probable, or possible AD (17, 18).

Ascertainment of Stroke

Stroke incidence was assessed through the continuous monitoring of hospital admissions in Framingham and by reviewing all available outside medical records and interim hospitalizations (19). Stroke was defined as focal neurological symptoms of rapid onset and presumed vascular origin lasting >24 hours or resulting in death within 24 hours. A committee comprising at least 3 FHS investigators, including at least 2 neurologists, adjudicated stroke diagnosis. The committee considered all available medical records, brain imaging, cerebrovascular imaging, and the assessment of the study neurologist who visited the participant.


The following dementia and/or stroke risk factors at Exam 1 in the FHS-OS cohort were utilized as baseline covariates in the study analysis: hypertension, diabetes mellitus, body mass index (BMI), waist circumference, alcohol consumption, current smoking status, blood lipids and education level. Hypertension was confirmed as systolic blood pressure>140 mm Hg, diastolic blood pressure>90 mm Hg, or use of antihypertensive medications. Individuals with fasting plasma glucose concentrations ≥7 mmol/L or who self-reported the use of antidiabetic medications were considered diabetic. BMI was calculated as weight in kilograms divided by the square of the participant’s height in meters at baseline examination. Waist circumference (in inches) was measured at the level of the umbilicus. Current smokers were defined as participants who smoked regularly during the year before the examination (yes or no) as assessed via questionnaire in all serial examination cycles. The educational level was assessed by medical interview. Total cholesterol and low-density lipoprotein cholesterol were measured after an overnight (>10 hours) fast.

Statistical Analysis

Descriptive statistics were performed for the 3 subgroups: participants with no exposure to secondhand smoke, 0-1 pack/day and >1 pack/day. Continuous and ordinal variables are expressed as the mean ± SD or median (interquartile range), respectively. The chi-square test was used to compare categorical variables, and Fisher’s exact test was used for categories with <5 observations. Follow-up for dementia and stroke was from the baseline examination to the time of incident event. For persons with no incident events, follow-up was censored at the time of death or the date the participant was last known to be dementia free. For survival analysis, Cox proportional hazards modeling was applied, and the following covariates were included: crude analysis; model 1 adjusted for age and sex; and model 2 adjusted for age, sex, and the following dementia risk factors: hypertension, smoking, diabetes and body mass index. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated. The same analysis was performed for Alzheimer’s disease and stroke. Analyses were performed using SPSS 23.0 (SPSS Inc., Chicago, IL, USA) and Stata statistical software, version 15 (Stata Corporation, College Station, Texas, USA). A 2-sided P<0.05 was considered statistically significant.




Of the 3765 participants of the FHS Offspring Cohort attending the baseline exam until older than 18 years old, the condition of parental smoking was ascertained in 3545 subjects. Thus, these participants were included in this study. Therefore, 552 subjects were excluded because 498 subjects lacked assessment of dementia or stroke, 4 subjects were diagnosed with dementia at baseline, 3 with stroke, and 50 subjects were lost to follow-up for dementia or stroke. Finally, 2993 individuals could be included in the analyses with 1683 (56.2%) subjects with no secondhand smoke exposure, 670 (22.4%) subjects with exposure to 0-1 packs/day and 640 (21.4%) subjects with over 1 pack/day. (Figure 1)

Figure 1. Selection of Study Participants in the Framingham Heart Study


Table 1 shows the characteristics of those included in the analysis. The three groups descriptively differed with regard to age, sex, systolic blood pressure, diastolic blood pressure, smoking status and history of hypertension. In detail, in trends from the lowest to the highest exposure of secondhand smoking, those exposed to secondhand smoke were more likely to be slightly younger, female, with an increased incidence of blood pressure and hypertension but not diabetes and differ metabolic indicators, including LDL and TC. Finally, those exposed to secondhand smoke were more likely to be smokers themselves, and no difference was shown in alcohol consumption.

Table 1. Characteristics of the study sample of the Framingham Heart Study at baseline secondhand smoke exposure assessment

Abbreviations: BMI, body mass index; WC, waist circumference. LDL, low-density lipoprotein, TC: total cholesterol.


Secondhand smoke exposure and risk for dementia and stroke

Among the FHS-OS cohort with available smoking ascertainment, 239 (8.0%) developed dementia, including 103 with AD dementia over a median follow-up of 31 years (interquartile range, 28 to 32 years), with an overall incidence rate of 3.32 per 1,000 person-years. In addition, 315 (10.5%) cases of incident stroke were identified over a median follow-up of 31 years (interquartile range, 30 to 32 years) with an overall incidence rate of 3.53 per 1,000 person-years.
Risks of incident dementia, AD dementia and stroke according to secondhand smoke exposure are presented in Table 2. After adjusting for offspring age, sex, BMI, diabetes, smoking and hypertension, parental smoking greater than 1 pack/day was associated with significantly increased risks of offspring dementia, AD dementia and stroke [HR 2.86 (2.00-4.09) for dementia; HR 3.13 (1.80-5.42) for AD dementia; HR 1.89 (1.37-2.61) for stroke] compared with subjects without exposure for secondhand smoke. These trends remained the same for participants with parental smoking between 0 and 1 pack/day [HR 1.79 (1.21-2.64) for all dementia; 1.97 (1.09-3.58) for AD dementia; 1.60 (1.17-2.19) for stroke]. In an additional sensitivity analysis, all analyses in the subgroup of nonsmokers in offspring were repeated, and results consistent with the whole population were obtained. (eTable 1 in the supplement) The cumulative incidence curves for all dementia, AD dementia and stroke stratified by groups with different levels of exposure to secondhand smoke showed consistent results with the Cox regression after adjusting for age and sex. (Figure 2.)

Table 2. Cumulative hazards based on secondhand smoke exposure

Abbreviations: AD, Alzheimer’s disease; HR, hazard ratio; CI, confidence interval; Model 1 Sex and Age; Model 2 in addition for hypertension, smoking, diabetes and body mass index.


Figure 1. Adjusted cumulative incidence of dementia and stroke based on secondhand smoke exposure



In this prospective community-based and multigenerational sample with a median follow-up time of almost 30 years, exposure to secondhand smoke was associated with increased risks of offspring dementia, AD dementia and stroke after adjustment for conventional risk factors and accounting for dementia or stroke clustering within families.
These findings highlight potential new mechanistic pathways for dementia and stroke risk that begin during childhood and an association between secondhand smoke exposure and dementia and/or stroke risk. These observations may also provide new information pertinent to smoking cessation and avoidance more than relationship may be mediated by a greater tendency among their posterity of smoking parents to smoke themselves (20), highlighting the harms that may be associated not only to irrelevant strangers but to close and the most vulnerable members of the family.
Despite the published health hazards of smoking and public awareness campaigns to reduce smoking, 52.1% of Chinese men and 14% of US adults continue to smoke (21, 22). Smoking remains the primary cause of preventable death with the number of attributable annual deaths expected to increase to 8 million by 2030 (23). However, we often overlook that secondhand smoke exposure reached 54.3% in the workplace and 57.1% at home in 2015 in China and is responsible for at least 41,000 deaths annually in the US (22, 24). While there have been numerous published reports of the deleterious effects of secondhand smoke exposure on chronic kidney disease (25), cancer and other cardiovascular conditions, such as coronary heart disease (26, 27), the risk of dementia and stroke secondary to is less well defined. A national cross-sectional study from the UK demonstrated that exposure to secondhand smoke may be associated with increased odds of cognitive impairment after adjustment for a wide range of established risk factors for cognitive impairment. A secondary analysis including 970 US participants in the Cardiovascular Health Cognition Study found that exposure to high levels of secondhand smoke alone would increase the risk of dementia in elderly individuals with a history of carotid artery stenosis but not in the general population. A prospective cohort study including 7000 permanent residents from six regions within Zhejiang Province, China showed that passive smoking exposure increased the risk of cognitive impairment in older adults, especially nonsmokers (28). To our knowledge, this is the first study to demonstrate a significant association between secondhand smoke exposure and all dementia, AD dementia and stroke development in a prospective observation of a large-scale cohort with accurate assessments for secondhand smoke based on well-organized original and offspring cohorts.
Given that smoking is widely recognized a risk factor for dementia and stroke, we hypothesize that these mechanisms should be no less relevant during passive smoke inhalation and that this may be even more critical during early human growth and development in the childhood and adolescent years. First, secondhand smoke is highly noxious and contains greater than 250 chemicals known to be harmful or carcinogenic (29, 30). Therefore, such exposure could negatively impact the brain and nerves in a direct manner that could result in dementia and stroke. Moreover, exposure to secondhand smoke is a previously established risk factor for coronary heart disease and diabetes mellitus (31), each of which is an important risk factor for cognitive impairment, dementia and stroke and thus may indirectly affect the risk (32, 33). Finally, regarding basic research, secondhand smoke exposure adversely affects endothelial function and contributes to vasoconstriction, atherogenesis, and thrombosis and may therefore compromise the blood supply to the brain (34, 35). Moreover, endothelial dysfunction may lead to the reduced clearance of β-amyloid protein, which is considered to be related to the pathogenesis of Alzheimer’s disease (36). Taken together, there are several mechanisms through which secondhand smoke could directly and indirectly affect dementia risk. In a large, dementia-free multigenerational cohort, our results provide further evidence that secondhand smoke is associated with an increased risk of all dementia, AD dementia and stroke after adjustment for confounders, including sex, age, BMI, diabetes and hypertension.
The strengths of the study include a large heterogeneous sample size; a unique longitudinal, community-based cohort spanning generations; and long-term and robust follow-up for our outcomes of interest. Moreover, the FHS and FHS-OS cohorts were renowned for their precise ascertainment of relevant covariates that may be potential confounders used in multivariable adjustment.


First, parental smoking status was unclear in some potentially exposed offspring participants, and the number of incident events was small, which may lead to selection bias. Second, the observational nature of our study prevents us from inferring causal links between exposure to secondhand smoke and the risks of dementia and stroke. Finally, although we addressed confounding in numerous ways, we cannot exclude the possibility of residual confounding, especially for certain dementia and/or risk factors that were not available to us based on public databases, such as genotype ApoE4 (37). Therefore, larger studies are needed to replicate and verify our results.



Our study demonstrates that early-life exposure to secondhand smoke was significantly associated with increased risks of dementia, AD dementia and stroke development. International policy debate on exposure to secondhand is a topic of major public health significance. Dementia and stroke are two of the largest global public health challenges facing our aging population. These findings may provide new evidence to reduce the risk of secondhand smoke exposure by enhancing public smoking restriction policies and motivating current smokers to quit as well as motivating potential smokers to avoid smoking altogether to maintain better health.


Acknowledgments: The authors thank the National Heart, Lung, and Blood Institute Framingham Heart Study, Framingham, MA, Harvard Medical School and Brigham and Women’s Hospital, Boston, MA, and the Second People’s Hospital of Yibin/West China Yibin Hospital, Yibin, Sichuan.

Data Availability: Data described in the manuscript, code book, and analytic code will not be made available because the authors are prohibited from distributing or transferring the data and codebooks on which their research was based to any other individual or entity under the terms of an approved NHLBI Framingham Heart Study Research Proposal and Data and Materials Distribution Agreement through which the authors obtained these data.

Author Contributions: All authors had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.Study concept and design: All authors. Acquisition, analysis, or interpretation of data: All authors. Drafting of the manuscript: SF Zhou. Critical revision of the manuscript for important intellectual content: KR Wang. Statistical analysis: KR Wang. Obtained funding: KR Wang. Supervision: KR Wang. Grant Support: None. All authors have read the journal’s authorship agreement, and the manuscript has been reviewed by and approved by all named authors.

Ethical Standards: The study procedures followed were in accordance with the ethical standards of the Institutional Review Board and the Principles of the Declaration of Helsinki.

Conflicts of Interest: The authors declare that they have no conflicts of interest.



1. Matthews FE, Stephan BC, Robinson L, Jagger C, Barnes LE, Arthur A, Brayne C, Cognitive F, Ageing Studies C (2016) A two decade dementia incidence comparison from the Cognitive Function and Ageing Studies I and II. Nat Commun 7, 11398.
2. Gao S, Burney HN, Callahan CM, Purnell CE, Hendrie HC (2019) Incidence of Dementia and Alzheimer Disease Over Time: A Meta-Analysis. J Am Geriatr Soc 67, 1361-1369.
3. Wesselman LMP, Schild AK, Hooghiemstra AM, Meiberth D, Drijver AJ, Leeuwenstijn-Koopman MV, Prins ND, Brennan S, Scheltens P, Jessen F, van der Flier WM, Sikkes SAM (2020) Targeting Lifestyle Behavior to Improve Brain Health: User-Experiences of an Online Program for Individuals with Subjective Cognitive Decline. J Prev Alzheimers Dis 7, 184-194.
4. Crous-Bou M, Minguillon C, Gramunt N, Molinuevo JL (2017) Alzheimer’s disease prevention: from risk factors to early intervention. Alzheimers Res Ther 9, 71.
5. Saif N, Yan P, Niotis K, Scheyer O, Rahman A, Berkowitz M, Krikorian R, Hristov H, Sadek G, Bellara S, Isaacson RS (2020) Feasibility of Using a Wearable Biosensor Device in Patients at Risk for Alzheimer’s Disease Dementia. J Prev Alzheimers Dis 7, 104-111.
6; Kalra A, Teixeira AL, Diniz BS (2020) Association of Vitamin D Levels with Incident All-Cause Dementia in Longitudinal Observational Studies: A Systematic Review and Meta-analysis. J Prev Alzheimers Dis 7, 14-20.
7. Wang TW, Asman K, Gentzke AS, Cullen KA, Holder-Hayes E, Reyes-Guzman C, Jamal A, Neff L, King BA (2018) Tobacco Product Use Among Adults – United States, 2017. MMWR Morb Mortal Wkly Rep 67, 1225-1232.
8. Kivipelto M, Mangialasche F, Ngandu T (2018) Lifestyle interventions to prevent cognitive impairment, dementia and Alzheimer disease. Nat Rev Neurol 14, 653-666.
9. Skipina TM, Soliman EZ, Upadhya B (2020) Association between secondhand smoke exposure and hypertension: nearly as large as smoking. J Hypertens 38, 1899-1908.
10. Tommasi S, Yoon JI, Besaratinia A (2020) Secondhand Smoke Induces Liver Steatosis through Deregulation of Genes Involved in Hepatic Lipid Metabolism. Int J Mol Sci 21.
11. Chen HJ, Li GL, Sun A, Peng DS, Zhang WX, Yan YE (2019) Age Differences in the Relationship between Secondhand Smoke Exposure and Risk of Metabolic Syndrome: A Meta-Analysis. Int J Environ Res Public Health 16.
12. Llewellyn DJ, Lang IA, Langa KM, Naughton F, Matthews FE (2009) Exposure to secondhand smoke and cognitive impairment in non-smokers: national cross sectional study with cotinine measurement. BMJ 338, b462.
13. Gardener H, Wright CB, Rundek T, Sacco RL (2015) Brain health and shared risk factors for dementia and stroke. Nat Rev Neurol 11, 651-657.
14. Kannel WB, Feinleib M, McNamara PM, Garrison RJ, Castelli WP (1979) An investigation of coronary heart disease in families. The Framingham offspring study. Am J Epidemiol 110, 281-290.
15. Dawber TR, Kannel WB, Lyell LP (1963) An approach to longitudinal studies in a community: the Framingham Study. Ann N Y Acad Sci 107, 539-556.
16. Lane CA, Barnes J, Nicholas JM, Sudre CH, Cash DM, Malone IB, Parker TD, Keshavan A, Buchanan SM, Keuss SE, James SN, Lu K, Murray-Smith H, Wong A, Gordon E, Coath W, Modat M, Thomas D, Richards M, Fox NC, Schott JM (2020) Associations Between Vascular Risk Across Adulthood and Brain Pathology in Late Life: Evidence From a British Birth Cohort. JAMA Neurol 77, 175-183.
17. American Psychatric Association.Arlington V (2000) Diagnostic and Statistical Manual of Mental Disorders.4th ed, American Psychiatric Publishing.
18. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer’s disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease. Neurology 34, 939-944.
19. Weinstein G, Beiser AS, Preis SR, Courchesne P, Chouraki V, Levy D, Seshadri S (2016) Plasma clusterin levels and risk of dementia, Alzheimer’s disease, and stroke. Alzheimers Dement (Amst) 3, 103-109.
20. Shastri SS, Talluri R, Shete S (2020) Disparities in Secondhand Smoke Exposure in the United States: National Health and Nutrition Examination Survey 2011-2018. JAMA Intern Med.
21. Chinese Center for Disease Control and Prevention released 2015 China adult tobacco survey report,,
22. (2014) In The Health Consequences of Smoking-50 Years of Progress: A Report of the Surgeon General, Atlanta (GA).
23. Force USPST, Owens DK, Davidson KW, Krist AH, Barry MJ, Cabana M, Caughey AB, Curry SJ, Donahue K, Doubeni CA, Epling JW, Jr., Kubik M, Ogedegbe G, Pbert L, Silverstein M, Simon MA, Tseng CW, Wong JB (2020) Primary Care Interventions for Prevention and Cessation of Tobacco Use in Children and Adolescents: US Preventive Services Task Force Recommendation Statement. JAMA 323, 1590-1598.
24. Zhao J, Li X, Stewart SL, Gao W, Qi F, Zhang L, Pang Z, Qiao Q, Ning F, Tong E (2019) Cigarette Smoking and Secondhand Smoke Exposure Before and After a Tobacco-Free Olympic Policy Period: Qingdao, China. Nicotine Tob Res 21, 1531-1538.
25. Jhee JH, Joo YS, Kee YK, Jung SY, Park S, Yoon CY, Han SH, Yoo TH, Kang SW, Park JT (2019) Secondhand Smoke and CKD. Clin J Am Soc Nephrol 14, 515-522.
26. Kim AS, Ko HJ, Kwon JH, Lee JM (2018) Exposure to Secondhand Smoke and Risk of Cancer in Never Smokers: A Meta-Analysis of Epidemiologic Studies. Int J Environ Res Public Health 15.
27. Juntarawijit C, Juntarawijit Y (2020) Cooking with biomass fuel and cardiovascular disease: a cross-sectional study among rural villagers in Phitsanulok, Thailand. F1000Res 9, 307.
28. He F, Li T, Lin J, Li F, Zhai Y, Zhang T, Gu X, Zhao G (2020) Passive Smoking Exposure in Living Environments Reduces Cognitive Function: A Prospective Cohort Study in Older Adults. Int J Environ Res Public Health 17.
29. St Helen G, Jacob P, 3rd, Peng M, Dempsey DA, Hammond SK, Benowitz NL (2014) Intake of toxic and carcinogenic volatile organic compounds from secondhand smoke in motor vehicles. Cancer Epidemiol Biomarkers Prev 23, 2774-2782.
30. Chaichalotornkul S, Nararatwanchai T, Narkpinit S, Dararat P, Kikuchi K, Maruyama I, Tancharoen S (2015) Secondhand smoke exposure-induced nucleocytoplasmic shuttling of HMGB1 in a rat premature skin aging model. Biochem Biophys Res Commun 456, 92-97.
31. Jiang L, Chang J, Ziogas A, Deapen D, Reynolds P, Bernstein L, Anton-Culver H (2019) Secondhand smoke, obesity, and risk of type II diabetes among California teachers. Ann Epidemiol 32, 35-42.
32. Biessels GJ, Despa F (2018) Cognitive decline and dementia in diabetes mellitus: mechanisms and clinical implications. Nat Rev Endocrinol 14, 591-604.
33. Deckers K, Schievink SHJ, Rodriquez MMF, van Oostenbrugge RJ, van Boxtel MPJ, Verhey FRJ, Kohler S (2017) Coronary heart disease and risk for cognitive impairment or dementia: Systematic review and meta-analysis. PLoS One 12, e0184244.
34. Morris PB, Ference BA, Jahangir E, Feldman DN, Ryan JJ, Bahrami H, El-Chami MF, Bhakta S, Winchester DE, Al-Mallah MH, Sanchez Shields M, Deedwania P, Mehta LS, Phan BA, Benowitz NL (2015) Cardiovascular Effects of Exposure to Cigarette Smoke and Electronic Cigarettes: Clinical Perspectives From the Prevention of Cardiovascular Disease Section Leadership Council and Early Career Councils of the American College of Cardiology. J Am Coll Cardiol 66, 1378-1391.
35. Liu J, Wang X, Narayan S, Glantz SA, Schick SF, Springer ML (2016) Impairment of Endothelial Function by Little Cigar Secondhand Smoke. Tob Regul Sci 2, 56-63.
36. Pietrzak RH, Lim YY, Neumeister A, Ames D, Ellis KA, Harrington K, Lautenschlager NT, Restrepo C, Martins RN, Masters CL, Villemagne VL, Rowe CC, Maruff P, Australian Imaging B, Lifestyle Research G (2015) Amyloid-beta, anxiety, and cognitive decline in preclinical Alzheimer disease: a multicenter, prospective cohort study. JAMA Psychiatry 72, 284-291.
37. Arnold M, Nho K, Kueider-Paisley A, Massaro T, Huynh K, Brauner B, MahmoudianDehkordi S, Louie G, Moseley MA, Thompson JW, John-Williams LS, Tenenbaum JD, Blach C, Chang R, Brinton RD, Baillie R, Han X, Trojanowski JQ, Shaw LM, Martins R, Weiner MW, Trushina E, Toledo JB, Meikle PJ, Bennett DA, Krumsiek J, Doraiswamy PM, Saykin AJ, Kaddurah-Daouk R, Kastenmuller G (2020) Sex and APOE epsilon4 genotype modify the Alzheimer’s disease serum metabolome. Nat Commun 11, 1148.



M. Zuin1,3, C. Cervellati1, G. Brombo1, A. Trentini2, L. Roncon3, G. Zuliani1


1. Department of Morphology, Surgery & Experimental Medicine, University of Ferrara, Ferrara, Italy; 2. Department of Biomedical and Specialist Surgical Sciences, University of Ferrara, Ferrara, Italy; 3. Department of Cardiology, Santa Maria delle Misericordia Hospital, Rovigo, Italy

Corresponding Author: Carlo Cervellati, PhD, Department of Morphology and Experimental Medicine, University of Ferrara, Via Luigi Borsari 46, I-44121, Ferrara, Italy;Tel. ++39-532-455441; Fax. ++39-532-455426; e-mail: crvcrl@unife

J Prev Alz Dis 2021;
Published online September 14, 2020,



Objective: To investigate whether high serum homocysteine (Hcy) levels is associated with the risk of developing Alzheimer’s disease (AD) by performing a meta-analysis based on updated published data.
Methods: We conducted a comprehensive research using Medline (Pubmed), Scopus, Web of Science and EMBASE databases to identify all prospective studies published any time to July 7, 2020 evaluating the association between elevated Hcy levels and AD risk.
Results: From an initial screening of 269 published papers, 9 prospective investigations conducted on a total of 7474 subjects with mean follow-up of 9.5 years (range: 3.7-10) were included in the meta-analysis. Eight seventy-five of these subjects converted to AD. Hcy was significantly higher in these individuals (HRadjusted:1.48, 95% CI:1.23-1.76, I2=65.6%, p<0.0001) compared with who did not convert to AD. There was a significant publication bias (Egger’s test, t=6.39, p=0.0003) and this was overcome by the trim and fill method, which allowed to calculate a bias-corrected imputed risk estimate of HRadjusted:1.20, 95% CI:1.01-1.44, Q value=41.92.
Conclusions: The present meta-analysis found that having higher Hcy increases the risk of AD in the elderly and this finding is consistent with the widely suggested role of this non-proteinogenic α-amino acid in AD neurodegeneration.

Key words: Alzheimer’s disease, homocysteine, meta-analysis, prospective studies.



Mounting epidemiological and clinical evidences have demonstrated a considerable overlap between Vascular dementia (VaD) and Alzheimer’s disease (AD) (1, 2). The emerging scenario highlights that cardiovascular disease (CVD), atherosclerosis, and cerebral microvasculature abnormalities mutually interact promoting neurodegeneration since the earliest stage of AD, and influencing the disease progression (3, 4). In support of this view, several studies have demonstrated the presence of an association between cardiometabolic risk factors and development AD, besides VAD (5–9).
In this regard, hyperhomocysteinemia (H-Hcy), which represents a well-established cardiovascular risk factor (10), represents an emblematic example in this frame. The first solid demonstration showing that increased H-Hcy is an independent risk factor for the development of AD, and more in general dementia, was presented in 2002 (11). Since then, several epidemiological studies have been consistent with this finding, (12, 13), suggesting Hcy as a potential target for both non- and pharmacological treatments (14). Unfortunately, the causality of H-Hcy in AD has not yet been definitely confirmed, although experimental evidence clearly suggests its implication in pathogenic mechanism of the neurodegenerative disease (15, 16). One of the most intriguing hypotheses linking H-Hcy and AD onset, is inspired by the role of Hcy in the metabolism of methionine, and the importance of the latter in phosphatidylcholine synthesis. Indeed, Hcy is a product of methionine catabolism, but it can also be recycled back to the essential amino acid by the vitamin-B12 dependent methionine synthase, as well as via a folate-independent pathway (17, 18). Owing the crucial role of these two vitamins in methionine synthesis, a deficiency of either of them can result in H-Hcy and low bioavailability of methionine (19). In turn, a decrease in methionine may cause a lower synthesis of phosphatidylcholine (methionine is a precursor of this phospholipid), which serves as important carrier for docosahexaenoic acid (DHA) through the blood-brain barrier. Importantly, DHA is the most abundant fatty acid in the brain and its deficiency is associated with AD (20).
The clinical relevance of the topic, prompted us to provide an updated systematic review and meta-analysis based on published prospective studies evaluating the role of H-Hcy and the risk of AD.



Search strategy

This study followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline (Supplementary file 1) (21). Data were obtained searching MEDLINE, Scopus and Web of Science and EMBASE for all prospective studies in English language and without age restrictions, published any time to July 7, 2020 evaluating the association between H-Hcy and the risk of AD in the elderly. The risk of AD due to H-Hcy was chosen as the primary outcome of the study.

Study selection

The selection of studies to be included in our analysis was independently conducted by 2 authors (MZ, GZ) in a blinded fashion. Any discrepancies in study selection was resolved consulting a third author (CC). The following MeSH terms were used for the search: “Homocysteine” OR “Hyperhomocisteinemia” AND “Alzheimer’s disease” OR “Dementia”. Moreover, we searched the bibliographies of target studies for additional references. Case reports, review articles, abstracts, editorials/letters, and case series with less than 10 participants were excluded. Data extraction was independently conducted by 2 authors (AT, MZ). Any disagreements were resolved by consensus after discussion. Studies were included in the present analysis if they were prospective investigations or prospective nested case-control studies assessing the relationship between H-Hcy and AD and the results expressed as hazard ratio (HR) and relative 95% confidence interval (CI).

Data extraction

For each investigation included into the final analysis, the following items were extracted: year of publication, country, sample size, male gender, mean follow-up duration, diagnostic criteria for AD, method used for the assessment of blood Hcy concentration and covariates used in the multivariate analyses of each manuscript. The quality of included studies was graded using the Newcastle-Ottawa quality assessment scale (22).

Statistical analysis

Continues variables were expressed as mean ± standard deviation (SD) or range while categorical variables were presented as numbers and relative percentages. From each study, the adjusted hazard ratio (aHR) and 95% confidence interval (CI) for the higher versus the lower Hcy category comparison was pooled using a random-effects model, while a traditional forest plot was adopted to visually evaluate the results. Statistical heterogeneity between groups was measured using the Higgins I2 statistic. Specifically, a I2=0 indicated no heterogeneity while we considered low, moderate, and high degrees of heterogeneity the values of I2 as <25%, 25–75% and above 75%, respectively. Moreover, tau-squared (τ2) was also calculated to see the extent of variation among the effects observed in different studies. To evaluate potential bias, both the Egger’s test and funnel plots were computed. In case of significant Egger’s test, the Begg’s rank correlation test was also carried out and the trim-fill method was used to re-calculate the pooled risk estimates. A p-value < 0.05 was considered statistically significant. All meta-analyses were conducted using Comprehensive Meta-Analysis software, version 3 (Biostat, USA).



A total of 269 articles were retrieved after excluding duplicates. The initial screening excluded 186 articles because they did not meet inclusion criteria, leaving 83 articles to assess for eligibility. After evaluation of the full-text articles, 74 were excluded and 9 prospective investigations met the inclusion criteria (Figure 1) (11, 23–29).

Figure 1. PRISMA flow chart


Overall, 7474 community-dwelling adults (mean age 71 years, 53% male), with a mean follow-up of 9.5 years were analysed (Table 1). Eight seventy-five of these subjects (11.7%) converted to AD.

Table 1. General characteristics of the studies included in the meat-analysis


The diagnostic criteria used in the studies were: the National Institute of Neurological and Communicative Disorders and Stroke (NINCDS) (11,25), the National Institute of Neurological and Communicative Disorders and Stroke of the United States and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) (23, 24, 26, 27), the revised Diagnostic and Statistical Manual of Mental Disorders criteria (DSM-III-R) (29,30), DSM IV (28) and the National Institute of Neurological Disorders and Stroke (NINDS) Association Internationale pour la Recherche et l’Enseignement en Neurosciences (AIREN) criteria (NINDS-AIREN) (23).
The different confounders considered for the estimation of Hazard Risk in each analysis are shown in Table 2. Age (n=9 studies), sex (n=8), Apo E4 (n=7), education (n=9) and body mass index (n=7) were the most considered covariates; surprisingly, the most important determinants of Hcy, folate and vitamin B12, in less than 50% of the investigations. The studies included into the meta-analysis resulted of moderate-high quality according to the NOS.


Table 2. Confounders considered in each study for the estimation of Hazard risk

CVD: Cardiovascular disease; Hb, Haemoglobin; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; BMI: Body mass index; MMSE: Mini-mental state; holo-TC: Holotranscobalamin; HT: Arterial hypertension; eGFR: Estimated glomerular filtration rate


The pooled analysis, based on a random effect model, revealed that subjects with higher vs. lower levels of blood Hcy had an increased risk of AD (Figure 2, HRadjusted:1.48, 95% CI:1.23-1.76, p<0.0001), with a moderate heterogeneity in effects size between the studies (I2=65.6%). However, as displayed by the Funnel Plot (Figure 3), there was a significant publication bias confirmed also by the Egger’s test (t=6.39, p=0.0003). To overcome this limitation, the trim and fill method calculated a bias-corrected imputed risk estimate of HRadjusted:1.20, 95% CI:1.01-1.44, Q value=41.92.

Figure 2. Forest plots investigating the risk of AD in patients with hyperhomocysteinemia


Figure 3. : Funnel plot for the risk of AD in patients with hyperhomocysteinemia



The result of the present meta-analysis confirms that higher concentration of blood Hcy increases the risk of developing AD in older individuals. The clinical value of Hcy is beyond its mere use as static biomarker; indeed, this cysteine homologue represents a well-known modifiable risk factor, especially in the field of cardiovascular prevention, as well as a potential therapeutic target.
H-Hcy has been found to be related with cognitive decline, global and regional brain atrophy (including hippocampus volumes), white matter damage, formation and/or accumulation of the major AD-neuropathological hallmarks, neurofibrillary tangles and neuritic plaques (31). Notably, some authors found that a nutritional model of B vitamin deficiency with Hcy cycle alteration could lead to increased amyloid β (Aβ) deposition, due to over-expression in presenilin 1 and β-secretase 1 activity (32). Similarly, Li and co-workers reported a dietary approach that leads to an increase in Hcy levels resulting in a typical AD phenotype where Aβ and tau neuropathology were accompanied by memory deficit (33). More recently, it has been shown that supraphysiological concentrations of Hcy (>0.5 µM) caused a decrease in synaptic proteins in AD animal model, with the concomitant increase in oxidative stress and excitatory transmission hyperactivity, which are all considered to be neurotoxic effects (34). Furthermore, it has been reported that H-Hcy plays a causal role in stroke (35), a frequent co-existing pathology and potent risk factor of AD (36, 37), and has deleterious effects on the cerebral vasculature, including blood brain barrier disruption (38), a well-recognized early event in AD pathogenesis (39).
A meta-analysis on studies published until June 2018 showed increase of 1 μmol/L in Hcy in the blood is linearly associated with a 15% increase in the relative risk of AD (40). Our work adds to those performed to Zhou et al, since we have considered around one thousand and six hundred more patients. Moreover, the cited authors performed a dose-response meta-analysis on the risk all-cause dementia (AD and vascular dementia), while our study aimed to confirm whether patients of general population with H-Hcy, were at higher risk of AD. Indeed, evaluating the risk of AD in terms of fixed increase of blood Hcy, as every 5 μmol/L results directly correlated with the baseline values. Conversely, it could be more useful for clinicians to establish a direct relationship between H-Hcy and AD in the evaluation of patients with dementia. Furthermore, whether the risk between H-Hcy and AD follow a linear or exponential growth, has not yet been defined.
H-Hcy remains a major and yet underrecognized risk factor for cognitive impairment and dementia in daily clinical practice (41). This is mostly due to the contrasting results of the clinical trials that failed to show a clear beneficial effect of Hcy-lowering B vitamins (B-6, B-12 and folic acid) supplementation on cognitive decline. However, some studies found that baseline Hcy levels could be predictive of the response with beneficial effects of B vitamins administration only in subjects with high baseline Hcy (42, 43). The effectiveness of B vitamins supplementation could depend on other endogen and exogen factors; therefore, it could be helpful to identify subgroups that are likely to benefit of such supplementation in clinical trials. Of particular relevance to this context, two studies reported a beneficial effect of B vitamins supplementation on brain atrophy and cognitive decline only on those subjects which had high baseline levels of plasma omega (ω)-3 fatty acids (FA) (44, 45). Interestingly, the recent findings of Jerenlen et al. clearly suggest that B vitamins, Hcy and ω-3 FA influence each other. In fact, FAs supplementation seems to be effective on cognitive and clinical outcomes performance only on those AD patients with low baseline levels of Hcy (46).
Our findings confirm that the assessment of Hcy level in serum is a promising tool for the evaluation of AD risk in general population. However, it is undeniable that any case of h-Hcy should be adequately interpreted in a multidimensional evaluation because it could be expression of an underlying causal condition (e.g. chronic renal failure, alcohol consumption, smoke, use of some medications) or a consequence of cognitive decline itself (e.g. malnutrition in demented patients). Our analysis has some limitations. Firstly, being based on observational studies, the possibility of remaining residual confounding items, due either to unmeasured or underestimated risk factors in the reviewed studies cannot be excluded, representing a potential source of biases. At the same manner, we cannot exclude that patients enrolled in the reviewed cohort might be treated with vitamin B supplementation. However, potential bias resulted mitigated by the fact that some of the reviewed studies demonstrated that H-Hcy remained associated with AD, after adjustment for vitamin B levels (24, 30). However, the relative long follow-up period of the studies considered, our findings are less prone to be biased due to potential reversed causalities over the time. Finally, the lack of standardized cut-offs for H-Hcy represents another important limitation in our findings and analysis.



Our meta-analysis found that H-Hcy increases the risk of AD in the elderly and this finding is consistent with the potential role of Hcy in promoting neurodegeneration. Although interventional studies analysing B vitamins supplementation in terms of prevention of Hcy-related cognitive decline have shown scant results, promotion of healthy lifestyle, screening of high-risk subjects and earlier therapeutic approaches, before neurological damages have occurred, could get better results.

Key points

1. Homocysteine might play an important role in Alzheimer’s disease-related neurodegeneration
2. In elderly, higher blood levels of homocysteine are associated with a greater risk of developing Alzheimer’s disease


Conflicts of interest: All authors declare that they have no conflicts of interest.

Data availability statement: Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Funfing information: The research reported did not receive any specific grant from funding agencies in the public, commercial, or not- for- profit sectors.

Approval by ethical committee: Not necessary (systematic review)



1. Ravaglia G, Forti P, Montesi F, Lucicesare A, Pisacane N, Rietti E, et al. Mild cognitive impairment: epidemiology and dementia risk in an elderly Italian population. J Am Geriatr Soc. 2008;56(1):51–8.
2. Villeneuve S, Jagust WJ. Imaging Vascular Disease and Amyloid in the Aging Brain: Implications for Treatment. J Prev Alzheimer’s Dis. 2015;2(1):64–70.
3. de la Torre JC. Is Alzheimer’s disease a neurodegenerative or a vascular disorder? Data, dogma, and dialectics. Lancet Neurol. 2004;3(3):184–90.
4. Iturria-Medina Y, Sotero RC, Toussaint PJ, Mateos-Pérez JM, Evans AC. Early role of vascular dysregulation on late-onset Alzheimer’s disease based on multifactorial data-driven analysis. Nat Commun. 2016;7(1):11934.
5. Cervellati C, Trentini A, Romani A, Bellini T, Bosi C, Ortolani B, et al. Serum paraoxonase and arylesterase activities of paraoxonase-1 (PON-1), mild cognitive impairment, and 2-year conversion to dementia: A pilot study. J Neurochem. 2015;135(2):395–401.
6. Hottman DA, Chernick D, Cheng S, Wang Z, Li L. HDL and cognition in neurodegenerative disorders. Neurobiol Dis. 2014;72 Pt A:22–36.
7. Craft S. The role of metabolic disorders in Alzheimer disease and vascular dementia: two roads converged. Arch Neurol. 2009;66(3):300–5.
8. Wood WG, Li L, Müller WE, Eckert GP. Cholesterol as a causative factor in Alzheimer’s disease: a debatable hypothesis. J Neurochem. 2014;129(4):559–72.
9. van Eersel MEA, Joosten H, Gansevoort RT, Slaets JPJ, Izaks GJ. Treatable Vascular Risk and Cognitive Performance in Persons Aged 35 Years or Older: Longitudinal Study of Six Years. J Prev Alzheimer’s Dis. 2019;6(1):42–9.
10. Chrysant SG, Chrysant GS. The current status of homocysteine as a risk factor for cardiovascular disease: a mini review. Expert Rev Cardiovasc Ther. 2018;16(8):559–65.
11. Seshadri S, Beiser A, Selhub J, Jacques PF, Rosenberg IH, D’Agostino RB, et al. Plasma Homocysteine as a Risk Factor for Dementia and Alzheimer’s Disease. N Engl J Med. 2002;346(7):476–83.
12. Ma F, Wu T, Zhao J, Ji L, Song A, Zhang M, et al. Plasma Homocysteine and Serum Folate and Vitamin B12 Levels in Mild Cognitive Impairment and Alzheimer’s Disease: A Case-Control Study. Nutrients. 2017;9(7):725.
13. Handing EP, Small BJ, Reynolds SL, Kumar NB. Impact of Dietary Factors and Inflammation on Cognition among Older Adults. J Prev Alzheimer’s Dis. 2015;2(4):220–6.
14. Elias MF, Sullivan LM, D’Agostino RB, Elias PK, Jacques PF, Selhub J, et al. Homocysteine and Cognitive Performance in the Framingham Offspring Study: Age Is Important. Am J Epidemiol. 2005;162(7):644–53.
15. McCully KS. Chemical pathology of homocysteine. IV. Excitotoxicity, oxidative stress, endothelial dysfunction, and inflammation. Ann Clin Lab Sci. 2009;39(3):219–32.
16. Bonetti F, Brombo G, Zuliani G. The relationship between hyperhomocysteinemia and neurodegeneration. Neurodegener Dis Manag. 2016;6(2):133–45.
17. Selhub J, Miller JW. The pathogenesis of homocysteinemia: interruption of the coordinate regulation by S-adenosylmethionine of the remethylation and transsulfuration of homocysteine. Am J Clin Nutr. 1992;55(1):131–8.
18. Froese DS, Fowler B, Baumgartner MR. Vitamin B 12 , folate, and the methionine remethylation cycle-biochemistry, pathways, and regulation. J Inherit Metab Dis. 2019;42(4):673–85.
19. Yamada K, Kawata T, Wada M, Isshiki T, Onoda J, Kawanishi T, et al. Extremely Low Activity of Methionine Synthase in Vitamin B-12–Deficient Rats May Be Related to Effects on Coenzyme Stabilization Rather than to Changes in Coenzyme Induction. J Nutr. 2000;130(8):1894–900.
20. Sugasini D, Yalagala PCR, Goggin A, Tai LM, Subbaiah P V. Enrichment of brain docosahexaenoic acid (DHA) is highly dependent upon the molecular carrier of dietary DHA: lysophosphatidylcholine is more efficient than either phosphatidylcholine or triacylglycerol. J Nutr Biochem. 2019;74:108231.
21. Moher D, Liberati A, Tetzlaff J, Altman DG, PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 2009;6(7):e1000097.
22. Hartling L, Milne A, Hamm MP, Vandermeer B, Ansari M, Tsertsvadze A, et al. Testing the Newcastle Ottawa Scale showed low reliability between individual reviewers. J Clin Epidemiol. 2013;66(9):982–93.
23. Chen S, Honda T, Ohara T, Hata J, Hirakawa Y, Yoshida D, et al. Serum homocysteine and risk of dementia in Japan. J Neurol Neurosurg Psychiatry. 2020;91(5):540–6.
24. Ravaglia G, Forti P, Maioli F, Chiappelli M, Montesi F, Tumini E, et al. Blood inflammatory markers and risk of dementia: The Conselice Study of Brain Aging. Neurobiol Aging. 2007;28(12):1810–20.
25. Luchsinger JA, Tang M-X, Shea S, Miller J, Green R, Mayeux R. Plasma homocysteine levels and risk of Alzheimer disease. Neurology. 2004 Jun 8;62(11):1972–6.
26. Ravaglia G, Forti P, Maioli F, Martelli M, Servadei L, Brunetti N, et al. Homocysteine and folate as risk factors for dementia and Alzheimer disease. Am J Clin Nutr. 2005;82(3):636–43.
27. Hooshmand B, Solomon A, Kareholt I, Leiviska J, Rusanen M, Ahtiluoto S, et al. Homocysteine and holotranscobalamin and the risk of Alzheimer disease: A longitudinal study. Neurology. 2010;75(16):1408–14.
28. Miwa K, Tanaka M, Okazaki S, Yagita Y, Sakaguchi M, Mochizuki H, et al. Increased Total Homocysteine Levels Predict the Risk of Incident Dementia Independent of Cerebral Small-Vessel Diseases and Vascular Risk Factors. J Alzheimer’s Dis. 2015;49(2):503–13.
29. Zylberstein DE, Lissner L, Björkelund C, Mehlig K, Thelle DS, Gustafson D, et al. Midlife homocysteine and late-life dementia in women. A prospective population study. Neurobiol Aging. 2011;32(3):380–6.
30. Kivipelto M, Annerbo S, Hultdin J, Bäckman L, Viitanen M, Fratiglioni L, et al. Homocysteine and holo-transcobalamin and the risk of dementia and Alzheimers disease: a prospective study. Eur J Neurol. 2009;16(7):808–13.
31. Smith AD, Smith SM, de Jager CA, Whitbread P, Johnston C, Agacinski G, et al. Homocysteine-lowering by B vitamins slows the rate of accelerated brain atrophy in mild cognitive impairment: a randomized controlled trial. PLoS One. 2010;5(9):e12244.
32. Fuso A, Nicolia V, Cavallaro RA, Ricceri L, D’Anselmi F, Coluccia P, et al. B-vitamin deprivation induces hyperhomocysteinemia and brain S-adenosylhomocysteine, depletes brain S-adenosylmethionine, and enhances PS1 and BACE expression and amyloid-β deposition in mice. Mol Cell Neurosci. 2008; 37(4):731-46
33. Li J-G, Chu J, Barrero C, Merali S, Praticò D. Homocysteine exacerbates β-amyloid pathology, tau pathology, and cognitive deficit in a mouse model of Alzheimer disease with plaques and tangles. Ann Neurol. 2014;75(6):851–63.
34. Montecinos-Oliva C, Arrázola MS, Jara C, Tapia-Rojas C, Inestrosa NC. Hormetic-Like Effects of L-Homocysteine on Synaptic Structure, Function, and Aβ Aggregation. Pharmaceuticals. 2020;13(2):24.
35. Gu SX, Sonkar VK, Katare PB, Kumar R, Kruger WD, Arning E, et al. Memantine Protects From Exacerbation of Ischemic Stroke and Blood Brain Barrier Disruption in Mild But Not Severe Hyperhomocysteinemia. J Am Heart Assoc. 2020;9(4):e013368.
36. de la Torre JC. Alzheimer Disease as a Vascular Disorder. Stroke. 2002;33(4):1152–62.
37. Cervellati C, Wood PL, Romani A, Valacchi G, Squerzanti M, Sanz JM, et al. Oxidative challenge in Alzheimer’s disease: state of knowledge and future needs. J Investig Med. 2016;64(1):21–32.
38. Beard RS, Reynolds JJ, Bearden SE. Hyperhomocysteinemia increases permeability of the blood-brain barrier by NMDA receptor-dependent regulation of adherens and tight junctions. Blood. 2011;118(7):2007–14.
39. Cervellati C, Trentini A, Pecorelli A, Valacchi G. Inflammation in Neurological Disorders: The Thin Boundary Between Brain and Periphery. Antioxid Redox Signal. 2020;33(3):191–210.
40. Zhou F, Chen S. Hyperhomocysteinemia and risk of incident cognitive outcomes: An updated dose-response meta-analysis of prospective cohort studies. Ageing Res Rev. 2019;51:55–66.
41. Price BR, Wilcock DM, Weekman EM. Hyperhomocysteinemia as a Risk Factor for Vascular Contributions to Cognitive Impairment and Dementia. Front Aging Neurosci. 2018;10:350.
42. Douaud G, Refsum H, de Jager CA, Jacoby R, Nichols TE, Smith SM, et al. Preventing Alzheimer’s disease-related gray matter atrophy by B-vitamin treatment. Proc Natl Acad Sci U S A. 2013;110(23):9523–8.
43. Jager CA, Oulhaj A, Jacoby R, Refsum H, Smith AD. Cognitive and clinical outcomes of homocysteine-lowering B-vitamin treatment in mild cognitive impairment: a randomized controlled trial. Int J Geriatr Psychiatry. 2012;27(6):592–600.
44. Jernerén F, Elshorbagy AK, Oulhaj A, Smith SM, Refsum H, Smith AD. Brain atrophy in cognitively impaired elderly: the importance of long-chain ω-3 fatty acids and B vitamin status in a randomized controlled trial. Am J Clin Nutr. 2015;102(1):215–21.
45. Oulhaj A, Jernerén F, Refsum H, Smith AD, de Jager CA. Omega-3 Fatty Acid Status Enhances the Prevention of Cognitive Decline by B Vitamins in Mild Cognitive Impairment. J Alzheimer’s Dis. 2016;50(2):547–57.
46. Jernerén F, Cederholm T, Refsum H, Smith AD, Turner C, Palmblad J, et al. Homocysteine Status Modifies the Treatment Effect of Omega-3 Fatty Acids on Cognition in a Randomized Clinical Trial in Mild to Moderate Alzheimer’s Disease: The OmegAD Study. J Alzheimers Dis. 2019;69(1):189–97.


K.V. Giudici1

1. Gerontopole of Toulouse, Institute of Aging, Toulouse University Hospital (CHU Toulouse), Toulouse, France.

Corresponding Author: Kelly Virecoulon Giudici, Gérontopôle of Toulouse, Institute of Aging, 37 Allée Jules Guesde, 31000 Toulouse, France, E-mail:

J Prev Alz Dis 2021;
Published online February 7, 2021,


The Clinical Trials on Alzheimer’s Disease (CTAD) 2020 conference was the stage for researchers from all over the world to present their recent and ongoing research focused on potential Alzheimer’s disease (AD) treatments and prevention of cognitive decline. Among a varied range of topics, nutritional aspects arose as possibilities of treatments towards the promotion of a healthy aging. Among the discussed themes, supplementation of omega-3 polyunsaturated fatty acids and multi-nutrient approaches were presented, suggesting that long-term supplementation (i.e., over 3 years) might be needed for observing positive effects on cognitive performance. Trials testing ketogenic agents and carbohydrate-restricted diet were also presented and showed promising effects on improving cognitive function of mild-cognitive impaired (MCI) and pre-diabetic individuals, respectively, in a short-term way (i.e. after 3 to 6 months). The combination of some of the nutritional approaches with physical activity interventions raises the question on whether they would individually perform in a similar way. Promising therapies involving nutrition appear to be safe and well tolerated by volunteers. Failures on achieving positive findings raise questions on whether they were driven by specific characteristics of the studied populations, insufficient doses or duration of treatment. Notwithstanding, current evidence on the applicability of nutrition-based approaches as AD treatments are encouraging but demand further research on the topic.



As the population’s life expectancy becomes longer and the prevalence of Alzheimer’s disease (AD) increases worldwide, measures protecting cognitive function and promoting healthy aging become of extreme relevance to public health. The Clinical Trials on Alzheimer’s Disease (CTAD) conference annually welcomes the most recent findings from clinical trials of AD treatments, bringing light to profitable discussions on their successes and failures, in order to allow future research to be better designed and more effective. Its most recent edition, which occurred virtually due to the COVID-19 pandemic, welcomed over 1,500 participants from more than 40 different countries from November 4th to 7th, 2020, and provided opportunities for researchers to share what they have learned from their latest and ongoing studies.
Among a wide range of investigations on treatments aiming to slow cognitive decline and fully understand the pathophysiological mechanisms leading to AD, trials targeting nutritional pathways stand out as possibilities of nonpharmacological interventions. The present article discuss the main findings of the studies that have used nutritional approaches, exploring the characteristics that could have possibly contributed to their efficacy or failure.


Single-nutrient approaches

The effects of omega-3 polyunsaturated fatty acids (PUFA) on fighting inflammation and oxidative stress are well stablished in literature (1, 2). Docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA), mainly, have been shown to have neuroprotective abilities through both vascular and neuronal mechanisms (3, 4), being associated with reduced white matter hyperintensities (WMH) accumulation and delayed neurodegeneration (5, 6). In this sense, trials based on omega-3 PUFA supplementation have been developed. The double-blind, placebo-controlled PUFA Trial (NCT01953705) provided marine omega-3 PUFA to North-American adults free of dementia but with suboptimum plasma omega-3 (<110μg/mL) and magnetic resonance imaging (MRI)-derived WMH burden (≥5cm3), aged 75 years and older, for 3 years (7). Participants took daily doses of 1,650mg of omega-3 PUFA (975mg of EPA and 675mg of DHA) or placebo. Data presented in CTAD 2020 showed that WMH progression was slowed over 3 years among participants that adhered to study protocol, while no effects were observed among the modified intention to treat (mITT) cohort presenting at least one follow-up MRI. No differences were seen in medial temporal lobe, total brain or ventricular volume changes, nor in executive function Z-score though. The given dose and duration appeared to be safe, since no differences in adverse events between active group and placebo were observed (8).
Another study testing the effects of omega-3 PUFA supplementation on cognitive function among community-dwelling older adults is the multicenter, randomized, double-blind controlled Multidomain Alzheimer Preventive Trial (MAPT) (NCT01513252), performed in France and Monaco with subjects aged 70 years or more reporting subjective memory complaints. MAPT also provided omega-3 PUFA supplementation over 3 years (in a daily dose of 225mg of EPA and 800mg of DHA, combined or not to a multidomain intervention based on nutritional counseling, exercise advice and cognitive training), however was unable to observe a protective role of interventions on cognitive function among its mITT population (9). The MAPT team then wondered if specific subgroups would respond differently to treatment. Indeed, participants of MAPT Study who were amyloid-β (Aβ) positive (as measured by positron emission tomography – PET scan) have been shown to respond to the multidomain intervention (combined or not with omega-3 supplementation) [10]. Defining amyloid status by PET is, however, less accessible and more expensive. Blood-based biomarkers, instead, have shown high reliability to predict neurodegeneration, as Aβ peptides – which are highly associated with PET amyloid status and Aβ in cerebrospinal fluid (CSF) – in a cost-effective and less invasive way (11–13). On the recent study presented at CTAD 2020 conference, participants had their plasma Aβ42 and Aβ40 variants assessed by a high-precision immunoprecipitation and liquid chromatography-mass spectrometry assay in a subsample of participants of MAPT Study, at the end of the first year of follow-up. While no effects of MAPT interventions were observed for cognitive function among the intention-to-treat (ITT) Aβ negative group (n=322) nor in the group with low Aβ42/40 ratio (i.e. Aβ positive, n=161), the per-protocol group of Aβ positive participants (n=154) presented lower cognitive decline after receiving the omega-3 supplementation combined with the multidomain intervention. The omega-3 supplementation alone, however, did not present effects on cognitive performance among the per-protocol Aβ positive group. In addition, the observed benefits did not persist after 2 years post-intervention (14).
Findings from the PUFA Trial and the MAPT Study point towards the utility of classifying individuals based on their omega-3 status and brain characteristics (such as WMH and amyloid burden) to optimize the early identification of at-risk subjects to whom future interventions would be more profitable. For amyloid status specifically, the assessment of blood Aβ peptides emerge as a promising method for screening non-demented populations with memory complaints. The daily dose of omega-3 PUFA should also be importantly considered. Based on these studies, it is possible that a higher dose (as the one provided to participants of the PUFA Trial) might be needed for achieving positive responses after 3 years. Moreover, findings suggest that omega-3 treatments may lose their effects upon discontinuity.


Multi-nutrient approaches

In addition to trials focusing only on omega-3 PUFA properties, two other trials presented at CTAD 2020 included nutritional approaches based on multi-nutrient supplementation. Such choice was stimulated by considering the multifactorial mechanisms leading to cognitive decline, and to previous investigations showing promising effects of the combination of nutrients in cognitive outcomes among animals (15, 16) and humans (17–19).
The LipiDiDiet Trial (Dutch Trial Register number NTR1705) is a double-blind, multi-center randomized controlled trial performed in Finland, Germany, Netherlands and Sweden with 311 older adults presenting prodromal AD (defined according to the International Working Group – IWG-1 criteria). Following an initial 24-month intervention composed of daily receiving a 125mL drink containing a multi-nutrient combination (named Fortasyn Connect) or an isocaloric placebo drink, participants could opt to continue in the trial for a maximum of 72 months of intervention, and another 24 months of observational follow-up. Previous publications reporting findings after 24 months of follow-up have shown that supplementation with the multi-nutrient drink (containing EPA, DHA, phospholipids, choline, uridine monophosphate, vitamins B6, B9, B12, C, E and selenium) had no significant effect on cognitive function measured by a composite neuropsychological test battery (NTB) (20). In CTAD 2020, researchers reported their findings over a total of 36 months of intervention following the initial randomization. After 3 years, significantly lower cognitive decline and brain atrophy were observed for those taking the multi-nutrient drink. According to the NTB and the clinically relevant measure of Clinical Dementia Rating – Sum of Boxes (CDR-SB), the active group declined respectively 45% and 60% less compared to the placebo group. In addition to these clinically detectable benefits, the multi-nutrient drink presented a good safety profile and high compliance throughout this study (21, 22).
The NOLAN Study (NCT03080675) has also tested the effectiveness of a multi-nutrient oral supplement on cognitive function and brain volume of community-dwelling older adults, as also on erythrocyte omega-3 and plasma homocysteine (biomarkers with a capacity to underlie an attenuation in cognitive decline during aging (23–25)). The nutritional blend was composed by two soft capsules and a powdered drink mix to be solved in cold water, containing omega-3 PUFA (DHA and EPA), choline, vitamins B1, B2, B3, B5, B6, B7, B9, B12, E, C, D, selenium and citrulline, and offered daily for one year to participants presenting subjective memory complaints. In the meanwhile, placebo identical doses were offered to the control group, composed of volunteers expressing the same concerns. After this period, the nutritional blend successfully increased erythrocyte omega-3 and reduced plasma homocysteine concentrations, while both biomarkers worsened in the other group. Results, however, did not support its use for preventing cognitive decline over a 1-year range. Marginal significance was observed for hippocampal atrophy in the left hemisphere (with the active group presenting a lower volume decrease) (26), raising the question if a longer follow-up would have brought positive effects on brain measures.
Indeed, the limited duration of the follow-up in the NOLAN study is believed to have prevented the identification of positive findings, what is reinforced by results of the LipiDiDiet Trial showing that 36 months, but not 24, were able to deliver positive effects of the multi-nutrient intervention on cognitive performance. Taken together, the presented results suggest that benefits may get apparent or more pronounced with long-term use, highlighting the need for longer trials. The dose of each nutrient may also have contributed to the lack of effects over 1 or 2 years.


Other dietary approaches

Another nutritional approach used in a trial presented at CTAD 2020 was the oral supplementation with ketone bodies. Brain function is mainly fueled by glucose. In the insufficiency of circulating glucose (which occurs after the depletion of glycogen stores caused by a long fasting state), ketones (acetoacetate and beta-hydroxybutyrate) are naturally produced to keep brain functioning (27). In mild cognitive impairment (MCI) and AD, brain glucose uptake is reduced (28, 29). In contrast, uptake of ketones has been shown to remain normal in both MCI and mild-moderate AD (28, 30, 31). Previous research have suggested that endogenous or exogenous sources of ketones may improve energy metabolism in both MCI and AD subjects (32). To test if such improvement might also benefit cognitive performance, the randomized, placebo-controlled Benefic Trial (NCT02551419) supplemented MCI subjects with an oral nutritional supplement composed of ketogenic medium chain triglycerides (kMCT), and compared them to controls taking placebo for 6 months. The nutritional supplement was a lactose-free skim milk emulsion containing 15g of kMCT to be taken twice a day. Similarly, an energy-equivalent placebo providing 13g of non-ketogenic vegetable oil was offered to the control group twice a day. Brain ketones, glucose PET and plasma ketone response were assessed in subgroups. After 6 months, brain ketone significantly increased in the intervention group, while no changes in plasma ketone response were observed for both groups. At the end of follow-up, global brain ketone uptake doubled and cognitive performance (as measured by tests comprising episodic memory, executive function and language) improved in the active group only, in a probable clinically meaningful way. However, this group also presented increased interleukine-8, but not other inflammatory markers. Overall, the consistent plasma ketone response suggested no metabolic adaptation to the oral kMCT drink after 6 months of treatment, with safety and good acceptability from participants (33, 34).
Also using a nutrition-based approach, the Blood Flow Improvement Trial (BFIT) (NCT03117829) was designed to test whether an intervention based on exercise and carbohydrate-restricted diet (CRD) would reduce insulin resistance (IR) and improve cognitive function. The potential relationship between IR and cognitive performance stands on evidence showing that type 2 diabetes is associated with increased risk of cognitive impairment (35) and of MCI conversion to dementia (36, 37), what is believed to occur through vascular damage and dysfunctions in glucose, insulin and amyloid metabolism (38). Such evidence raises the question if reversing IR in midlife would benefit cognition. To test this hypothesis, the BFIT Study was a stepped wedge cluster randomized trial that performed a 12-week intervention composed of exercise and CRD among 29 pre-diabetic, cognitively unimpaired participants (mean age 57.9 ± 5.1 years), followed by a 6-month post-intervention phase. The exercise arm of treatment consisted of 50 minutes of supervised moderate intensity aerobic exercise three times per week, unsupervised exercise twice a week, and weekly classes of behavioral change to promote program adherence. The CRD was instructed by a dietitian and contained less than 100g of carbohydrates per day. Participants were told to avoid or restrict grains, sugars, legumes, starchy vegetables and fruits, but not carbohydrate from dairy, condiments, nuts and seeds, and were instructed to further keep the regimen for six months more. Exercise and CRD successfully improved markers of glucose metabolism, executive function and verbal memory, even after 6 months post-intervention (39). However, the lack of a control group in this study should be noted as a limitation. Ongoing analyses with data of this trial will investigate the impact of intervention on cerebral blood flow, and test whether improvements in cerebral blood flow and cardiorespiratory fitness would mediate enhancements in cognitive performance.
The Benefic Trial presenting a ketogenic approach with MCI volunteers suggests this to be a potential treatment pathway to reduce the risk of MCI progressing towards AD. In a similar way, findings from the BFIT Study based on physical activity and carbohydrate restriction approach with pre-diabetic subjects raise the possibility that preventing pre-diabetes progression towards diabetes would simultaneously prevent cognitive impairment. Future investigations with different populations are required to test whether the CRD without the exercise arm would mimic the benefits, and to help establishing these nutrition-based possibilities as part of the new generation of AD treatments.


Take home lessons

● Single-nutrient approaches based on omega-3 PUFA supplementation suggested that long-term supply (i.e., more than 3 years) might be needed for observing positive effects on cognitive performance. Findings from the PUFA Trial and the MAPT Study also raise the possibility that treatments may differ according to omega-3 status and brain characteristics (such as WMH and amyloid burden) of individuals.
● Multi-nutrient approaches comprising omega-3 PUFA, choline, minerals and vitamins are believed to promote synergistic actions that increase effect magnitude on protecting cognitive function. At CTAD 2020, trials testing this novel approach provided mixed evidence, suggesting that medium to long-term follow-ups might be needed in order to identify positive effects. In the LipiDiDiet Trial, reduced progression of hippocampal, ventricular and whole brain atrophy were observed after 36 months, but not after 2 years of intervention. The NOLAN Study was unable to find positive effects in cognitive function over a 1-year frame, but successfully increased erythrocyte omega-3 and decreased plasma homocysteine.
● Supplements driving ketogenic status in the Benefic Trial were shown to directly enhance brain energy status by the increased supply of ketones, resulting in improvements in cognitive performance of older adults with MCI after 6 months, and to be safe. The ketogenic approach for protecting cognitive function deserves further investigations.
● Among people with impaired glucose regulation, a combined 12-week intervention of carbohydrate-restricted diet and exercise was shown to reduce insulin resistance and improve cognitive function, even after 6 months post-intervention. Whether the dietary arm of the BFIT Study would provide similar benefits without the increased physical activity demands additional exploring.
● Nutrition-based treatments were offered as oral supplements. Products were well tolerated and no indication for safety concerns was observed. None of the provided nutrients achieved their tolerable upper intake level (UL). On the other hand, the lack of success in some studies comes up with the possibility that the provided dose of nutrients may have been insufficient.

In spite of the unquestionable high value of drug treatments, nutritional approaches stand out as alternatives to prevent and/or slow cognitive decline, contributing to define the next generation of AD preventive measures and treatments, in order to promote a healthier aging process. Optimizing treatments through nutrition might be especially interesting in the sense that, differently from drugs with specific targets, nutrients are widely recognized by different types of cells and are part of several metabolic mechanisms. Considering that older people are at higher risk of malnutrition [40] and micronutrients deficiency (41, 42), interventions based on nutrients may also help treating and/or preventing other comorbidities typically developed with the advancing of age. The exact way on how the aforementioned interventions or similar ones may do that deserves further exploring.
By joining worldwide leaders in the treatment of AD together, the CTAD conference allowed profitable discussions on candidate therapeutics and methodological issues related to this field of research, bringing crucial elements for optimizing the development of effective AD treatments. The next CTAD conference, which will be held in Boston (USA) from November 9th through 12th, 2021, is expected to bring additional and novel elements that will help designing successful clinical trials and increasing diversity of future AD treatments.


Abbreviations: Aβ: amyloid-β; AD: Alzheimer’s disease; BFIT: Blood Flow Improvement Trial; CDR-SB: Clinical Dementia Rating – Sum of Boxes; CRD: carbohydrate-restricted diet; CSF: cerebrospinal fluid; CTAD: Clinical Trials on Alzheimer’s Disease; DHA: docosahexaenoic acid; EPA: eicosapentaenoic acid; IR: insulin resistance; ITT: intention to treat; IWG: International Working Group; kMCT: ketogenic medium chain triglycerides; MAPT: Multidomain Alzheimer Preventive Trial; MCI: mild cognitive impairment; mITT: modified intention to treat; MRI: magnetic resonance imaging; PET: positron emission tomography; PUFA: polyunsaturated fatty acids; UL: tolerable upper intake level; WMH: white matter hyperintensities.

Conflicts of interest: None.

Open Access: This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits use, duplication, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.



1. Giordano E, Visioli F. Long-chain omega 3 fatty acids: molecular bases of potential antioxidant actions. Prostaglandins Leukot Essent Fatty Acids. 2014;90(1):1–4.
2. Shibabaw T. Omega-3 polyunsaturated fatty acids: anti-inflammatory and anti-hypertriglyceridemia mechanisms in cardiovascular disease. Mol Cell Biochem. 2020 Nov 11; doi: 10.1007/s11010-020-03965-7. Online ahead of print.
3. Dyall SC. Long-chain omega-3 fatty acids and the brain: a review of the independent and shared effects of EPA, DPA and DHA. Front Aging Neurosci. 2015;7:52.
4. Chitre NM, Moniri NH, Murnane KS. Omega-3 Fatty Acids as Druggable Therapeutics for Neurodegenerative Disorders. CNS Neurol Disord Drug Targets. 2019;18(10):735–49.
5. Bowman GL, Silbert LC, Howieson D, et al. Nutrient biomarker patterns, cognitive function, and MRI measures of brain aging. Neurology. 2012;78(4):241–9.
6. Tan ZS, Harris WS, Beiser AS, et al. Red blood cell ω-3 fatty acid levels and markers of accelerated brain aging. Neurology. 2012;78(9):658–64.
7. Bowman GL, Silbert LC, Dodge HH, et al. Randomized Trial of Marine n-3 Polyunsaturated Fatty Acids for the Prevention of Cerebral Small Vessel Disease and Inflammation in Aging (PUFA Trial): Rationale, Design and Baseline Results. Nutrients. 2019;11(4):735.
8. Bowman GL, Murchison C, Silbert L, et al. Effects of omega-3 (n-3) polyunsaturated fatty acids (PUFA) on cerebral white matter hyperintensities, medial temporal lobe atrophy and white matter integrity in older nondemented adults: A 3-year randomizedcontrolled phase 2 trial. J Prev Alz Dis 2020; 7(S1):S18.
9. Andrieu S, Guyonnet S, Coley N, et al. Effect of long-term omega 3 polyunsaturated fatty acid supplementation with or without multidomain intervention on cognitive function in elderly adults with memory complaints (MAPT): a randomised, placebo-controlled trial. Lancet Neurol. 2017;16(5):377–89.
10. Delrieu J, Payoux P, Carrié I, et al. Multidomain intervention and/or omega-3 in nondemented elderly subjects according to amyloid status. Alzheimers Dement J Alzheimers Assoc. 2019;15(11):1392–401.
11. Nakamura A, Kaneko N, Villemagne VL, et al. High performance plasma amyloid-β biomarkers for Alzheimer’s disease. Nature. 2018;554(7691):249–54.
12. Bateman RJ, Blennow K, Doody R, et al. Plasma Biomarkers of AD Emerging as Essential Tools for Drug Development: An EU/US CTAD Task Force Report. J Prev Alzheimers Dis. 2019;6(3):169–73.
13. Wang X, Sun Y, Li T, Cai Y, Han Y. Amyloid-β as a Blood Biomarker for Alzheimer’s Disease: A Review of Recent Literature. J Alzheimers Dis JAD. 2020;73(3):819–32.
14. Delrieu J, Vellas B, Cantet C, Bateman R, Andrieu S. Multidomain intervention and/or omega 3 in non-demented subjects according to plasma Aβ42/40 ratio: cognitive impact at 3 and 5 years in a subgroup analysis from the randomized clinical MAPT trial. J Prev Alz Dis 2020; 7(S1):S48-S49.
15. Pan Y, Araujo JA, Burrows J, et al. Cognitive enhancement in middle-aged and old cats with dietary supplementation with a nutrient blend containing fish oil, B vitamins, antioxidants and arginine. Br J Nutr. 2013;110(1):40–9.
16. Pan Y, Kennedy AD, Jönsson TJ, Milgram NW. Cognitive enhancement in old dogs from dietary supplementation with a nutrient blend containing arginine, antioxidants, B vitamins and fish oil. Br J Nutr. 2018;119(3):349–58.
17. Smith AD, Smith SM, de Jager CA, et al. Homocysteine-lowering by B vitamins slows the rate of accelerated brain atrophy in mild cognitive impairment: a randomized controlled trial. PloS One. 2010;5(9):e12244.
18. Oulhaj A, Jernerén F, Refsum H, Smith AD, de Jager CA. Omega-3 Fatty Acid Status Enhances the Prevention of Cognitive Decline by B Vitamins in Mild Cognitive Impairment. J Alzheimers Dis JAD. 2016;50(2):547–57.
19. Li M, Li W, Gao Y, et al. Effect of folic acid combined with docosahexaenoic acid intervention on mild cognitive impairment in elderly: a randomized double-blind, placebo-controlled trial. Eur J Nutr. 2020 Aug 28; doi: 10.1007/s00394-020-02373-3. Online ahead of print.
20. Soininen H, Solomon A, Visser PJ, et al. 24-month intervention with a specific multinutrient in people with prodromal Alzheimer’s disease (LipiDiDiet): a randomised, double-blind, controlled trial. Lancet Neurol. 2017;16(12):965–75.
21. Hartmann T, Solomon A, Visser P, et al. LipiDiDiet results: 3-year evaluation of Fortasyn Connect in individuals with prodromal Alzheimer’s Disease. 2020; J Prev Alz Dis 2020; 7(S1):S15-S16.
22. Soininen H, Solomon A, Visser PJ, et al. 36-month LipiDiDiet multinutrient clinical trial in prodromal Alzheimer’s disease. Alzheimers Dement. 2021;17(1):29-40.
23. Smith AD, Refsum H. Homocysteine, B Vitamins, and Cognitive Impairment. Annu Rev Nutr. 2016;36:211–39.
24. Setién-Suero E, Suárez-Pinilla M, Suárez-Pinilla P, Crespo-Facorro B, Ayesa-Arriola R. Homocysteine and cognition: A systematic review of 111 studies. Neurosci Biobehav Rev. 2016;69:280–98.
25. Hooper C, De Souto Barreto P, Pahor M, Weiner M, Vellas B. The Relationship of Omega 3 Polyunsaturated Fatty Acids in Red Blood Cell Membranes with Cognitive Function and Brain Structure: A Review Focussed on Alzheimer’s Disease. J Prev Alzheimers Dis. 2018;5(1):78–84.
26. Giudici KV, Guyonnet S, Cantet C, et al. A 1-year randomized controlled trial of a nutritional blend to prevent cognitive decline among community-dwelling older adults: the NOLAN Study. J Prev Alz Dis 2020; 7(S1):S43-S44.
27. Owen OE, Morgan AP, Kemp HG, Sullivan JM, Herrera MG, Cahill GF. Brain metabolism during fasting. J Clin Invest. 1967;46(10):1589–95.
28. Castellano C-A, Nugent S, Paquet N, et al. Lower brain 18F-fluorodeoxyglucose uptake but normal 11C-acetoacetate metabolism in mild Alzheimer’s disease dementia. J Alzheimers Dis JAD. 2015;43(4):1343–53.
29. Cunnane SC, Courchesne-Loyer A, St-Pierre V, et al. Can ketones compensate for deteriorating brain glucose uptake during aging? Implications for the risk and treatment of Alzheimer’s disease. Ann N Y Acad Sci. 2016;1367(1):12–20.
30. Croteau E, Castellano CA, Fortier M, et al. A cross-sectional comparison of brain glucose and ketone metabolism in cognitively healthy older adults, mild cognitive impairment and early Alzheimer’s disease. Exp Gerontol. 2018;107:18–26.
31. Cunnane SC, Trushina E, Morland C, et al. Brain energy rescue: an emerging therapeutic concept for neurodegenerative disorders of ageing. Nat Rev Drug Discov. 2020;19(9):609–33.
32. Avgerinos KI, Egan JM, Mattson MP, Kapogiannis D. Medium Chain Triglycerides induce mild ketosis and may improve cognition in Alzheimer’s disease. A systematic review and meta-analysis of human studies. Ageing Res Rev. 2020;58:101001.
33. Cunnane S, Fortier M, Castellano A, et al. Ketones improve brain energetics and cognitive performance in mild cognitive impairment: Final results of the 6-month Benefic trial in MCI. J Prev Alz Dis 2020; 7(S1):S10-S11.
34. Fortier M, Castellano C-A, St-Pierre V, et al. A ketogenic drink improves cognition in mild cognitive impairment: Results of a 6-month RCT. Alzheimers Dement. 2020 Oct 26; doi: 10.1002/alz.12206. Online ahead of print.
35. Xue M, Xu W, Ou Y-N, et al. Diabetes mellitus and risks of cognitive impairment and dementia: A systematic review and meta-analysis of 144 prospective studies. Ageing Res Rev. 2019;55:100944.
36. Cooper C, Sommerlad A, Lyketsos CG, Livingston G. Modifiable predictors of dementia in mild cognitive impairment: a systematic review and meta-analysis. Am J Psychiatry. 2015;172(4):323–34.
37. Pal K, Mukadam N, Petersen I, Cooper C. Mild cognitive impairment and progression to dementia in people with diabetes, prediabetes and metabolic syndrome: a systematic review and meta-analysis. Soc Psychiatry Psychiatr Epidemiol. 2018;53(11):1149–60.
38. Biessels GJ, Staekenborg S, Brunner E, Brayne C, Scheltens P. Risk of dementia in diabetes mellitus: a systematic review. Lancet Neurol. 2006;5(1):64–74.
39. Heston MB, Gaitan JM, Ma Y, et al. Exercise and carbohydrate-restricted diet associates with improved insulin resistance and cognitive performance. J Prev Alz Dis 2020; 7(S1):S89.
40. Fávaro-Moreira NC, Krausch-Hofmann S, Matthys C, et al. Risk Factors for Malnutrition in Older Adults: A Systematic Review of the Literature Based on Longitudinal Data123. Adv Nutr. 2016;7(3):507–22.
41. Borg S ter, Verlaan S, Hemsworth J, et al. Micronutrient intakes and potential inadequacies of community-dwelling older adults: a systematic review. Br J Nutr. 2015;113(8):1195–206.
42. Vural Z, Avery A, Kalogiros DI, Coneyworth LJ, Welham SJM. Trace Mineral Intake and Deficiencies in Older Adults Living in the Community and Institutions: A Systematic Review. Nutrients. 2020;12(4):1072.




C. Roberts1, J. Kaplow2, M. Giroux2, S. Krause2, M. Kanekiyo2


1. Eisai Ltd., Hatfield, UK; 2. Eisai Inc., Woodcliff Lake; NJ, USA

Corresponding Authors: Claire Roberts, Eisai Ltd., Hatfield, UK, Email:, Phone: +44 8456 761 590

J Prev Alz Dis 2021;
Published online January 25, 2021,



BACKGROUND/OBJECTIVES: Elenbecestat, an oral BACE-1 inhibitor that has been shown to reduce Aβ levels in cerebrospinal fluid, was investigated in two global phase 3 studies in early AD. Here we report on differences observed in characteristics of APOE ε4 and amyloid positive subjects in the large screening cohort.
DESIGN: Screening was performed in 5 sequential tiers over a maximum of 80 days, as part of placebo controlled, double blind phase 3 studies.
SETTING: Subjects were evaluated at sites in 7 regions (29 countries).
PARTICIPANTS: Overall, 9758 subjects were screened.
INTERVENTION: All screened subjects that were eligible received either placebo or 50 mg QID elenbecestat post randomisation.
MEASUREMENTS: Gender, disease staging, APOE ε4 status, amyloid status, amyloid positron emission tomography (PET) standard uptake value ratio (SUVr) and amyloid PET Centiloid (CL) values were determined for screened subjects; by country and region.
RESULTS: In this program, 44% of subjects were APOE ε4 positive. Frequency of females was similar in both APOE ε4 positive and negative groups. However, early mild AD subjects were slightly higher in the APOE ε4 positive group compared with the APOE ε4 negative group. 56% of subjects were amyloid positive. The mean age in the amyloid positive group was slightly higher than the amyloid negative group. The gender distribution was similar between amyloid groups. A lower number of mild cognitive impairment was observed in the amyloid positive group along with a higher number of early mild AD. APOE ε4 positive subjects were higher in amyloid positive group compared to the amyloid negative group. China had the lowest APOE ε4 and amyloid positivity rates with Western Europe and Oceania performing best. Subjects received florbetapir, florbetaben or flutemetamol amyloid PET tracer. Amyloid negative and positive subjects CL values were normally distributed around their respective means of 1.5 CL and 83 CL. However, there was an appreciable overlap in the 20-40 CL range.
CONCLUSIONS: In this large cohort of cognitively impaired subjects, subject demographics characteristics were comparable regardless of APOE genotype or amyloid positivity. APOE ε4 positivity and amyloid positivity varied by country and by geographical region.

Key words: MissionAD, Amyloid, APOE, elenbecestat.


Alzheimer’s disease (AD) is characterised by the deposition of amyloid-beta (Aβ) aggregates and neurofibrillary tangles in the brain (1). The amyloid cascade hypothesis proposes that Aβ aggregates trigger the spreading of tau-related neurofibrillary tangles and subsequent neuronal degeneration (2). Recently, it has been shown that Aβ accumulation precedes tau and seems to accelerate neocortical tau pathology (3). In addition, amyloid accumulation has been shown to be poorly correlated with cognitive impairment (4), while tau accumulation shows a high correlation to neurodegeneration and loss of cognitive function (5). Aβ is produced when amyloid precursor protein (APP) is cleaved sequentially by β-site APP–cleaving enzyme 1 (BACE-1; also referred to as β-secretase) and γ-secretase (6). Inhibition of BACE-1 is a potential therapeutic strategy for slowing the progression of Alzheimer’s disease by reducing the production of Aβ.
Elenbecestat is an oral BACE-1 inhibitor that has been shown to reduce the Aβ level in the cerebrospinal fluid (CSF). Aβ (1-x) is a compilation of all amyloid isoforms that are cleaved by the BACE enzyme at the 1 position, that include at least the first 24 amino acids. CSF Aβ (1-x) decreased by 62% compared to placebo in healthy volunteer phase 1 study (E2609-A001-002) at 50mg; increasing to 74% at 100mg and 85% at 400mg (7). This was confirmed in patients with mild cognitive impairment (MCI) and early mild AD in an elenbecestat phase 2 study (Study E2609-G000-201)[8] indicating an average of 69% decrease in CSF Aβ(1-x) with 50mg of elenbecestat.
Elenbecestat was investigated in two global phase 3 studies (E2609-G000-301 or MissionAD1; E2609-G000-302 or MissionAD2) in early AD. The population consisted of subjects with a diagnosis of MCI due to AD and no more than 25% diagnosed as early stage mild dementia due to AD.
The studies were recommended to terminate early by the programme Data Safety Monitoring Board in September 2019, following an unfavourable benefit-risk profile of elenbecestat. At the time of termination approximately 959 subjects had reached 12 months of treatment (500 placebo, 459 elenbecestat). There was no evidence of potential efficacy and the adverse event profile was worse than placebo. At that time elenbecestat was the only BACE inhibitor programme ongoing, following similar outcomes with BACE inhibitor studies with verubecestat (9-10), lanabecestat (11), atabecestat (12-13), and umibecestat (14).
At the time of the early termination the MissionAD studies were fully recruited with 2212 randomised subjects. This has resulted in a large cohort of 9758 screened subjects, establishing eligibility through cognitive assessments, laboratory assessments, MRI safety and amyloid status; representing a large volume of information for the Alzheimer’s research community. Here we report on differences observed in characteristics of APOE ε4 and amyloid positive subjects in this large screening cohort. Preliminary data was presented at AAIC and CTAD conferences during the past two years (15-17).



The MissionAD screening process was performed in 5 sequential tiers over a maximum of 80 days. Cognitive assessments, medical history and clinical diagnosis were determined in tier 1. Questionnaires were administered in tier 2 to establish baseline quality of life and to assess any suicidality risk. Laboratory assessments, including APOE ε4 status, were conducted in tier 3. An MRI scan was done at tier 4 and finally an amyloid PET scan or a CSF sample was taken to determine amyloid status in tier 5.

Amyloid PET Scans

Three amyloid PET tracers were used for amyloid assessment depending on availability of the tracer manufacturing facilities: florbetapir (Amyvid™), florbetaben (Neuraceq™) or flutemetamol (Vizamyl™). Florbetaben was prioritised if more than one tracer was available at an imaging facility. Florbetaben and flutemetamol emission acquisitions require a 20-minute scan, 90-110-minute post injection. Florbetapir emission acquisition requires a 20 minute scan, 50-70 minutes post injection.
Amyloid PET status was assessed centrally by visual read, by a radiologist blinded to cognitive status. The label of each tracer defines the number of positive regions to claim positivity on a visual read. All scans were analysed by readers trained on the guidelines established by the manufacturer. Reading for visual positivity includes the following regions for analysis:
• Florbetapir (Amyvid™) Frontal cortex (excluding midline), medial frontal cortex (including Anterior cingulate), parietal cortex (excluding midline), medial parietal cortex (precuneus and/or posterior cingulate), temporal cortex, and occipital cortex.
• Florbetaben (Neuraceq™) Lateral temporal, frontal lobes, posterior cingulate/precuneus, parietal lobes.
• Flutemetamol (Vizamyl™) Frontal lobes (axial & sagittal views), posterior cingulate and precuneus (sagittal & coronal views), temporal lobes – lateral regions (axial views – coronal views as supportive), parietal lobes – lateral regions (coronal views – axial views as supportive), striatum (axial views – sagittal views as supportive).

Amyloid PET SUVr data was calculated as mean composite SUVr (a simple average of cingulate, frontal, parietal, and temporal corticies) and mean composite SUVr including occipital region (a simple average of cingulate, frontal, parietal, temporal, and occipital corticies) using whole cerebellum as reference region.
In addition, amyloid PET data was combined across the three tracers using the Centiloid methodology for the mean composite SUVr. The Centiloid project has driven the creation of a 100-point scale termed “Centiloid (CL),” which is an average value of zero in “high certainty” amyloid negative subjects and an average of 100 in “typical” AD patients (Klunk et al., 2015;

Centiloid Conversion

The following equations were used to calculate the mean composite SUVr in Centiloid for each tracer as derived by Bioclinica (18, 19): Florbetapir: 205.72 * mean composite SUVr – 209.63; Florbetaben: 175.57 * mean composite SUVr – 173.21; Flutemetamol: 145.58 * mean composite SUVr – 139.29. All tracers: Mean composite SUVr with whole cerebellum region used as reference region is used.

CSF Amyloid Analysis

CSF sample collection was an option in this study, to determine baseline amyloid burden status. This could replace or be in addition to a baseline amyloid scan. During the study two platforms were utilized for analysis; Aβ(1-42) <250 pg/mL from Alzbio3 run at the ADNI core lab (Dr. Leslie Shaw) and the Lumipulse™ platform from Fujirebio, using a total tau:Aβ(1-42) ratio greater than 0.37 to indicate positive amyloid status. The Lumipulse CSF cutpoint was established (20); note: >0.37 ratio was set prior to the incorporation of the issued IRMM Aβ(1-42) reference standard.

APOE Status Determination

APOE genotyping was performed using a real time PCR Taqman assay, developed and performance determined by Brooks Life Sciences. The subjects were categorized into APOE ε4 positive if they had at least one ε4 allele, and negative if they did not have an ε4 allele.


Overall, 9758 subjects were screened in MissionAD, of which 5710 had a known APOE status, 4121 had a known amyloid status (PET or CSF), 4077 had a known APOE and amyloid status, 3492 had a known amyloid status and available SUVr data, and 2212 subjects were randomised (Table 1). The majority of subjects, that reached tiers 3-5, had a MMSE score ≥24, CDR Global score of 0.5, and a cognitive impairment of ≥1 standard deviation from age-adjusted norms in the International Shopping List Task.

Table 1. Recruitment status


APOE Genotype

Of the 5,710 screened subjects with known APOE genotype, 44% were APOE ε4 positive (Table 2). The mean age in both the APOE groups was 71 years. Frequency of females was similar in both APOE ε4 positive and negative groups (51% and 52%). Early mild AD subjects were slightly higher in the APOE ε4 positive group compared with the APOE ε4 negative group (15% and 10%).

Table 2. Demography of APOE ε4 and amyloid positive and negative screened subjects

*The percentages are based on the subjects with ApoE4 status. **The percentage are based on the subjects with amyloid status (PET or CSF). Other percentages are based on each group.


Amyloid Status

Of the 4,077 screened subjects with known APOE and amyloid status (as determined either by PET or CSF), 56% were amyloid positive (Table 2). The mean age in the amyloid positive group was slightly higher (72 years c.f. 69 years). The gender distribution was similar between amyloid groups. A lower number of MCI was observed in the amyloid positive group (84% and 90%) along with a higher number of early mild AD (16% and 8%). APOE ε4 positive subjects were higher in amyloid positive group compared to the amyloid negative group (64% and 22%).

APOE and Amyloid Status Varies Depending on Region & Country

29 countries spanning the 7 regions of North America (Canada, USA), South America (Argentina, Chile, Mexico), Western Europe & Oceania (Australia, Austria, Denmark, Finland, France, Germany, Greece, Italy, Portugal, South Africa, Spain, UK), Eastern Europe (Bulgaria, Croatia, Czech Republic, Hungary, Poland, Russia, Slovakia), China, Japan, and Other Asia (Singapore, South Korea, Taiwan) participated in the studies.
Mean APOE ε4 positivity across the countries was 48%. This was lowest in Mexico < Taiwan < Greece < Bulgaria < Slovakia, but care needs to be used where numbers are too low to interpret. The highest rate was in France > Australia > Finland > Hungary > South Africa (Table 3). Mean amyloid positivity was 64% across the countries. This was lowest in Singapore < Croatia < Poland < China < Taiwan and highest in Greece > France > Italy > Hungary > Australia (Table 3). Grouping the countries into regions reduced variability. The mean APOE ε4 positivity and amyloid positivity across the regions was 46% and 59%, respectively. China had the lowest APOE ε4 and amyloid positivity rates, while Western Europe and Oceania the highest rates (Figure 1).

Table 3. APOE and amyloid status by country

Figure 1. APOE and amyloid status by region (Dotted red lines indicate regional means)


Amyloid PET Status

53% of Subjects screened with a known amyloid status and amyloid PET SUVr data were amyloid positive. Overall, 1563 (45%) subjects were APOE ε4 positive, 1816 (52%) were female, and 3028 (87%) were MCI and 420 (12%) mild early AD (Table 2).

Amyloid PET SUVr

In total, 386, 2548 and 558 subjects received florbetapir, florbetaben or flutemetamol amyloid PET tracer, of which 218 (56%), 1292 (51%) and 330 (59%) were determined visually to be amyloid positive, respectively. Amyloid PET mean composite (SD; min – max) SUVr for amyloid negative subjects were 1.03 (0.11; 0.80 – 1.48), 0.99 (0.08; 0.70 – 1.45) and 0.98 (0.11; 0.80 – 1.38) for florbetapir, florbetaben or flutemetamol, respectively. Amyloid PET mean composite (SD; min – max) SUVr for amyloid positive subjects were 1.41 (0.18; 0.94 – 1.94), 1.47 (0.22; 0.80 – 2.22) and 1.49 (0.19; 0.95 – 1.99) for florbetapir, florbetaben or flutemetamol, respectively. Including the occipital region in the mean composite SUVR had very little impact on the data.

Amyloid PET Conversion to Centiloid Mean Composite

Centiloid (CL) conversion, combining data for the 3 tracers, calculated a mean (SD) of 1.5 (15.2) CL for amyloid negative subjects (n=1652) and 82.7 (36.7) CL for amyloid positive (n=1840). No differences were observed in the individual CL values between the tracers.

Mean Composite Centiloid Distribution

Amyloid negative and positive subjects CL values were normally distributed around their respective means of 1.5 CL and 83 CL. However, there was an appreciable overlap in the 20-40 CL range; minimum and maximum values of -50 and 95 CL for amyloid negative and 33 and 217 CL for amyloid positive subjects (Figure2).
There was no obvious impact of gender on the distribution of CL, with mean (SD) composite of 85 (35.6) CL and 81 (37.8) CL for amyloid positive, and 3.2 (16.1) CL and 0 (14.0) CL for amyloid negative, for females and males respectively.
More amyloid positive subjects were APOE ε4 positive (65%) compared to APOE ε4 negative (35%), with the mean 10 CL higher in APOE ε4 positive (86 CL, SD 34.2) than APOE ε4 negative (76 CL, SD 40.4) in the amyloid positive population. A majority (77%) of the APOE ε4 positive subjects were also amyloid positive.
Of the 420 mild early AD subjects 69% were amyloid positive, while in the 3028 MCI subjects 51% were amyloid positive. There was a shift in the mean by approximately 9 CL in the amyloid positive cohort, 81 CL (SD 36.6) for MCI subjects and 90 CL (SD 36.8) for mild early AD subjects.

Figure 2. Distribution of mean composite Centiloid for amyloid negative and positive subjects



Alzheimer’s Disease clinical trials, due to the complexity required to identify the correct population, include many screening assessments that encompass not only traditional elements but also multiple cognitive, biomarker and imaging assessments. Especially in large global studies, there is an expectation of a high screen failure rate, while tiered screening is often required in order to reduce unnecessary burden on subjects and control costs.
For MissionAD the screening approach was to conduct the cognitive assessments and study specific clinical diagnosis in the first tier to eliminate the highest proportion of subjects not suitable for the trial. Approximately 41% of subjects screen failed due to cognitive or suicidality assessments. About 17% further subjects failed due to lab assessments. Finally, in the last tier a further 19% of subjects screen failed when amyloid status was determined to be negative. Overall, the screen fail rate in the early AD population (MCI and early mild AD) of MissionAD was 77%. This screen failure rate is in the same range as reported in studies with similar populations. For example, the verubecestat APECS study in prodromal AD had a SF rate of 68% (10), the AMARANTH and DAYBREAK studies with lanabecestat in early and mild AD had SF rates of 68% and 70%, respectively (11). The slightly higher rate in MissionAD may be explained by the focus on recruiting a higher proportion of MCI subjects (>75%) in this program.
When assessed at the end of the screening cascade, amyloid positivity (amyloid PET visual read or amyloid CSF) was a global average of 56%. Of note, the positivity rates varied from country to country. It is important to understand where rates may be low, so that recruitment enrichment strategies can be employed in those areas. Countries where the positivity rates were low may be reflective of populations that more readily report subjective cognitive decline or the “worried well”, lack of access to technology, or lack of access to care. Socioeconomic status and different healthcare systems may result in a greater focus on subjective cognitive symptoms. Lowest rates of positivity in MissionAD were observed in North America and China. APOE ε4 positivity also varied by country and was lowest in North America and China. Summation of these components supported amyloid positivity as highly correlated with APOE ε4 positivity.
These data reinforce the need to be aware of the variability in APOE ε4 and amyloid positivity when designing clinical trials, and the importance of having strategies available to mitigate. Various approaches were employed in this trial to maximise eligible subjects, however, the screen failure rate for MissionAD was still 77% (2212 randomised out of the 9758 screened subjects).
This study allowed the use of three different amyloid PET tracers with florbetaben being used in 75% of the subjects in this study. We presented the SUVr data as a mean composite with and without the occipital region for all tracers. However, it should be noted that per the manufacturers label the occipital region is appropriate only for florbetaben. Data indicate that the tracers were comparable and could be grouped for comparisons by using the Centiloid methodology supporting the use of multiple tracers in a global trial to expand countries able to participate.
Amyloid negative and positive subjects Centiloid values were normally distributed with an appreciable overlap in the 20-40 CL range. Distribution was shifted to higher Centiloid values by APOE ε4 positive status and mild early AD diagnosis. Similar results have been shown by other groups when examining large cohort studies (21) and have revealed similar, bimodal distributions of CL in the population studied (22). These results reveal similar factors influenced amyloid positivity as was evident with amyloid PET visual read.
In this large cohort of cognitively impaired subjects, just under half were positive for APOE ε4 and just over half were amyloid positive. APOE ε4 positivity and amyloid positivity varied by country and by geographical region. Subject demographics characteristics were comparable regardless of APOE genotype or amyloid positivity. A higher APOE ε4 or amyloid positivity rate generally reflected a higher identification of early AD subjects in most regions. APOE ε4 positive subjects are more likely to have elevated amyloid.


Acknowledgments: Study participants and all sites that took part in MissionAD. Editorial support provided by JD Cox, PhD and Mayville Medical Communications.

Funding: Funding for the studies, analyses, and editorial support was provided by Eisai Inc.

Disclosures: All authors are, or were, employees of Eisai Inc or Eisai Ltd.

Ethical standards: The study was performed in full compliance with International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use and all applicable local Good Clinical Practice and regulations.

Open Access: This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits use, duplication, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.



1. Masters CL, Bateman R, Blennow K, Rowe CC, Sperling RA, Cummings JL. Alzheimer’s disease. Nat Rev Dis Primers 2015;1:15056.
2. Selkoe DJ, Hardy J. The amyloid hypothesis of Alzheimer’s disease at 25 years. EMBO Mol Med 2016;8:595–608.
3. Hanseeuw BJ, Betensky RA, Jacobs HI, et al. Association of Amyloid and Tau With Cognition in Preclinical Alzheimer Disease: A Longitudinal Study. JAMA Neurol. 2019;76(8):915-924.
4. Jack CR, Knopman DS, Jagust WJ, et al. Hypothetical model of dynamic biomarkers of the Alzheimer’s pathological cascade. Lancet Neurol. 2010;9(1):119-28.
5. Duyckaerts C, Brion JP, Hauw JJ, Flament-Durand J. Quantitative assessment of the density of neurofibrillary tangles and senile plaques in senile dementia of the Alzheimer type. Comparison of immunocytochemistry with a specific antibody and Bodian’s protargol method. Acta Neuropathol. 1987;73(2) :167-170.
6. Yan R, Vassar R. Targeting the β secretase BACE1 for Alzheimer’s disease therapy. Lancet Neurolog. 2014 ;13:319–29.
7. Albala B, Kaplow J, Lai R, Matijevic M, Aluri J, Satlin A. CSF Amyloid Lowering in Human Volunteers after 14 days Oral Administration of the Novel BACE1 Inhibitor E2609. 2012. AAIC abstract published July 2012 (oral presentation).
8. Majid O, Gee M, Albala B, et al. Dose-related reductions of CSF amyloid β (1-x) by E2609, a novel BACE1 inhibitor, in patients with mild cognitive impairment due to Alzheimer’s disease (AD) and mild-to-moderate AD dementia. CTAD 2016;San Diego: P3-28 (poster presentation).
9. Egan MF, Kost J, Tairot PN et al. Randomised trial of verubecestat for mild-to-moderate Alzheimer’s disease. N. Eng. J. Med. 2018;378(18): 1691-1703.
10. Egan MF, Kost J, Voss T, et al. Randomized Trial of Verubecestat for Prodromal Alzheimer’s Disease. N Engl J Med. 2019;380(15):1408-1420.
11. Wessels AM, Tariot PN, Zimmer JA, et al. Efficacy and Safety of Lanabecestat for Treatment of Early and Mild Alzheimer Disease: The AMARANTH and DAYBREAK-ALZ Randomized Clinical Trials. JAMA Neurol. 2020 Feb 1;77(2):199-209.
12. Henley D, Raghavan N, Sperling R, Aisen P, Raman R, Romano G. Preliminary Results of a Trial of Atabecestat in Preclinical Alzheimer’s Disease. N Engl J Med. 2019;380(15):1483-1485.
13. Novak G, Streffer JR, Timmers M, et al. Long-term safety and tolerability of atabecestat (JNJ-54861911), an oral BACE1 inhibitor, in early Alzheimer’s disease spectrum patients: a randomized, double-blind, placebo-controlled study and a two-period extension study. Alzheimers Res Ther. 2020;12(1):58.
14. Graf A, Borowsky B, Tariot P, et al. Alzheimer’s Prevention Initiative Generation Program: Update And Next Steps. 2019. CTAD abstract [OC11].
15. Roberts C, Kaplow J, Giroux M, Krause S, Kanekiyo M. Amyloid burden assessed by three amyloid PET tracers in the elenbecestat MissionAD phase 3 program. 2020. AAIC abstract published, July 2020 (poster presentation).
16. Roberts C, Kanekiyo M, Albala B. Amyloid positive subject characteristics in the elenbecestat MissionAD phase 3 program. 2019. CTAD abstract published, December 2019 (poster presentation).
17. Roberts C, Kanekiyo M, Kaplow J, Albala B. APOE4 status and amyloid burden differences across regions in the elenbecestat MissionAD phase 3 program. 2019. AAIC abstract published, July 2019 (poster presentation).
18. Klunk WE, Koeppe RA, Price JC, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimers Dementia. 2015;11(1):1-15.
19. Adamczuk K, Sampat M, Bracoud L, Runkle M, Gorman B, Scott D. Centiloid scale in practice: Effects of different SUVr reference regions and comparison of centiloid cut-offs. Alzheimer’s & Dementia. 2019;15(7) suppl: 223-224.
20. Kaplow J, Vandijck M, Gray J, et al. Concordance of Lumipulse cerebrospinalfluid t-tau/Aβ42 ratio with amyloid PET status. Alzheimer’sDement. 2020;16:144–152.
21. Mormino E, Betensky RA, Hedden T, et al. Amyloid and APOE ε 4 interact to influence short-term decline in preclinical Alzheimer’s disease. Neurology. 2014;82:1760-1867.
22. La Joie R, Ayakta N, Seeley WW, Borys E et al. Multisite study of the relationships between antemortem [11C]PiB-PET Centiloid values and post-mortem measures of Alzheimer’s disease neuropathology. Alzheimer’s & Dementia. 2018; 1-12.


M. Canevelli1, N. Vanacore2, A. Blasimme3, G. Bruno1, M. Cesari4

1. Sapienza University, Rome, Italy; 2. National Center for Disease Prevention and Health Promotion, National Institute of Health, Rome, Italy; 3. Department of Health Sciences and Technology, ETH Zürich, Switzerland; 4. IRCCS Istituti Clinici Maugeri, University of Milan, Italy.

Corresponding Author: Matteo Cesari, MD, PhD, Geriatric Unit, IRCCS Istituti Clinici Scientifici Maugeri, Via Camaldoli 64 – 20138 Milan – Italy, Phone: +39 02 50725136, Twitter: @macesari, email:

J Prev Alz Dis 2020;
Published online December 31, 2020,



The management of frailty in older persons is not easy, implying interventions beyond the simple prescription of medications. Biological complexity, multimorbidity, polypharmacy, and social issues often hamper the possibility to directly translate the evidence coming from research into clinical practice. Frailty indeed represents the most relevant cause of the “evidence-based medicine issue” influencing clinical decisions in geriatric care. Today, patients with Alzheimer’s disease (AD) are much older and frailer than some decades ago. They also tend to have more drugs prescribed. In parallel, research on AD has evolved over the years, hypothesizing that anticipating the interventions to the earliest stages of the disease may provide beneficial effects (to date, still lacking). In this article, we argue that, by focusing exclusively on “the disease” and pushing to anticipate its detection (sometimes even before the appareance of its clinical manifestations) may overshadow the person’s values and priorities. Research should be developed for better integrating the concept of aging and frailty in the design of clinical trials in order to provide results that can be implemented in real life. On the other hand, clinicians should be less prone to the easy (but unsupported by evidence) pharmacological prescription.

Key words: Overtreatment, dementia, cognition, geriatrics, prevention.

The management of frailty in older persons is not easy, mainly because it requires interventions beyond the simple prescription of medications (1–5). Biological complexity, multimorbidity, polypharmacy, and social issues often hamper the possibility to translate the evidence coming from research into clinical practice directly. Frailty indeed represents the most relevant cause of the “evidence-based medicine issue”, which has traditionally complicated geriatric medicine (3, 6, 7). Moreover, the outcomes often used in clinical trials conducted in adult and/or non-frail individuals are often meaningless in older persons with frailty (8, 9). What is more, the priorities of older persons with chronic conditions are usually found on aspects that are not necessarily related to the cure of a given disease or the life extension (10, 11). Instead, considerations about the quality of life play a critical role that is routinely neglected in medical decision-making (8). Lack of regard for patient-centered care as well as the absence of evidence specifically supporting the clinical management of frail older persons often lead to the need of making decisions with the addition of a good dose of common sense.
At the same time, the increasing diagnostic capacity and the hyper-specialization of medicine (arguably a correlate of a culture that privileges functional integrity over life quality and emotional fulfillment) (12) have substantially contributed to the widespread phenomena of overdiagnosis and overtreatment (13). In particular, these problems of modern medicine stem from the incapacity to deal with the individual’s multidimensional complexity and the lack of integrated models of care. The former, caused by the heterogeneity of the underlying aging process (14, 15), is responsible for the “one-size-fits-all” paradigm pervasively promoted in medicine. The latter causes an approach to the patient that proceeds by siloes, failing to 1) realize the complexity of the system (16, 17), and 2) adequately integrate available clinical information with the person’s life plans, expectations, and preferences (3). After all, it is well-established that, while the robustness of scientific evidence tends to fade in the oldest and frailest individuals, these are paradoxically the ones who are the most exposed to polypharmacy (18). In this scenario, we cannot ignore the detrimental aspect of the defensive medicine19. Clinicians today identify the prescription of tests and medications as the final goal of their mission, and patients are not aware that sometimes “less is more” (20).
The field of Alzheimer’s disease (AD) has substantially been exposed to these risks. AD is a condition of old age. Today, AD patients are much older and frailer than some decades ago (21). They also tend to have more drugs prescribed. The complexity of the condition is also exemplified by the substantial role played by social factors (e.g., literacy) in its clinical expression (22).
Research on AD has been evolving over the years, hypothesizing that the anticipation of the interventions to the earliest stages of the disease may provide those beneficial effects that, to date, are still lacking. For this reason, a myriad of preclinical conditions has been developed. Constructs as mild cognitive impairment (23), prodromal AD (24), subjective cognitive decline (25)… have been designed for research purposes, in particular for defining conditions to target with experimental drugs. Although sometimes explained the rationale behind the design of these pre-AD conditions, their adoption in the clinical setting has always been quite rapid – a phenomenon known as “diagnostic creep” (26). This leads to an earlier diagnosis of AD (and its different forms), generating a more extended life “with the disease”, more heterogeneity in the biological/clinical profile, more patients to treat, and higher costs for the healthcare systems. By focusing exclusively on “the disease”, the push to anticipate its detection even before its preclinical manifestations appear (27) overshadows the person’s values and priorities.
Furthermore, it also seems that the risk of preciously treating individuals presenting early signs/symptoms of AD is never considered for its potential of reducing the effect size of the intervention. In other words, there is the possibility that an effective molecule might appear less effective than it is simply because administered to a population including false positive (i.e., non-diseased) individuals. Moreover, anticipating the diagnosing of a condition that, to date, has still no disease-modifier treatment is fraught with a host of not yet settled ethical issues (28–30). In particular, does an anticipated AD diagnosis really expand individual autonomy and self-determination (as many seem to argue), or is it instead conducive to self-depreciation, stigmatization, and further social isolation?
Moreover, let us consider some of the implications of treating AD preclinical phases as actual clinical entities. Recently, Egan et al. (31) published the results from a randomized, double-blind, placebo-controlled trial aimed at evaluating the effect of verubecestat, a beta-site amyloid precursor protein-cleaving enzyme-1 (BACE-1) inhibitor, in prodromal AD. Beyond the negative effect of the drug on cognitive function, what stood out from the report was the theoretically unexpected, high proportion of participants (i.e., 46.0 %) taking cholinesterase inhibitors or memantine at the study baseline. Such a high prevalence in the treatment of prodromal AD is unjustified and quite worrisome, given the likely inclusion of many potentially reversible cases among the participants (32). It is furthermore not negligible that this overtreatment occurs with medications that:
1) have proven to be ineffective at delaying/halting the progression to overt dementia (33);
2) may expose the individual to adverse drug reactions and even worsen the person’s functioning (34); and
3) are not approved for such use by international regulatory agencies.
The absence of alternative therapeutic options cannot justify these interventions, which should be considered as ethically, clinically, and economically inappropriate/unsustainable. Research should thus be developed for better integrating the concept of aging and frailty in the design of clinical trials in order to provide results that can be implemented in real life (8, 35–38). On the other hand, clinicians should be less prone to the easy (but unsupported by evidence) pharmacological prescription.


Conflicts of interest: Marco Canevelli is supported by a research grant of the Italian Ministry of Health (GR-2016-02364975) for the project “Dementia in immigrants and ethnic minorities living in Italy: clinical-epidemiological aspects and public health perspectives” (ImmiDem). Matteo Cesari has received honoraria for presentations at scientific meetings and/or research funding from Nestlé. No specific conflict of interest declared by the other authors.



1. de Kerimel J, Tavassoli N, Lafont C, et al. How to Manage Frail Older Adults in the Community? Proposal of a Health Promotion Program Experienced in a City of 16,638 Inhabitants in France. J Frailty Aging. 2018;7(2):120-126. doi:10.14283/jfa.2017.47
2. Sinclair AJ, Abdelhafiz A, Dunning T, et al. An International Position Statement on the Management of Frailty in Diabetes Mellitus: Summary of Recommendations 2017. J Frailty Aging. 2018;7(1):10-20. doi:10.14283/jfa.2017.39
3. Cesari M, Marzetti E, Thiem U, et al. The geriatric management of frailty as paradigm of “The end of the disease era”. Eur J Intern Med. 2016;31:11-14. doi:10.1016/j.ejim.2016.03.005
4. Negm AM, Kennedy CC, Thabane L, et al. Management of Frailty: A Systematic Review and Network Meta-analysis of Randomized Controlled Trials. J Am Med Dir Assoc. 2019;20(10):1190-1198. doi:10.1016/j.jamda.2019.08.009
5. Muscedere J. The Need to Implement Frailty in the International Classification of Disease (ICD). J Frailty Aging. 2020;9(1):2-3. doi:10.14283/jfa.2020.2
6. Balducci L. Evidence-based or evidence-biased medicine? J Med Pers. 2013;11:99-100.
7. Evans J. Evidence-based and evidence-biased medicine. Age Ageing. 1995;24(6):461-463.
8. Canevelli M, Remoli G, Toccaceli Blasi M, et al. Ongoing Research Protocols for the Pharmacological Treatment of Neuropsychiatric Symptoms in Dementia. J Frailty Aging. Published online July 7, 2020. doi:10.14283/jfa.2020.36
9. Guralnik J, Bandeen-Roche K, Bhasin S a. R, et al. Clinically Meaningful Change for Physical Performance: Perspectives of the ICFSR Task Force. J Frailty Aging. 2020;9(1):9-13. doi:10.14283/jfa.2019.33
10. Fried TR, Street RL, Cohen AB. Chronic Disease Decision Making and “What Matters Most.” J Am Geriatr Soc. 2020;68(3):474-477. doi:10.1111/jgs.16371
11. Renne I, Gobbens RJ. Effects of frailty and chronic diseases on quality of life in Dutch community-dwelling older adults: a cross-sectional study. Clin Interv Aging. 2018;13:325-334. doi:10.2147/CIA.S156116
12. Cesari M, Pérez-Zepeda M, Marzetti E. Frailty and multimorbidity: different ways of thinking about geriatrics. J Am Med Dir Assoc. 2017;18(4):361-364. doi:10.1016/j.jamda.2016.12.086
13. Canevelli M, Bruno G, Remiddi F, et al. Spontaneous Reversion of Clinical Conditions Measuring the Risk Profile of the Individual: From Frailty to Mild Cognitive Impairment. Front Med (Lausanne). 2017;4:184. doi:10.3389/fmed.2017.00184
14. Cesari M. Frailty and Aging. J Frailty Aging. 2012;1:3-6.
15. Cesari M, Vellas B, Gambassi G. The stress of aging. Exp Gerontol. 2013;48(4):451-456. doi:10.1016/j.exger.2012.10.004
16. Kiiti Borges M, Oiring De Castro Cezar N, Silva Santos Siqueira A, Yassuda M, Cesari M, Aprahamian I. The relationship between physical frailty and mild cognitive impairment in the elderly: a systematic review. J Frailty Aging. 2019;8(4):192-197. doi:10.14283/JFA.2019.29
17. Maltais M, de Souto Barreto P, Rolland Y, Vellas B. The Association of ApoE ε4 Status with Lower Limb Function and Handgrip Strength in Older Adults. J Frailty Aging. 2019;8(2):62-66. doi:10.14283/jfa.2019.7
18. Muth C, Blom JW, Smith SM, et al. Evidence supporting the best clinical management of patients with multimorbidity and polypharmacy: a systematic guideline review and expert consensus. J Intern Med. 2019;285(3):272-288. doi:10.1111/joim.12842
19. Snyder Sulmasy L, Weinberger SE. Better care is the best defense: high-value clinical practice vs defensive medicine. Cleve Clin J Med. 2014;81(8):464-467. doi:10.3949/ccjm.81a.14037
20. Dam A, Datta N, Mohanty U, Khanna P. Pharmaceutically Less and Holistically More. JAMA Intern Med. 2013;173(11):948. doi:10.1001/jamainternmed.2013.475
21. Canevelli M, Cesari M, Trobia F, et al. Sociodemographic and Clinical Changes Over Time of Individuals Evaluated for Cognitive Disturbances: Good or Bad News. J Am Med Dir Assoc. 2015;16(12):1095-1096. doi:10.1016/j.jamda.2015.09.013
22. Arce Rentería M, Vonk JMJ, Felix G, et al. Illiteracy, dementia risk, and cognitive trajectories among older adults with low education. Neurology. 2019;93(24):e2247-e2256. doi:10.1212/WNL.0000000000008587
23. Petersen R. Clinical trials for early (pre-dementia) Alzheimer’s disease: a case for mild cognitive impairment. J Nutr Health Aging. 2010;14(4):304-305.
24. Dubois B, Feldman HH, Jacova C, et al. Advancing research diagnostic criteria for Alzheimer’s disease: the IWG-2 criteria. Lancet Neurol. 2014;13(6):614-629. doi:10.1016/S1474-4422(14)70090-0
25. Jessen F, Amariglio R, van Boxtel M, et al. A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement. 2014;10(6):844-852. doi:10.1016/j.jalz.2014.01.001
26. McGregor MJ, Martin D. Testing 1, 2, 3: is overtesting undermining patient and system health? Can Fam Physician. 2012;58(11):1191-1193, e615-617.
27. Yassine HN. Targeting prodromal Alzheimer’s disease: too late for prevention? Lancet Neurol. 2017;16(12):946-947. doi:10.1016/S1474-4422(17)30372-1
28. Mattsson N, Brax D, Zetterberg H. To know or not to know: ethical issues related to early diagnosis of Alzheimer’s disease. Int J Alzheimers Dis. 2010;2010. doi:10.4061/2010/841941
29. Howe EG. Ethical Issues in Diagnosing and Treating Alzheimer Disease. Psychiatry (Edgmont). 2006;3(5):43-53.
30. Galvin JE. Screening for Mild Cognitive Impairment: There is the Will but Is There a Way? J Prev Alzheimers Dis. 2020;7(3):144-145. doi:10.14283/jpad.2020.16
31. Egan MF, Kost J, Voss T, et al. Randomized Trial of Verubecestat for Prodromal Alzheimer’s Disease. New England Journal of Medicine. 2019;380(15):1408-1420. doi:10.1056/NEJMoa1812840
32. Canevelli M, Grande G, Lacorte E, et al. Spontaneous Reversion of Mild Cognitive Impairment to Normal Cognition: A Systematic Review of Literature and Meta-Analysis. J Am Med Dir Assoc. Published online August 5, 2016. doi:10.1016/j.jamda.2016.06.020
33. Raschetti R, Albanese E, Vanacore N, Maggini M. Cholinesterase inhibitors in mild cognitive impairment: a systematic review of randomised trials. PLoS Med. 2007;4(11):e338. doi:10.1371/journal.pmed.0040338
34. Schneider LS, Insel PS, Weiner MW. Treatment With Cholinesterase Inhibitors and Memantine of Patients in the Alzheimer’s Disease Neuroimaging Initiative. Arch Neurol. 2011;68(1):58-66. doi:10.1001/archneurol.2010.343
35. Archibald M, Kitson A, Frewin D, Visvanathan R. Transdisciplinary Research In Frailty: Knowledge Translation To Inform New Models of Care. J Frailty Aging. 2017;6(2):62-64. doi:10.14283/jfa.2017.6
36. Rowe J, Fried L. Incorporating frailty into clinical practice and clinical research. J Frailty Aging. 2013;2(3):126-127. doi:10.14283/jfa.2013.17
37. Beard JR. Linking Geroscience and Integrated Care to Reinforce Prevention. J Prev Alzheimers Dis. 2020;7(2):68-69. doi:10.14283/jpad.2020.13
38. Takeda C, Guyonnet S, Sumi Y, Vellas B, Araujo de Carvalho I. Integrated Care for Older People and the Implementation in the INSPIRE Care Cohort. J Prev Alzheimers Dis. 2020;7(2):70-74. doi:10.14283/jpad.2020.8




F. McDougall1, C. Edgar2, M. Mertes3, P.Delmar3, P. Fontoura3, D. Abi-Saab3, C.J. Lansdall3, M. Boada4,5, R. Doody1,3


1. Genentech, South San Francisco, USA; 2. Cogstate Ltd, London, UK; 3. F. Hoffmann-La Roche Ltd, Basel, Switzerland; 4. Research Center and Memory Clinic, Fundació ACE, Institut Català de Neurociències Aplicades, Universitat Internacional de Catalunya, Barcelona, Spain; 5. Networking Research Center on Neurodegenerative Diseases (CIBERNED), Instituto de Salud Carlos III, Madrid, Spain.

Corresponding Author: Fiona McDougall, Genentech 620 E Grand Ave, South San Francisco, CA 94080, USA,

J Prev Alz Dis 2020;
Published online December 21, 2020,



BACKGROUND: The Clinical Dementia Rating–Sum of Boxes (CDR-SB) has been proposed as a primary outcome for use in prodromal AD trials. However, the psychometric properties of this, and of other commonly used measures, have not been well-established in this patient population.
OBJECTIVE: To describe the psychometric properties of commonly used efficacy measures in a clinical trial of prodromal AD.
SETTING: Data were gathered as part of a two-year clinical trial.
PARTICIPANTS: Patients had biomarker confirmed prodromal AD.
MEASUREMENTS: Clinical Dementia Rating (CDR), Functional Activities Questionnaire (FAQ), Alzheimer’s Disease Assessment Scale – Cognition Subscale 11 and 13 (ADAS-Cog), Mini Mental State Exam (MMSE), and Free and Cued Selective Reminding Test (FCSRT-IR [words]). Assessments were conducted at least every 24 weeks.
RESULTS: For the CDR-SB, test-retest reliability was good (intra-class correlation coefficient [ICC]=0.83); internal consistency was 0.65 at baseline but above 0.8 at later assessments. Relationships between the CDR-SB and other measures were as expected (higher correlations with more closely related constructs), and the CDR-SB differentiated between patients with different severities of dementia (-2.9 points difference between CDR-Global Score 0.5 and 1, P<.0001). Floor and ceiling effects on the CDR-SB total score were minimal; however, at baseline there were ceiling effects in the personal care domain. Further detail is provided on the psychometric properties of ADAS-Cog, MMSE, FCSRT-IR and FAQ in this population.
CONCLUSION: The psychometric properties of the CDR-SB are adequate in prodromal AD and continued use is warranted in clinical trials. However, there remains scope for improvement in the assessment of functional constructs and development of novel measures should continue.

Key words: Clinical dementia rating, prodromal Alzheimer’s disease, psychometric testing.



A number of clinical trials of potential disease-modifying treatments in Alzheimer’s Disease (AD) are now targeted at early stage disease, where it is thought that there will be the greatest benefit to patients. By targeting the disease at the prodromal and mild dementia stages (also referred to in this paper as “early AD”), it is hoped to slow progression before extensive, irreversible neurodegeneration occurs. Since AD may be viewed as a continuum with preclinical, prodromal and dementia stages (mild, moderate and severe), the dementia diagnosis itself may be an important milestone in progression, but not one that represents a natural or stark differentiating boundary in terms of underlying pathophysiology. Diagnostic criteria for prodromal AD (pAD) (1) and mild cognitive impairment (MCI) due to AD (2) are now established; however, existing outcome measures to assess efficacy were mainly developed and validated for overt dementia and so may be unsuitable for clinical trials in this earlier patient population.
The FDA and EMA have both called for novel approaches to assess efficacy, recognizing the limitations of existing instruments in the earliest stages of AD. The FDA guidance and EMA guidelines (3, 4) have stated that clinical trials in the dementia stage of AD should use a co-primary approach, in which a treatment should demonstrate efficacy on both a cognitive measure and a functional measure (3, 5). This has been described as intending to ensure “that a clinically meaningful effect was established by a demonstration of benefit on the functional measure and that the observed functional benefit was accompanied by an effect on the core symptoms of the disease as measured by the cognitive assessment” (3). However, in the early stages of AD (stages 3 and 4), spanning pAD and mild AD dementia (mAD), it is recognized in both the FDA draft guidance (3) and the EMA guideline (6) that measurement may be more challenging. As independent research has shown, current assessment tools may have limited sensitivity due to ceiling effects and slow rates of progression (7, 8). Co-primary outcomes are not well established at this early stage, and whilst the principle behind the co-primary approach still holds, it has been suggested by regulators and others that application in practice could be achieved by integrated cognitive and functional endpoints, such as the Clinical Dementia Rating – Sum of Boxes (CDR-SB) score (6, 9-11). The CDR is intended to measure “the influence of cognitive loss on the ability to conduct everyday activities” (12). Whilst it has been hypothesized that the CDR may be broken down into ‘cognitive’ and ‘functional’ items (10, 13), the original intent was a unitary underlying construct (14) . Thus, it may be that observed statistical relationships supporting separate cognitive and functional items result from other properties, such as disease severity, or the use of information from the patient versus that from the caregiver-informant.
Studies have consistently reported high internal consistency for the CDR-SB across the AD spectrum, including clinically defined prodromal populations (CDR-GS = 0.5) (10, 15). Inter-rater reliability of the CDR-SB in a prodromal population is unclear, with some studies reporting low inter-rater agreement in populations with earlier non-biomarker confirmed AD dementia (13, 14, 16). Although many clinical trials in early/prodromal AD have used the CDR-SB as a primary endpoint, including studies of crenezumab (NCT02670083, NCT03114657), gantenerumab (NCT03443973, NCT03444870), aducanumab (NCT02484547, NCT02477800), BAN2401 (NCT03887455), and verubecestat (NCT01953601), a comprehensive assessment of the psychometric properties of the CDR-SB in this population is lacking.
To our knowledge, there are no studies describing the test-retest reliability according to gold-standard intra-class correlation coefficient (ICC) for a biomarker-confirmed prodromal population; a critical gap in the evidence needed to support use of the CDR-SB as a primary endpoint in AD clinical trials.
Investigation of the psychometric properties of commonly used outcome measures in the pAD clinical trials population is a critical step in confirming that assessments are fit for purpose, and for identifying potential gaps/areas for further development. Here, we describe traditional psychometrics, including test-retest reliability, of cognitive and functional assessments in a pAD trial population from SCarlet RoAD (NCT01224106; WN25203), a Phase 3, multicenter, randomized, double-blind, placebo-controlled study. In addition to amnestic MCI, subjects recruited to SCarlet RoAD were required to have evidence of amyloid pathology as demonstrated by low levels of Aβ(1–42) in cerebrospinal fluid (CSF). We also explore suitability of the CDR-SB as a single primary endpoint. Such properties should be established for the planned context of use (17, 18), and this paper is intended to do so for multinational, pAD, clinical trials. Furthermore, estimates are dataset dependent and a range of published estimates across different contexts of use may be informative. There are three main points of distinction from prior studies on this topic; some studies have used data from a single country only (15), while others have used observational cohorts (10) and defined AD based on clinical rather than biomarker criteria (10). This paper therefore adds to the existing literature by providing a comprehensive evaluation of the psychometric properties of key clinical outcome assessments in a multinational, clinical trial, biomarker confirmed prodromal AD population. The analysis presented is based on data from the SCarlet RoAD trial that evaluated low dose gantenerumab in patients with prodromal AD.



Data source and patients

The data were gathered as part of a Phase III, multicenter, randomized, double-blind, placebo-controlled, parallel-group, two-year study to evaluate the effect of subcutaneous gantenerumab (RO4909832) on cognition and function in prodromal Alzheimer’s disease (pAD) conducted across 24 countries.
The primary objective of the trial was to evaluate the effect of gantenerumab versus placebo on the change from baseline to week 104 in the Clinical Dementia Rating scale Sum of Boxes (CDR-SB). All measures were translated and linguistically validated as per industry guidelines (19). CDR raters received comprehensive training prior to study start and refresher trainings at regular interval during the study. The information to make each rating was obtained through a semi-structured interview of the subject and a reliable informant. The study required that each subject have a study partner who, in the investigator’s judgment, had frequent and sufficient contact with the subject so as to be able to provide accurate information regarding the subject’s cognitive and functional abilities and who agreed to accompany the subject to clinic visits for scale completion. As far as possible, raters and study partners remained unchanged during the conduct of the study. Other assessments were rated by qualified site staff who were trained and, when necessary, certified to administer the assessments. Whenever possible, the CDR rater did not assess the other cognitive scales.
Inclusion criteria were modelled on International Working Group (IWG) criteria, which redefined AD as a clinicobiological syndrome that can be identified prior to the onset of dementia by an amnestic syndrome of the hippocampal type and supportive evidence from biomarkers (20). Key inclusion criteria were: age 50-85; recent gradual decline in memory (informant); abnormal memory function based on the Free and Cued Selective Reminding Test (FCSRT: free recall <17, or total recall <40, or [free recall <20 and total recall <42]); Mini-Mental Status Exam (MMSE) score ≥24; and Clinical Dementia Rating – global score (CDR-GS) of 0.5 with memory box score of 0.5 or 1; CSF Aβ(1-42) ≤600 ng/L as measured by the central laboratory.

Study Design and Outcome measures

The study consisted of an 8-week screening period, a double-blind treatment phase of 100 weeks, a final assessment at week 104, and follow-up visits at 16 and 52 weeks after the last dose. Participants were recruited from clinical sites, some of which were memory centers. Subjects meeting all eligibility criteria during screening were randomized 1:1:1 to receive either placebo,
105 mg, or 225 mg gantenerumab subcutaneously every four weeks. Assessments at screening included the CDR, the Functional Activities Questionnaire (FAQ), the Alzheimer’s Disease Assessment Scale – Cognition Subscale 11 and 13 item version (ADAS-Cog 11, ADAS-Cog 13 respectively), the MMSE, and the Free and Cued Selective Reminding Test (FCSRT-IR [words]) (Table 1). These assessments (listed in Table 1) were also conducted at baseline, and at 24-weeks interval (at weeks 24, 52, and 76) including final assessment at Week 104. The MMSE, ADAS-Cog, and FCSRT were obtained at additional 12 weeks intervals (at weeks 12, 36, 64, and 88), and the CDR was also obtained at weeks 64 and 88.

Table 1. Clinical Outcome Assessments

ClinRO: Clinician-rated Outcome Assessment; PerfO: Performance-based Outcome Assessment


Clinical Dementia Rating scale

The CDR was originally developed as a staging tool to categorize dementia severity into normal, questionable, mild, moderate or severe. Clinicians rate the severity of symptoms across six domains following a semi-structured interview with the subject and a reliable informant or collateral source (e.g., family member). There are three cognition domains (Memory, Orientation, Judgment & Problem Solving) and three functional domains (Community Affairs, Home & Hobbies, and Personal Care) (12, 21). The response options for each domain describe five degrees of impairment: 0=None; 0.5=Questionable (not present in the Personal care domain); 1=Mild; 2=Moderate; 3=Severe. The CDR-Global score, which determines dementia stage, is rated from 0-3. The Sum of Boxes score is a continuous measure of dementia severity and ranges from 0-18. For completeness, we report the psychometric properties of the CDR-Cognition and CDR-Function domains individually, acknowledging that the CDR was not intended for this purpose and that each domain contains few items (3 items) for traditional analyses (e.g. Chronbach’s α).

Alzheimer Disease Assessment Scale-Cognition (ADAS-Cog)

The ADAS-Cog-11 (22) is a structured scale that evaluates memory (word recall, word recognition), reasoning (following commands), language (naming, comprehension), orientation, ideational praxis (placing letter in envelope) and constructional praxis (copying geometric designs). Ratings of spoken language, language comprehension, word finding difficulty, and ability to remember test instructions are also obtained. The test is scored in terms of errors, with higher scores reflecting poorer performance. Scores can range from 0 (best) to 70 (worst). The 13-item version also includes Delayed Word Recall and Number Cancellation tasks, with scores ranging from 0 to 85.

Mini Mental State Exam (MMSE)

The MMSE consists of a set of standardized questions to evaluate possible cognitive impairment and help stage the severity level of this impairment. The questions target five areas; orientation, short term memory retention, attention, short term recall and language (23). The MMSE is scored as the number of correctly completed items with lower scores indicative of poorer performance and greater cognitive impairment. The total score ranges from 0 (worst) to 30 (best).

Functional Activities Questionnaire (FAQ)

The FAQ is an informant-based assessment in which caregivers rate abilities on 10 activities of daily living (ADLs) (24). The 10 items are scored as Dependent = 3, Requires assistance = 2, Has difficulty but does by self = 1, Normal = 0. The total score ranges from 0 to 30 with higher scores indicating worse functioning. The FAQ has demonstrated good sensitivity and specificity in differentiating MCI from very mild AD, by reflecting very mild functional impairment (25) .

Free and Cued Selective Reminding Test – Immediate Recall (FCSRT-IR)

The FCSRT-IR is a measure of memory under conditions that control attention and cognitive processing in order to obtain an assessment of memory unconfounded by normal age-related changes in cognition (26, 27). The FCSRT-IR used cards with four written words corresponding to a specific category cue, with immediate recall after each card followed by a cued recall using the category cue (28). Abnormal memory function according to FCSRT-IR was defined as a free recall score<17 (sum of free recall items), a total recall score<40 (sum of free recall and cued recall items), or a free recall score<20 and total recall score<42. FCSRT-IR performance has been associated with preclinical and early dementia in several longitudinal epidemiological studies. The CDR, ADAS-Cog, MMSE and FAQ may be viewed as composites, where a total score is based on the sum of item responses and individual items are intended to assess different cognitive and/or functional domains or concepts. Whilst total scores are also derived for FCSRT, items/words are not interpreted as individually meaningful.

Statistical methods

All screening and baseline analyses were conducted on the total sample. Analyses that included Week 52 and/or Week 104 data were conducted in the placebo group only to remove the potential impact of treatment from the evaluation of psychometrics. Patients were included in the analysis if they had completed measures at a given time point.

Test-retest reliability

Test–retest reliability is used to assess the degree to which a measure provides stable scores over time, assuming that the underlying condition of patients has not changed. This aspect of reliability was evaluated by intra-class correlation coefficients (ICCs, Shrout & Fleiss classification Random set 2, 1) (29) between the screening and baseline visits i.e. an interval approximately 8 Weeks (up to 12 Weeks was allowed for FCSRT-IR). Subjects were expected to remain clinically stable over this interval, whereas for longer intervals (baseline to Week 52, Week 52 to Week 104), clinical progression would be expected. Intra-class correlation coefficients that exceed 0.70 are generally assumed to be adequate (30).

Internal consistency

Internal consistency refers to the degree of association between the individual items that comprise a composite measure, and was measured by Cronbach’s α, which generally increases as the inter-correlation amongst test items increases (31). As a general rule, >0.7 is considered an appropriate target for internal consistency (30, 32, 33). Internal consistency was not calculated for the FCSRT-IR outcomes since these are essentially single item constructs.

Construct validity

Construct validity refers to the extent to which a measure adequately assesses an intended concept and may be evaluated by the association to other measures of both similar and different concepts. Relationships between the measures were examined in cross-section, using scores at baseline and change from baseline scores at Week 104. Spearman correlation coefficients (with Fisher’s adjustment) were used to test the correlation between continuous variables. It was expected that objective cognitive measures would be inter-correlated (≥0.4), as would functional measures, but that correlation between cognition and function measures may be lower. The following thresholds were used to assess the strength of the relationship: <0.2: Weak, ≥0.2 to <0.4: Modest, ≥0.4 to <0.6: Moderate, ≥0.6 to 0.8: Strong; ≥0.8: Very strong (34, 35).

Ability to detect change (responsiveness to decline)

As AD is a progressive neurodegenerative disease, decline over time may be used as a way to assess ability to detect change. As an effect size metric, standardized response means (SRM) for the change from baseline in the placebo arm were calculated as SRM = mean change divided by the standard deviation of change, at Week 104. For convenience, we considered values ≥0.2 to <0.5 as low and ≥0.5 to <0.8 as moderate responsiveness (36) (Table 3). Ceiling and floor effects were determined according to the proportion that received the highest and lowest scores at baseline.

Known groups validity

The difference between CDR Global score = 0.5 (Questionable dementia) and CDR Global Score = 1 (Mild dementia) groups were calculated, for each of the variables. This evaluation was conducted at Week 52 (with the exception of FCSRT for which Week 104 was used as the only available time-point), since this maximized the sample size in both CDR-GS = 0.5 and CDR-GS = 1 groups; CDR-GS of 0.5 was an inclusion criterion at screening and a reduced sample size was available at Week 104. Independent samples t-tests were used to assess the statistical significance of the between groups differences. A significant difference between groups is generally considered to reflect reasonable known groups validity.




Seven hundred and ninety-seven subjects received allocated treatment (All Patients), mean age 70.4 years (SD 7.2), mean years of education 12.5 years (SD 4.5), 43.2% male. Two-hundred and sixty-six were randomized to placebo, with 104 completing the Week 104 visit (Placebo Arm), mean age 68.5 years (SD 6.8), mean years of education 12.3 years (SD 4.7), 43.8% male. The clinical characteristics of both populations are summarized in Table 2. Countries with the highest enrollment (>3%) included: the United States (14.3%), Spain (12.6%), Canada (7.3%), the United Kingdom (7.1%), Germany (6.9%), Italy (6.9%), France (6.4%), Australia (6.1%), Mexico (5.9%), Argentina (4.5%), and the Netherlands (4.1%).

Table 2. Clinical characteristics

*For FAQ, ADAS-Cog11, and ADAS-Cog13 n=221; †For FAQ, ADAS-Cog11, and ADAS-Cog13 n=105, FCSRT-IR assessments n=100; ADAS-Cog-11: the Alzheimer’s Disease Assessment Scale – Cognition Subscale 11 item version, ADAS-Cog-13: the Alzheimer’s Disease Assessment Scale – Cognition Subscale 13 item version, CDR-Cognition: Clinical Dementia Rating Cognition domain, CDR-Function: Clinical Dementia Rating Function domain, CDR-SB: Clinical Dementia Rating-Sum of Boxes, FAQ: the Functional Activities Questionnaire, FCSRT-IR: Free and Cued Selective Reminding Test – Immediate recall, MMSE: Mini Mental State Exam, SD: standard deviation.


Floor and Ceiling Effects

At the total score level, floor and ceiling effects were within acceptable ranges (Figure 1). At the item level, for all composite measures (i.e. CDR-SB, FAQ, ADAS-Cog and MMSE), notable ceiling effects (≥20% of patients at ceiling) were evident, showing that a large proportion of the enrolled pAD patient population were unimpaired in several of the items and/or domains assessed by these instruments (Figure 1). In addition, delayed word recall (ADAS-Cog and MMSE recall items) showed evidence of a floor effect.

Figure 1. Floor and Ceiling effects by item and total scores at baseline

All Patients. Dashed line represents threshold for notable floor or ceiling effect.


Test-retest reliability

Test–retest reliability for the total scores was generally >0.7, with the exception of ADAS-Cog11 (0.67), MMSE (0.52) and FCSRT-IR Cued Recall (0.68) (Table 3).

Table 3. Intraclass correlation coefficients, internal consistency, responsiveness and clinical validity

*All p values less than .0001; evaluation was conducted at Week 52 for all measures, with the exception of FCSRT for which Week 104 was used, in order to maximize n; ^Negative value due to scoring direction (lower score = worse cognition); α, standardized Cronbach’s alpha; ADAS-Cog-11: the Alzheimer’s Disease Assessment Scale – Cognition Subscale 11 item version, ADAS-Cog-13: the Alzheimer’s Disease Assessment Scale – Cognition Subscale 13 item version, CDR-Cognition: Clinical Dementia Rating Cognition domain, CDR-Function: Clinical Dementia Rating Function domain, CDR-SB: Clinical Dementia Rating-Sum of Boxes, FAQ: the Functional Activities Questionnaire, FCSRT-IR: Free and Cued Selective Reminding Test – Immediate recall, MMSE: Mini Mental State Exam. Internal consistency not appropriate for FCSRT.


Internal consistency

Internal consistency at screening and baseline was <0.5 for CDR-Cognition and CDR-Function, 0.65 for CDR-SB, 0.63 and 0.68 for ADAS-Cog11, and ADAS-Cog13, respectively; and 0.8 for FAQ. Chronbach’s α tended to increase over the study, exceeding 0.7 for most measures at later timepoints (Table 3). The exception was the MMSE, which had very low internal consistency at baseline, rising to 0.66 at Week 104.

Construct validity

Inter-correlation of scores at baseline and change from baseline to Week 104 are reported in Table 4. CDR-SB was most strongly correlated with FAQ (0.6 at baseline and change from baseline). However, CDR-SB and FAQ were not strongly correlated with the cognitive measures, ADAS-Cog, MMSE or FCSRT (all correlations ≤0.4), with the exception of the correlation between change in CDR-SB and ADAS-Cog13 change at Week 104 (0.5). Both CDR ‘cognition’ and ‘function’ items were equally well correlated with function as measured by FAQ. However, a low degree of correlation was seen between CDR function and ADAS-Cog for the baseline scores only. As expected, the objective cognitive tests, ADAS-Cog, MMSE and FCSRT tended to be more highly correlated with each other.

Table 4. Inter-correlation of scores

ADAS-Cog-11: the Alzheimer’s Disease Assessment Scale – Cognition Subscale 11 item version, ADAS-Cog-13: the Alzheimer’s Disease Assessment Scale – Cognition Subscale 13 item version, CDR-Cognition: Clinical Dementia Rating Cognition domain, CDR-Function: Clinical Dementia Rating Function domain, CDR-SB: Clinical Dementia Rating-Sum of Boxes, FAQ: the Functional Activities Questionnaire, FCSRT-IR: Free and Cued Selective Reminding Test – Immediate recall, MMSE: Mini Mental State Exam.


Responsiveness to decline/Ability to detect change

Sensitivity to change was evaluated as the SRM (Table 3). Of the total scores, ADAS-Cog and FCSRT-IR were the least responsive, whilst CDR-SB, FAQ and MMSE were the most responsive. Importantly, CDR-SB and CDR cognition were free from floor and ceiling effects at Week 104, which may influence SRM. CDR function showed 25.6% at ceiling, FAQ 9.1% at ceiling and 0.3% at floor and MMSE 1.6% at ceiling, suggesting modest impact on SRM.

Known Groups Validity

For the evaluation of known groups validity, large and statistically significant (P<0.0001) differences were evident between subjects with a CDR-Global score of 0.5 versus those with a score of 1, for all measures
(Table 3). CDR-SB, CDR-Function and CDR-Cognition all had Cohen’s effect sizes of greater than 2 (≥small effect size).



The Clinical Dementia Rating was devised as a global dementia-staging tool, taking into account results of clinician testing of cognitive performance and a rating of cognitive behavior in everyday activities, in six major categories of cognitive performance. Impairment is scored as decline from the person’s previous level due to cognitive loss alone, not impairment due to other factors, such as physical impairment, depression, or personality change (21). The CDR is considered to be a face valid measure of “the influence of cognitive loss on the ability to conduct everyday activities” (12). The CDR-SB has gained prominence in recent times as a single primary endpoint for clinical trials in early AD. Results from the crenezumab discontinued Ph III trial show a robust decline in CDR-SB score over 24 months in patients with both prodromal and mild AD, suggesting that, when enriching for “fast progressors” who are impaired on the FCSRT , the CDR-SB is sensitive to decline in early (prodromal-mild) AD (37). Whilst psychometric properties of the CDR-SB have been explored in early AD dementia (10), they have not been explored in a biomarker confirmed pAD population, and test-retest reliability, critical to repeated assessments, has not previously been published to our knowledge.
These data further support the psychometric properties of the CDR-SB, in demonstrating adequate test-retest reliability, a good degree of internal consistency especially over time, and also construct validity in terms of association to instrumental activities of daily living measured by the FAQ. Importantly, these properties are now confirmed in a pAD clinical trial population. Although there was evidence for ceiling effects in individual domains/items at the baseline assessment, this did not have a major impact on sensitivity to decline, and CDR-SB showed a greater degree of responsivity than ADAS-Cog and FCSRT-IR. However, SRM was lower than previously reported over two years in an early AD population derived from ADNI data (0.71 in this report, versus 1.03 in ADNI), which may result from differences in inclusion criteria (10). There were floor and ceiling effects at the item level for all composite measures; this may be an important consideration with respect to the coverage of relevant concepts and for potential sensitivity to disease progression in the early stages of the disease. Floor effects suggest that even at the early stages of disease, delayed free recall assessments may be markedly impaired, again impacting potential sensitivity.
Previous reports have focused on inter and intra-rater reliability. The novel finding for test-retest is of particular value, given the importance of reliability in clinical trial use. Strong test-retest reduces measurement error, which increases the likelihood of detecting true treatment effects. There was an improvement in internal consistency from baseline for all measures over the course of the study. One possible explanation for improved internal consistency may be ’other’ reliability, such as improved intra-rater reliability and reliability of subject and informant report, as all parties become more familiar with the scales and have more data available to inform them. In addition, regression to the mean, or disease progression bias could result in greater homogeneity of scores between items over time. The internal consistency for CDR-SB at Week 52 and 104 (Cronbach’s alpha 0.84 and 0.90, respectively) was similar to that observed in the French REAL.FR cohort study of patients with very mild-to-moderate AD (0.88) (15).
Specific to construct validity of the CDR, Tractenberg et al previously observed that in an AD dementia population, change in ‘cognitive’ items showed a modest correlation with change in MMSE and a low correlation with change in ADL, and ‘functional’ items the opposite pattern (13). Along with results from principal components analyses, this was seen as supportive of separate cognitive and function domains. In the present data, a correlation was observed between both the CDR cognition and function domains and FAQ, for both baseline and change scores (all 0.5). This may be seen as supportive of overall convergent validity with the FAQ (25). Inter-correlation of CDR-SB and FAQ may be driven by measurement of function and some direct overlap in item content and the use of informant report in both assessments. Cedarbaum et al (2013), found correlations with FAQ tended to be higher than with ADAS-Cog11 or ADAS-Cog13 for both cognitive (0.63, 0.55, and 0.59, respectively) and function domains (0.58, 0.42, and 0.45, respectively) in subjects with early or mild AD at baseline in the ADNI study (10). In addition, although their factor analysis showed some support for separate domains, there was overlap for “Judgment and problem solving” and “Community affairs” items in several cases, and a differential pattern based on disease severity was observed. Thus, the CDR may not capture function and cognition as separate domains but still address both, consistent with the original unitary measurement concept (“the influence of cognitive loss on the ability to conduct everyday activities”). Furthermore, low internal consistency reliability of these scores suggests there may be too few items for them to be reliable as separate measures.
For the FAQ and ADAS-Cog, adequate test-retest reliability and a good degree of internal consistency were observed. Both measures demonstrated construct validity in terms of association to related measures (CDR to FAQ and ADAS-Cog to MMSE and FCSRT). This is an important finding for the FAQ, given the lack of validation evidence for the scale beyond discriminative ability (38). Though there was evidence for ceiling effects with individual items at the baseline assessment for both measures, this did not have a major impact on sensitivity to decline for FAQ as the SRM for FAQ (0.73) was greater than for the other measures.
For the MMSE, assessment of psychometric properties was impacted by its use as a screening criterion, with only scores of between 24 and 30 out of 30 possible at screening. This would initially reduce range of scores and variance, decreasing power to obtain high alpha coefficients and impact the ability to adequately assess scale properties at the screening and baseline assessments in particular. Thus, caution is warranted in the interpretation of the results. Overall MMSE did show good sensitivity to decline, comparable to CDR-SB and FAQ (SRM=−0.71), with orientation to time as the single greatest contributory item (SRM=−0.63). This prominence of orientation as sensitive to decline across the different scales is consistent with other data, which has shown orientation to be sensitive to disease progression and important for inclusion in novel composite outcomes (39).
For the FCSRT-IR, adequate test-retest reliability was also observed, though this was also utilized as a screening inclusion criterion. Whilst the measures of free, cued, and total recall were relatively free from ceiling and floor effects, this did not translate to sensitivity to decline and SRMs were −0.2 for cued, −0.5 for free and −0.46 for total recall. Therefore, there may be limited additional value in FCSRT-IR as a longitudinal outcome measure in this patient population.
There are some limitations which could impact the generalizability of these findings. The study population was derived from a clinical trial, in which CDR-Global Score was one of the inclusion criteria, and thus we cannot rule out the possibility that this influenced the reporting of the CDR domains at baseline. Additionally, the CDR-Global score was used to define questionable dementia and mild dementia for the known groups validity analysis of CDR-SB. This may have impacted the analysis, as the CDR-SB and global score may be interrelated. Although industry standards were followed with regards to translation, we did not formally evaluate whether psychometric properties differed by culture or language. Furthermore, this was a biologically homogenous population with low levels of Aβ(1-42) and different results may be found in a more heterogeneous sample. This study population may have been subject to selection bias due to initial study recruitment methods (e.g. site selection), as well as individual interest in participating in a clinical trial. Loss to follow-up will also have impacted the representativeness of longitudinal analyses. Finally, further work is required to establish what constitutes a meaningful change on the CDR-SB in prodromal AD.
In conclusion, CDR-SB showed adequate psychometric properties in the pAD population and its sensitivity to decline over time further support its utility as a clinical trial outcome measure. In addition, its conceptual basis as a measure of the influence of cognitive loss on the ability to conduct everyday activities was supported by the construct validity data. These data reinforce the continued use of CDR-SB as a single primary outcome measure in early AD clinical trials, such as the phase III gantenerumab GRADUATE program and the phase III BAN2401 Clarity AD trial. In addition, validity and reliability of the other assessments, particularly the FAQ, is further supported.
Efforts to develop novel cognitive and functional assessments free from ceiling and floor effects and with greater sensitivity to change in this population should continue. However, given the adequate psychometrics of the CDR-SB, its clinical relevance, and the lack of a clearly established relationship between other objective cognitive endpoints and clinical benefit, at least for patients with early AD, there is good reason to continue to employ the CDR-SB in treatment trials.


Acknowledgements: The authors would like to acknowledge to important contributions of the SCarlet RoAD Investigators, Patients and their Families participating globally in Argentina, Australia, Belgium, Brazil, Canada, Chile, the Czech Republic, Denmark, Finland, France, Germany, Italy, Mexico, the Netherlands, Poland, Portugal, Russia, Spain, South Korea, Sweden, Switzerland, Turkey, the United Kingdom and the United States. This work was supported by F. Hoffmann-La Roche Ltd, Basel, Switzerland.

Conflicts of interest: FM is an employee of Genentech Inc., South San Francisco, USA. MM, PD, PF, DAS, and CJL are employees of F. Hoffmann-La Roche Ltd, Basel, Switzerland. PD owns stock in F. Hoffmann-La Roche Ltd. RD is an employee and owns stock in Genentech Inc. and F. Hoffmann-La Roche Ltd. MB has received grants from Merck & Co., Inc. related to the submitted work (paid to the institution); she has received grants from Araclon, Biogen Research Ltd, Bioberica, Grifols, Lilly S.A, Merck Sharp & Dohme, Nutricia SRL, Oryzon Genomics, Piramal Imaging Ltd, Schwabe Farma Iberica SLU and Merck & Co, Inc. within the last 36 months outside the submitted work (paid to the institution); she has served as a consultant or provided scientific advisory board services and/or given lectures for Roche, Araclon, Bioberica, Grifols, Kyowa Hakko Kirin, Laboratorios Servier, Lilly, S.A., Merck Sharp & Dohme, Nutricia SRL, Schwabe Farma Iberica SLU.

Ethical Standards: Institutional Review Boards (IRBs) approved the SCarlet RoAD study, and all participants gave informed consent before participating.

Data sharing statement: Qualified researchers may request access to individual patient-level data through the clinical study data request platform: Further details on Roche’s criteria for eligible studies are available here: For further details on Roche’s Global Policy on the Sharing of Clinical Information and how to request access to related clinical study documents, see here:

Open Access: This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits use, duplication, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.



1. Dubois B, Feldman HH, Jacova C, Hampel H, Molinuevo JL, Blennow K, et al. Advancing research diagnostic criteria for Alzheimer’s disease: the IWG-2 criteria. Lancet Neurol. 2014;13(6):614-29.
2. Albert MS, DeKosky ST, Dickson D, Dubois B, Feldman HH, Fox NC, et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):270-9.
3. FDA. Early Alzheimer’s Disease: Developing Drugs for Treatment Guidance for Industry, Draft Guidance. 2018.
4. EMA. Discussion paper on the clinical investigation of medicines for the treatment of Alzheimer´s disease and other dementias. In: CHMP, editor. London: European Medicines Agency; 2014. p. 33.
5. FDA. Multiple Endpoints in Clinical Trials. In: (CDER) CfDEaR, (CBER) CfBEaR, editors. Rockville, MD2017.
6. EMA. Guideline on the clinical investigation of medicines for the treatment of Alzheimer’s disease In: CHMP, editor. London2018.
7. Jekel K, Damian M, Wattmo C, Hausner L, Bullock R, Connelly PJ, et al. Mild cognitive impairment and deficits in instrumental activities of daily living: a systematic review. Alzheimers Res Ther. 2015;7(1):17.
8. Wang J, Logovinsky V, Hendrix SB, Stanworth SH, Perdomo C, Xu L, et al. ADCOMS: a composite clinical outcome for prodromal Alzheimer’s disease trials. J Neurol Neurosurg Psychiatry. 2016.
9. FDA. Guidance for Industry Alzheimer’s Disease: Developing Drugs for the Treatment of Early Stage Disease DRAFT GUIDANCE. In: CDER, editor. 2013.
10. Cedarbaum JM, Jaros M, Hernandez C, Coley N, Andrieu S, Grundman M, et al. Rationale for use of the Clinical Dementia Rating Sum of Boxes as a primary outcome measure for Alzheimer’s disease clinical trials. Alzheimers Dement. 2013;9(1 Suppl):S45-55.
11. Kozauer N, Katz R. Regulation of drugs for early Alzheimer’s disease. N Engl J Med. 2013;369(3):288.
12. Morris JC. The Clinical Dementia Rating (CDR): current version and scoring rules. Neurology. 1993;43(11):2412-4.
13. Tractenberg RE, Weiner MF, Cummings JL, Patterson MB, Thal LJ. Independence of changes in behavior from cognition and function in community-dwelling persons with Alzheimer’s disease: a factor analytic approach. J Neuropsychiatry Clin Neurosci. 2005;17(1):51-60.
14. Morris JC, Ernesto C, Schafer K, Coats M, Leon S, Sano M, et al. Clinical dementia rating training and reliability in multicenter studies: the Alzheimer’s Disease Cooperative Study experience. Neurology. 1997;48(6):1508-10.
15. Coley N, Andrieu S, Jaros M, Weiner M, Cedarbaum J, Vellas B. Suitability of the Clinical Dementia Rating-Sum of Boxes as a single primary endpoint for Alzheimer’s disease trials. Alzheimer’s & dementia : the journal of the Alzheimer’s Association. 2011;7(6):602-10.e2.
16. Rockwood K, Strang D, MacKnight C, Downer R, Morris JC. Interrater reliability of the Clinical Dementia Rating in a multicenter trial. J Am Geriatr Soc. 2000;48(5):558-9.
17. FDA. Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims:Guidance for Industry. 2009.
18. Powers JH, 3rd, Patrick DL, Walton MK, Marquis P, Cano S, Hobart J, et al. Clinician-Reported Outcome Assessments of Treatment Benefit: Report of the ISPOR Clinical Outcome Assessment Emerging Good Practices Task Force. Value Health. 2017;20(1):2-14.
19. Wild D, Grove A, Martin M, Eremenco S, McElroy S, Verjee-Lorenz A, et al. Principles of good practice for the translation and cultural adaptation process for patient-reported outcomes (PRO) measures: report of the ISPOR task force for translation and cultural adaptation. Value in Health. 2005;8(2):94-104.
20. Dubois B, Feldman HH, Jacova C, Dekosky ST, Barberger-Gateau P, Cummings J, et al. Research criteria for the diagnosis of Alzheimer’s disease: revising the NINCDS-ADRDA criteria. Lancet Neurol. 2007;6(8):734-46.
21. Berg L. Clinical Dementia Rating (CDR). Psychopharmacol Bull. 1988;24(4):637-9.
22. Rosen WG, Mohs RC, Davis KL. A new rating scale for Alzheimer’s disease. Am J Psychiatry. 1984;141(11):1356-64.
23. Folstein MF, Folstein SE, McHugh PR. «Mini-mental state». A practical method for grading the cognitive state of patients for the clinician. Journal of psychiatric research. 1975;12(3):189-98.
24. Vellas B, Andrieu S, Sampaio C, Coley N, Wilcock G, European Task Force G. Endpoints for trials in Alzheimer’s disease: a European task force consensus. Lancet Neurol. 2008;7(5):436-50.
25. Pfeffer RI, Kurosaki TT, Harrah CH, Jr., Chance JM, Filos S. Measurement of functional activities in older adults in the community. J Gerontol. 1982;37(3):323-9.
26. Grober E, Buschke H. Genuine memory deficits in dementia. Developmental Neuropsychology. 1987;3:13-36.
27. Grober E, Hall C, McGinn M, Nicholls T, Stanford S, Ehrlich A, et al. Neuropsychological strategies for detecting early dementia. J Int Neuropsychol Soc. 2008;14(1):130-42.
28. Grober E, Sanders AE, Hall C, Lipton RB. Free and cued selective reminding identifies very mild dementia in primary care. Alzheimer Disease and Associated Disorders. 2010;24(3):284.
29. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull. 1979;86(2):420-8.
30. Nunnally JC, Bernstein IH. Psychometric Theory. 3rd. ed. ed. New York: McGraw-Hill; 1994.
31. Cronbach LJ. Coefficient alpha and the internal structure of tests. Psychometrika. 1951;16(3):297-334.
32. DeVon HA, Block ME, Moyle-Wright P, Ernst DM, Hayden SJ, Lazzara DJ, et al. A psychometric toolbox for testing validity and reliability. Journal of Nursing scholarship. 2007;39(2):155-64.
33. DeVellis RF. Scale development: Theory and applications: Sage publications; 2016.
34. Swinscow T. Correlation and regression. Statistics at square one. 9 ed1997.
35. Bartz AE. Basic Statistical Concepts. 4 ed. Upper Saddle River, NJ: Merrill; 1999.
36. Middel B, Van Sonderen E. Statistical significant change versus relevant or important change in (quasi) experimental design: some conceptual and methodological problems in estimating magnitude of intervention-related change in health services research. International Journal of Integrated Care. 2002;2.
37. Sink K, Djakovic S., Smith JW, Hu N, Mackey H, Ostrowitzki S, et al. FCSRT inclusion criteria support recruitment of a population with early Alzheimer’s disease likely to progress over 24 months: results from the CREAD trial. Clinical Trials in Alzheimer’s Disease; San Diego, USA2019.
38. Kaur N, Belchior P, Gelinas I, Bier N. Critical appraisal of questionnaires to assess functional impairment in individuals with mild cognitive impairment. Int Psychogeriatr. 2016;28(9):1425-39.
39. Vellas B, Bateman R, Blennow K, Frisoni G, Johnson K, Katz R, et al. Endpoints for Pre-Dementia AD Trials: A Report from the EU/US/CTAD Task Force. J Prev Alzheimers Dis. 2015;2(2):128-35.


H.-T. Li1, S.-X. Yuan1, J.-S. Wu2, X.-Z. Zhang3, Y. Liu4, X. Sun1 and For the Alzheimer’s Disease Neuroimaging Initiative†

1. State Key Laboratory of Bioelectronics, School of Biological Science and Medical Engineering, Southeast University, Nanjing, P. R. China; 2. School of Geography and Biological Information, Nanjing University of Posts and Telecommunications, Nanjing, P.R. China; 3. Department of Biostatistics, Rollins School of Public Health, Emory University, Atlanta, USA; 4. The First Affiliated Hospital of Nanjing Medical University, Jiangsu Province Hospital, Nanjing, P.R. China

Corresponding Author: Xiao Sun, State Key Laboratory of Bioelectronics, School of Biological Science and Medical Engineering, Southeast University, Nanjing 210096, P. R. China,

J Prev Alz Dis 2020;
Published online November 23, 2020,



Background: Alzheimer’s Disease (AD) is a neurodegenerative brain disease in the elderly. Recent studies have revealed the heterogeneous nature of AD. Mild Cognitive Impairment (MCI) is the prodromal stage of AD.
Objectives: In this study, we identified subtypes of MCI based on genetic polymorphism and gene expression.
Methods: We utilized the two types of omics data, namely genetic polymorphism and gene expression profiling, derived from 125 MCI patients’ peripheral blood samples from the ADNI-1 dataset. Similarity network fusion (SNF) algorithm was implemented to cluster MCI patient subtypes. And 185 MCI patients in ADNI-2 were utilized to evaluate the effectiveness of this method. Two MCI subtypes were identified by implementing the SNF algorithm.
Results: We used Kaplan-Meier analysis and log-rank testing for the conversion from MCI to AD between two subtypes, and p-value is 4.58×10-3. In addition, we compared patients among two MCI subtypes by the following factors: the changes in Alzheimer’s Disease cognitive scales and MRI image; significantly enriched pathways based on differentially expressed genes. This study proved that MCI is a heterogeneous disease by concluding that AD development in two MCI subtypes is significantly different.
Conclusions: MCI patients with different molecular characteristics have different risks converting to AD. In addition to evaluating statistics, genetic polymorphism and gene expression profiling from MCI patients’ peripheral blood are non-invasiveness and cost-effectiveness markers to identify MCI subtypes for clinical application.

Key words: Alzheimer’s disease, mild cognitive impairment, molecular subtyping, similarity network fusion.



Alzheimer’s disease (AD) is a chronic degenerative brain disease and the most common cause of dementia in the elderly. According to statistics, about 10% of people older than 65 suffer from AD (1). Due to the lack of understanding of its causes, effective drugs or treatments of AD is yet not invented.
AD is a complex and heterogeneous disease caused by multiple different genetic factors (2). Recently, more and more studies, such as clinicopathologic (3), atrophy patterns on magnetic resonance imaging (MRI) (4) and amyloid-β fibril polymorphism on solid-state nuclear magnetic resonance (ssNMR) (5), have supported the hypothesis on the existence of distinctive AD molecular subtypes. For example, the rapidly progressive form in which neurodegeneration occurs within months and a typical prolonged-duration form are two AD clinical subtypes that been well recognized. Recently, some researchers have found that different AD clinical subtypes were correlated with fibril formations subtypes by researching on 37 brain samples from 18 deceased Alzheimer’s patients obtained by using ssNMR (5). Lately, another research assigned 4,050 people with late-onset AD into six subgroups according to their cognitive functioning at the time of diagnosis and then utilized genetic data to find the biological differences across these subgroups (6). This study supported the biological coherence of cognitively defined subgroups. With more in-depth studies of Alzheimer’s subtypes, new diagnostic criteria, and treatment of AD that target specific kinds of AD subtypes can be expected.
Mild Cognitive Impairment (MCI) is known as the prodromal stage of AD. MCI is a neurological disorder in which an elderly has mild but measurable changes in cognition. It is worth mentioning that not all people with MCI will develop AD. Studies suggest that MCI patients progress to AD at a rate of approximately 10% every year (7). Early identification of high-risk subtypes MCI patients appears to be significant and may enable a more effective, preventive treatment, thereby increasing the possibility of delaying even avoiding conversion from MCI to AD.
For the above reasons, we believe that MCI is a heterogeneous disease. Identifying the subtypes of MCI is critical for implementing precision medicine approaches and for ultimately developing successful subtype-specific drugs for AD. And classifying MCI patients into meaningful subtypes may provide better targeted treatment to delaying or preventing the conversion from MCI to AD. Genetic factors play an important role in MCI and AD (2). However, to our knowledge, the molecular subtyping of MCI based on integrative multi-omic data was not taken into consideration among current studies. Therefore, in this study, we took advantage of the two types of omics data, including genetic polymorphism and gene expression, derived from 125 MCI patients’ peripheral blood samples from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) to identify the MCI patient subtypes (8). We used the Similarity Network Fusion (SNF) algorithm to cluster the two types of omics data to determine the subtypes of 125 MCI patients (9). For testing the effectiveness and reliability of the SNF algorithm, 185 MCI patients from ADNI-2 were identified the subtype by the label propagation algorithm (9, 10). The flow chart of our research is illustrated in Figure 1. To prove the biological and clinical significance of subtyping patients based on our method, these different subtypes were compared by the following factors: the time difference of the conversion from MCI to AD; cognitive scales and MRI image; significantly enriched pathways based on differentially expressed genes separately.

Figure 1. Flow chart of our research. (a) The Similarity Network Fusion (SNF) algorithm is used to integrate SNP and gene expression data for subtype identification of MCI patients; (b) The label propagation algorithm is applied to predict the subtype of any new patient from ADNI-GO/2 for testing the effectiveness and reliability of the SNF algorithm



Genomic data and imaging data

Data used in this study were downloaded from ADNI. ADNI was a multi-site study proposed by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB) and the Food and Drug Administration (FDA) in 2003. This organization is holding an ongoing, longitudinal, multicenter study. Its primary goal is to test whether clinical, imaging, genetic, and biochemical biomarkers are effective in clinical trials of MCI and AD. The first stage of ADNI, as known as ADNI-1, was completed in 2010 (8). More up-to-date and detail information is available at
In this article, we used combinations of multi-omics data (genetic polymorphism and gene expression) from the ADNI-1 and ADNI-GO/2 study to identify the MCI molecular subtypes and to predict the conversion from MCI to AD. 125 MCI patients’ SNP and gene expression data were downloaded from ADNI-1 for identification MCI subtypes. Meanwhile, 185 MCI patients were downloaded from ADNI-GO/2 as an independent verification dataset for predicting the subtype of any new patients. The information on MCI patients is listed in Supplementary Excel file 1. Both profiling were collected from peripheral blood samples. ADNI-1 and ADNI-GO/2 subjects were genotyped using the Human 610-Quad BeadChip and Illumina Human Omni Express BeadChip, respectively. Only SNP markers were analyzed for subsequent analysis. Quality control steps were performed on genetic polymorphism using the software package named PLINK (11), release v1.90b.5. SNPs with missing rate >0.05, minor allele frequency < 0.05, and Hardy–Weinberg equilibrium P < 10−3 were excluded from the genetic polymorphism set. Then the SNP data was applied by using the IMPUTE2 program for imputing the missing data with NCBI 1000 Genomes build 37 (UCSC hg19) as the reference panel (12). The Affymetrix Human Genome U219 Array was carried out for expression profiling, which contains 530,467 probes. Thenceforth, we used an R package named RMA for the normalization of gene expression microarray data (13). Finally, 49,293 transcripts were kept in this study.
There are various clinical/cognitive assessment scores from ADNI that are useful to compare clinical information between two subtypes of patients, including Mini Mental State Examination (MMSE), Clinical Dementia Rating Sum of Boxes (CDR-SB) and Activities of Daily Living Score (from the Functional Activities Questionnaire, FAQ). In addition, we downloaded T1 weighted MRI images in NIFTI format from 125 MCI patients’ baseline, 24-month follow-up data set in ADNI, and structural MRI scan applied inversion recovery-fast spoiled gradient recalled (IR-SPGR) for researching two clusters of MCI patients’ differences in areas of brain atrophy. VBM analyses were performed using the SPM12 toolkit (Statistical Parametric Mapping software, spm12) running under MATLAB 2013a (14).

MCI subtype identification based on similarity network fusion

We applied the similarity network fusion (SNF) algorithm to cluster the MCI patient subtypes (9). SNF is an integrated characterization of genomic profiling at multiple levels for subtype identification. The advantage of using SNF is that it is based on complementarity in multiple genomic data types. First, the SNF algorism uses a similarity measure to constructs a patient-by-patient similarity network for each genomic data type. The nodes of the network for each data type represent patients and the weighted edges are equivalent to pairwise sample similarities. Next, the network fusion step updates every network using a nonlinear method named message-passing theory. Each iteration makes these networks more similar to each other. After many iterations, multiple networks converge to a fusion network. Finally, the fusion network is clustered into several subtypes based on spectral clustering methods. The illustrative example of SNF steps are shown in Figure S1. Some patients (002_S_0729, 010_S_0161 and 011_S_1282 from cluster-1; while 005_S_0546, 027_S_1045 and 037_S_0150 from cluster-2) were used as examples to explain the clustering process of the SNF method (Figure S1 (d)).
More formally speaking, given n MCI patients and M omics (SNP and expression data in this study), the sample×sample similarity graph G=(N, W) is constructed, where node set N represents the samples x1,x2,…,xn and the edges weight W(i, j) represents the weight between xi and xj. W is defined by:

where d(m) (xi,xj) is the Euclidean distance between sample xi,xj for the m-th omic. α is a hyperparameter and α=0.8 in this study. ε is expressed as below:

where Ki is the number of neighbours of xi and Ki=30 in this study, mean (ε(xi, Ki)) is the average distance between xi and each of its neighbors. ε is introduced to eliminate the scaling problem.
A transition probability matrix is constructed between all MCI patients initially by:

Meanwhile, a transition probability matrix between nearest neighbors is defined by:

where Ni represent a set of i’s k nearest neighbors in matrices with measurements from the m-th omic.
Then, the matrix P is updated based on message-passing theory iteratively between the k nearest neighbors by formula:

where Pq(m) is the matrix for omic m at iteration q. The iterative process means that the connection information of different networks is exchanged to achieve the final uniform network.
After completing the network fusion, low-weight edges in each network disappear, and high-weight edges are retained. SNF reduces the noise among these steps, which makes this method robust to noise and the data heterogeneity. Finally, based on spectral clustering methods, namely minimize RatioCut, the fusion network is clustered into several subgroups. Such subgroups are considered as our resulting subtypes. The details of SNF reference (9).

Any new MCI patient’ subtype prediction based on label propagation

We adopted label propagation algorithm which is a simple iterative semi-supervised learning algorithm based on network structure to identify the subtype of the new MCI patient (9, 10). Assume n patients have been determined into y subtypes by the SNF method with a fused network F. To predict the subtype of a new patient, a similarity matrix F=[F s;s’ 1] is constructed, where s is the similarities vector calculated by SNF. Define a (n+1)×(n+1) probabilistic transition matrix T:

where Tij is the probability of jumping from node j to i. Also we define a (n+1)×y label matrix Y, whose i-th row representing the label probabilities of node yi. We iterate the propagation process as follows:
Repeat the following steps:

This process will converge usually in 1000 iterations. And we can predict the subtype of the new patient given by converged Y.


Clustering of MCI patients

We downloaded 138 MCI patients’ gene expression profiling and 361 MCI patients’ genetic polymorphism data from the ADNI-1 dataset. The number of MCI patients with both genetic polymorphism and gene expression was 125. Hence, we used these MCI patients in this article for integrating the two types of omics data to identify MCI patient subtypes. Moreover, 276 MCI patients’ SNP data and 302 gene expression profiling were downloaded from the ADNI-GO/2 dataset. 185 MCI patients who have SNP data, gene expression data, and clinical follow-up data for greater than 36 months were selected as an independent verification set to evaluate the effectiveness of this method. Table S1 shows the characteristics of the MCI patients included in this study.
The subtypes of MCI patients in the ADNI-1 dataset were identification based on SNF method (9). In the beginning, quality control steps were performed on genetic polymorphism using the software package named PLINK (11) and gene expression profiling using an R package named RMA (13) as described in method. Then, we utilized SNF to cluster MCI patients using both SNP and gene expression profiling after quality control. SNFtool R package (v2.3.0) was applied with the parameters K = 30, alpha = 0.8, T = 20 (9). Spectral clustering implemented in the SNFtool package was run on the SNF fused similarity matrix to obtain the groups that each corresponding to k=2 to 5.
After executing the SNF algorithm, we chose the best number of clusters according to two main approaches of the spectral clustering method. One is the connectivity of the network, and the other is to make use of the structure of eigenvectors of the Laplacian L (9). However, the optimal number of clusters based on the connectivity of the network is 2, the best number decided by the other approaches is 3. Therefore, we used the highest average silhouette score as an assistance approach to decide the optimal number of clusters. The silhouette score represents the coherence of clusters to evaluate whether patients are more similar within subtypes. In other words, the silhouette score condenses the cluster quality for each patient’s omics data into a single score that ranges from 1.0 to -1.0. Hence, we had identified two subtypes. The number of patients in cluster-1 is 61, and cluster-2 has 64 patients.
To prove the biological and clinical significance of subtyping patients based on the SNF method, we applied the label propagation algorithm to assign new patients to subtypes in the ADNI-2 datasets (9, 10). Genotype data of MCI patients from the ADNI-GO/2 dataset were downloaded, quality controlled, imputed to the Illumina 610Quad platform and combined. Genotype imputation was conducted to estimate unobserved genotypes. Impute2 software was used with NCBI 1000 Genomes build 37 (UCSC hg19) as the reference panel (12). After executing the label propagation algorithm to 185 patients in ADNI-GO/2, 60 MCI patients were identified in cluster-1, while 125 patients were identified in cluster-2. The detail information on the subtypes of MCI patients in the ADNI-1 and ADNI-GO/2 dataset is listed in Supplementary Excel file 1.

Two MCI subtypes supported by clinical manifestations

We first examined the time difference of the conversion from MCI to AD between two subtypes of patients. Because the exact date of conversion to AD was not known, we used the midpoint between the last follow-up without an AD diagnosis and the first follow-up with an AD diagnosis for analyses. Subjects who did not convert were censored at the time of their last interview. We performed a Kaplan-Meier analysis on MCI of these two clusters. As is shown in Figure 2(a), P-value is 4.58×10-3, demonstrating a significantly different amount of time is consumed for MCI-to-AD conversion between two clusters. Patients that develop the disease more rapidly (red solid line) were cluster-1 MCI patients, and the others (blue dashed line) were cluster-2 MCI patients.

Figure 2. The Kaplan-Meier plot analysis on MCI of the two clusters of clinical data. X axis represents time past after MCI patients participating the study, while Y axis represents estimated percentages of stable MCI patients. The red solid line represents cluster-1 MCI patients in ADNI-1 (a) and ADNI-GO/2 (b), while the blue dashed line represents cluster-2 MCI patients in ADNI-1 (a) and ADNI-GO/2 (b)


We also considered the changes in Alzheimer’s Disease cognitive scales. Cognitive function status was measured by the Mini-Mental State Examination (MMSE) (rating 0–30, higher scores indicate good cognitive function), the Clinical Dementia Rating Sum of Boxes (CDR-SB) (rating 0–25, with higher scores representing greater impairmen) and the Functional Assessment Questionnaire (FAQ) (range 0–30, with higher scores representing greater impairment) in two years for two MCI subtypes of patients (8). As is shown in Figure 3(a), cognitive decline in cluster-1 MCI patients tends to be more remarkable than that of cluster-2 over 24 months.
To test the effectiveness and reliability of the SNF algorithm through its application on ADNI-GO/2 patients, we examined the time difference of the conversion from MCI to AD between two subtypes of all MCI patients. As is shown in Figure 2(b), this gives a log-rank P-value of 2.26×10-4. And three AD cognitive scales were also displayed in two years for two MCI subtypes of patients in the ADNI-2 dataset, which is shown in Figure 3(b). The scores change trends of all three cognitive scales in ADNI-GO/2 are similar to the ADNI-1 dataset. Thus, it proved the validity of the SNF method for subtyping MCI patients based on integrative genetic polymorphism and gene expression. Meanwhile, the cluster-1 subtypes having the worse prognosis than the cluster-2 subtypes.

Figure 3. Changes in AD cognitive scales (MMSE, CDR, FAQ) in two years for two MCI subtypes in ADNI-1 (a) and ADNI-GO/2 (b). X axis represents time past after MCI patients participating the study, while Y axis represents Alzheimer’s Disease cognitive scales score. Cognitive decline in cluster-1 MCI patients (red) is tend to be more remarkable than that of cluster-2 (blue) over 24 months

Two MCI subtypes supported by MRI image

We further analyzed the MRI images to illustrate the difference between two clusters of MCI patients’ ADNI baseline and 24-month follow-up MRI dataset using voxel-based morphometry (VBM) analyses in atrophy areas (15). VBM analysis has been developed for characterizing differences in the local composition of brain tissue using MRI and is not restricted to previously called region-of-interest measurements.
Firstly, we normalized images with the voxel sizes of 1.5×1.5×1.5mm3 because it could preserve the total amount of signal in the images. After normalization, T1-weighted images were segmented into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) using default option parameters on SPM12’s unified segmentation procedure. After that, we transformed patients’ images to the Montreal Neurological Institute (MNI) co-ordinate space using a template. Cognitive impairment is related to the MRI of GM decline on longitudinal analysis. Hence, on GM images, the spatial normalization approach was performed with the diffeomorphic anatomical registration using the exponentiated Lie algebra (DARTEL) algorithm (16). Subsequently, the images were smoothed with a 10-mm full-width at half-maximum isotropic Gaussian smoothing kernel. The results of GM images were analyzed with the two-sample t-test. For voxels in GM probability maps between baseline and 24 months, we selected those voxels with P<0.05 corrected by False Discovery Rate (FDR), and only regions of more than 100 contiguous selected voxels were considered in the analysis. To analysis the result of GM atrophy origins, we utilized the predefined anatomical masks obtained from an extension to the SPM package – XjView toolbox ( and the automated anatomical labeling (AAL, (17).
Based on the current official anatomical nomenclature proposed by Guilherme et al., the brain structure was divided into six lobes: frontal, parietal, occipital, temporal, insular, and limbic (18). The atrophic number of significantly different voxel regions is shown in Table 1. The result of the above steps was characterized by XjView.The comparison of cluster-1 (a) and cluster-2 (b) MCI patients’ regions of gray matter atrophy between baseline and 24-month follow-up MRI images are shown in Figure 4(a,b).

Figure 4. . Display of voxels with significantly brain areas of decreased gray matter intensity in each cluster. Images are 3D render view of (a) cluster-1 and (b) cluster-2 in sagittal, coronal and transversal. And paired images are MCI patients’ baseline MRI images compared to those of 24-month follow-up using VBM analyses. Colored voxels show regions that were significant in the analyses with p<0.05 corrected by FDR, and regions threshold of 100 contiguous voxels. The color brighter (yellow) indicates the more significant area of brain atrophic voxels in 24 month. (c) The atrophic size of significantly different voxel bunches within six lobes in 24 months

Figure 4(c) reveals that the atrophic size of significantly different voxel bunches of cluster-1 MCI patients in 24 months are apparently larger than that of cluster-2 MCI patients. In addition, the proportion of the atrophic voxels in six lobes accounted for 46.26% of total number of brain atrophic voxels in cluster-1, while in cluster-2 this ratio is 25.00%. This result indicates that not only was the atrophy of voxels in cluster-1 patients significantly more than that of cluster-2 patients, but also the location of atrophy was also concentrated in the functional areas of the brain. Therefore, by comparing the MRI images of cluster-1 and cluster-2 MCI patients collected from two-year data, one can see that AD development of cluster-1 patient is faster than that of cluster-2. Hence, this proves the usefulness of the subtype classification in clinical.

Two MCI subtypes supported by gene annotation

Subsequently, differential expressions of mRNA of MCI cluster-1, cluster-2 compared with the cognitively normal samples were each computed using R package named limma (19). Adjust-P value< 0.05 served as the screening conditions for the significant differences. The significantly different expression gene-set of cluster-1 had 3156 genes, while that of cluster-2 had 178 genes. We applied the functional annotation tool of “Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment” and “Gene ontology (GO)” in Enrichr, which was an integrative web-based software application that included many new gene-set libraries for gene and sequence annotations (20). Enrichr provided an adjustment P-value and combined score to annotate the biological significance of differentially expressed genes. An adjust p-value, 0.1, was chosen as significant thresholds upon filtering the pathway data. Because of too many biological processes in GO, a threshold of the combined score was considered. Common GO analyses were performed with a cut-off of combined score 20 and adjust p-value 0.1. The definition of the combined score in EnrichR is to integrate both p-value and z-score with the formula c = z-score•log(p-value), where c is the combined score, represented by p-value computed using the Fisher exact test, and z-score computed by assessing the deviation from the expected rank. The significant pathways and biological processes of differentially expressed genes between cluster-1, cluster-2, and control are shown in Figure 5. The full list of KEGG pathways and GO enrichment analysis information is in Supplementary Excel file 2.

Figure 5. The enriched significant KEGG pathways and GO biological processes bubble plot of differentially expressed genes (DEG) with FDR<0.05. (a) cluster-1 DEG KEGG enrichment, (b) cluster-2 DEG KEGG enrichment, (c) cluster-1 DEG GO enrichment, (d) cluster-2 DEG GO enrichment. The size of the dots represents the count of DEG in the corresponding pathways or GO terms. Y axis represents the enrichment pathways and biological processes. (a, b) X axis represents the opposite of the logarithm of p-value for each pathway, and (c, d) X axis represents the combine score which is defined by Enrichr for each biological process


The most remarkable pathways of cluster-1 are the following: RNA degradation, Amino sugar and nucleotide sugar metabolism, and RNA transport. These pathways are related to a wild range of biological processes. Meanwhile, the significant pathways of cluster-1 were predominated by immune system-related biological fields, such as B cell receptor signaling, TNF signaling and some microbial infection pathway (Epstein-Barr virus infection, Shigellosis and Legionellosis). More research results showed that inflammation is closely related to AD. In the brain, immune system cells called microglia is activated by the presence of toxic amyloid-β and tau proteins (21). Microglia tries to get rid of the remnants of inflammasomes in tiny clumps. However, these remnants continued to spread new amyloid-β clusters as well as aggravating the state of AD. Notably, Epstein-Barr virus infection is also one of the significant pathways. Epstein-Barr virus is known to be the one of herpes viruses. Recent research indicated that herpes viruses abundance was significantly associated with modulators of APP metabolism which revealed viral regulation of AD risk by multiscale networks (22). And insulin signaling pathways have a close relationship with AD. AD has been considered as a metabolic dysfunction disease associated with impaired insulin signaling (23). Proteolytic processes contribute to the amyloid cascade, and proteolysis of tau may be critical to neurofibrillary degeneration, which correlates with AD (24).
The significant biological processes in GO enrichment of cluster-1 are the regulation of transcription and protein catabolic process. These biological processes are closely correlated. Regulation of the transcription, DNA-templated is any process that modulates the frequency, rate or extent of cellular DNA-templated transcription. Regulation of transcription of AD genes might be an important player in the neurodegenerative process. For example, the APP gene is ubiquitously expressed in a variety of tissues, with the highest expression shown in neuronal cells. The abnormally expressed APP will lead to an increased amount and deposition of the amyloid β peptide (Aβ) in the brain triggering AD-related neuronal degeneration (25). Mutant forms of ubiquitin may inhibit proteolysis within neurons, making these cells susceptible to inclusion formation. Therefore, some researchers hold the contention that neurodegenerative diseases collectively referred to as “ubiquitin protein catabolic disorders”. Especially, similar to the KEGG analysis of cluster-1 MCI patients, the significant biological processes are also associated with the immune system. For instance, some biological processes are related to neutrophil and macroautophagy (26). Neutrophils are key components for early innate immunity. Blood samples from AD patients with dementia revealed that the neutrophil hyperactivation was associated with increased reactive oxygen species production as well as the levels of intravascular neutrophil extravascular traps. Moreover, neutrophil phenotype may have a close relationship with the rate of cognitive decline (26).
The cluster-2 significantly enriched pathways mainly consisted of neuronal signaling-related pathways, such as endocytosis and synaptic vesicle cycle. For instance, the synaptic vesicle cycle plays an important role in the biological process of exocytosis and endocytosis. It facilitates a series of events achieving chemical neurotransmission between functionally related neurons. Some study results demonstrated that considerable changes in the expression and functions of presynaptic proteins attributed in parts to direct effects of amyloid-β production and toxicity on the synaptic vesicle cycle (27). In addition, endocytosis is critical for the normal processing of APP, which is central to AD pathogenesis (28).
The most remarkable GO biological process of cluster-2 is the regulation of vascular associated smooth muscle cell migration. The degenerated smooth muscle cells express increased amounts of amyloid β-precursor protein deposition in the medial layer of the cerebral vessel wall and produce Aβ peptide (29). And the low-density lipoprotein particle receptor catabolic process is another important biological process in cluster-2. This biological process results in the breakdown of a low-density lipoprotein particle receptor molecule, a macromolecule that undergoes combination with a neurotransmitter to initiate a change in cell function. The disorder in this biological process could impair the neurotransmitter-triggered signal transduction appearing in AD.



AD is a neurodegenerative brain disease that yet has no available effective medications or supplemental treatment. Studies have shown that AD is a heterogeneous disease. In this article, we integrated two types of omics data (genetic polymorphism and gene expression profiling) of MCI patients to identify subtypes with biological and clinical significance by the SNF method. We performed SNF, the integrative clustering of multiple genomic data algorithms, to cluster MCI patients. Experimental studies were conducted on subtypes of MCI patients, and we showed that multi-omics data define subtypes characterized by biological and clinical significance.
We utilized the SNF method to identify MCI patient subtypes based on multi-omics characteristics (9). SNF has been used to cluster subtype of specific cancer patients, and satisfactory results have been achieved. After executing the SNF algorithm, we identified two MCI subtypes. By comparing clinical information between two subtypes of patients, we considered the changes in two years on AD cognitive scales (MMSE, CDR, and FAQ) and MRI images in atrophy areas based on VBM. We found that the molecular subtypes of MCI are remarkably different in clinical information. It is necessary to lay the foundation for the precision treatment of MCI patients.
To study the difference in the disease mechanism of cluster-1 and cluster-2, differential expressions of MCI cluster-1, cluster-2 mRNA compared with the cognitively normal samples were computed correspondingly. And the differential expression genes in cluster-1 are significantly more than that of cluster-2. We conjecture that the risk factors of AD in cluster-1 are more complicated. Subsequently, we applied the functional annotation tool of KEGG and GO in Enrichr for enrichment analysis based on these genes. In cluster-1 MCI patients, there are some microorganisms (such as gram-negative bacterium and herpes viruses (22)) that can escape immune responses. These microorganisms activated immune responses, such as microglia, to clear the toxic proteins and widespread remnants from dying cells. Furthermore, these remnants continue to spread new amyloid-β clusters causing inflammatory storms (21). Above is the reason that MCI in cluster-2 patients may have synaptic failure and degeneration conditions. For example, the reduction in synaptic vesicle proteins has been shown to have a strong association with the clinical symptoms of dementia (27). We speculated that it is the storm caused by inflammasomes in the brain that result in cluster-1 MCI patients to develop the disease more rapidly than cluster-2 patients. Also, the perturbations of many other pathways have associated with the cause of AD. For example, Moriguchi et al. proposed that AD may be brain diabetes, and insulin signaling pathway is an important pathway for causing AD (23). And perturbation of pathways such as protein processing in endoplasmic reticulum, inositol phosphate metabolism and fubiquitin mediated proteolysis pathways will contribute to the amyloid cascade, which closely related to senile plaques and thus causing AD (24). The cluster-2 significantly enriched pathways mainly consisted of neuronal signaling-related pathways, and some scholars considered AD as a synaptic dysfunction caused by diffusible oligomeric assemblies of the amyloid-β protein (27). Both cluster-1 and cluster-2 enriched KEGG pathways of significantly differentially expressed genes have the endocytosis pathway. Hence, we speculated that endocytosis is the basic molecular mechanism of AD.
SNP data and mRNA expression profiling collected from patients’ peripheral blood have the characteristics of non-invasiveness and cost-effectiveness markers to identify MCI subtypes for clinical application. Clinical decisions will most likely be dictated by the genetic characteristics of AD patients in the coming years. We believe our method can effectively identify the subtypes of MCI patients, and can be applied in clinical in the future. Tailoring our method based on individual genetic characteristics will help doctors and researchers develop better therapeutic strategies and save many of MCI patients from receiving unnecessary toxic therapy. Further study should take into account the factors that can influence gene expression. For example, some other pathologies, influencing the expression of certain genes, may be present in elderly MCI patients. It may have an impact on the subtyping of MCI patients.
Two experiments can illustrate the clinical relevance of our method. For the first experiment, the expression data of 44 AD patients at baseline from the ADNI dataset were downloaded. We performed a hierarchical clustering analysis of patients with AD and patients of the two subtypes of MCI based on expression data using a similarity measure in SNF. The results are shown in the following Figure S2. This figure clearly shows that most AD patients are clustered with MCI cluster-1 patients. For the other experiment, 27 patients with AD at baseline in the ADNI dataset were downloaded. We applied the label propagation algorithm to assign new patients to subtypes. The subtype labels of these MCI patients were listed in Table S2. To test the effectiveness and reliability of our method, three AD cognitive scales were also displayed in 24-month for two subtypes of AD patients. As is shown in Figure S3, cognitive decline in cluster-1 MCI patients tends to be more remarkable than that of cluster-2 over 24 months, which is similar to the MCI patients in the ADNI dataset.
Hence, we believe our method can effectively identify the subtypes of MCI patients, and can be applied in clinical in the future. We look forward to potential collaborations with doctors and experimental biologists. We hope that the subtyping of MCI patients predicted with our model, will demonstrate its medical and therapeutic meaning. Besides, different types of data share complementary information, which is robust to noise and data heterogeneity (9). In the future, other types of biological data, such as DNA methylation and miRNA expression, can be integrated to explore biological patterns related to identify MCI subtypes. And classifying MCI patients into meaningful subtypes may improve the forecasting performance to proposing a method for predicting the conversion from MCI to AD (30).

Availability of data and material: Data used in this study are available through the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (

Author Contributions: HTL, SXY were involved with conception, design, and interpretation of data. HTL and JSW were involved with data analysis. XS, YL and XZZ provided general overall supervision of the study. XS acquired funding. All authors contributed to drafting and critical revision of the manuscript and have given final approval of the version to be published.

Funding: This research was sponsored by the National Natural Science Foundation of China (61972084, 81830053) and the Key Research and Development Program of Jiangsu province of China (BE2016002-3).

Acknowledgements: Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database ( As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. ADNI investigators include (complete listing available at ADNI_Authorship_list.pdf).

Conflicts of Interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Ethical Standards: Data involved in this study came from the Alzheimer’s disease neuroimaging initiative (ADNI) database. All procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the Helsinki Declaration of 1975.





1. Hebert LE, Weuve J, Scherr PA, Evans DA. Alzheimer disease in the United States (2010–2050) estimated using the 2010 census. Neurology 2013;80:1778–1783
2. Bagyinszky E, Youn YC, An SSA, Kim S. The genetics of Alzheimer’s disease. Clin Interv Aging 2014;9:535–551
3. Murray ME, Graff-Radford NR, Ross OA, Petersen RC, Duara R, Dickson DW. Neuropathologically defined subtypes of Alzheimer’s disease with distinct clinical characteristics: a retrospective study. Lancet Neurol 2011;10:785–796
4. Park J-Y, Na HK, Kim S, et al. Robust Identification of Alzheimer’s Disease subtypes based on cortical atrophy patterns. Sci Rep 2017;7:43270
5. Qiang W, Yau W-M, Lu J-X, Collinge J, Tycko R. Structural variation in amyloid-β fibrils from Alzheimer’s disease clinical subtypes. Nature 2017;541:217–221
6. Mukherjee S, Mez J, Trittschuh EH, et al. Genetic data and cognitively defined late-onset Alzheimer’s disease subgroups. Molecular psychiatry 2018
7. Mitchell AJ, Shiri-Feshki M. Rate of progression of mild cognitive impairment to dementia–meta-analysis of 41 robust inception cohort studies. Acta Psychiatr Scand 2009;119:252–265
8. Weiner MW, Aisen PS, Jack CR, et al. The Alzheimer’s Disease Neuroimaging Initiative: Progress report and future plans. Alzheimers Dement 2010;6:202–11.e7
9. Wang B, Mezlini AM, Demir F, et al. Similarity network fusion for aggregating data types on a genomic scale. Nat Methods 2014;11:333–337
10. Zhu X, Ghahramani Z. Learning from Labeled and Unlabeled Data with Label Propagation. 2002
11. Purcell S, Neale B, Todd-Brown K, et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. The American Journal of Human Genetics 2007;81:559–575
12. Howie B, Fuchsberger C, Stephens M, Marchini J, Abecasis GR. Fast and accurate genotype imputation in genome-wide association studies through pre-phasing. Nat Genet 2012;44:955–959
13. Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res 2003;31:e15
14. Ashburner J, Friston KJ. Voxel-based morphometry—the methods. Neuroimage 2000;11:805–821
15. Wright IC, McGuire PK, Poline JB, et al. A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. Neuroimage 1995;2:244–252
16. Ashburner J. A fast diffeomorphic image registration algorithm. Neuroimage 2007;38:95–113
17. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 2002;15:273–289
18. Ribas GC. The cerebral sulci and gyri. Neurosurg Focus 2010;28:E2
19. Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015;43:e47
20. Kuleshov MV, Jones MR, Rouillard AD, et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 2016;44:W90-97
21. Abbott A. Is ‘friendly fire’ in the brain provoking Alzheimer’s disease? Nature 2018;556:426
22. Readhead B, Haure-Mirande J-V, Funk CC, et al. Multiscale Analysis of Independent Alzheimer’s Cohorts Finds Disruption of Molecular, Genetic, and Clinical Networks by Human Herpesvirus. Neuron 2018;99:64-82.e7
23. Moriguchi S, Ishizuka T, Yabuki Y, et al. Blockade of the KATP channel Kir6.2 by memantine represents a novel mechanism relevant to Alzheimer’s disease therapy. Molecular Psychiatry 2018;23:211–221
24. Dickson DW. Apoptotic mechanisms in Alzheimer neurofibrillary degeneration: cause or effect? J Clin Invest 2004;114:23–27
25. Theuns J, Van Broeckhoven C. Transcriptional regulation of Alzheimer’s disease genes: implications for susceptibility. Hum Mol Genet 2000;9:2383–2394
26. Dong Y, Lagarde J, Xicota L, et al. Neutrophil hyperactivation correlates with Alzheimer’s disease progression. Ann Neurol 2018;83:387–405
27. Ovsepian SV, O’Leary VB, Zaborszky L, Ntziachristos V, Dolly JO. Synaptic vesicle cycle and amyloid β: Biting the hand that feeds. Alzheimer’s & Dementia 2018;14:502–513
28. Karch CM, Goate AM. Alzheimer’s Disease Risk Genes and Mechanisms of Disease Pathogenesis. Biological Psychiatry 2015;77:43–51
29. Van Nostrand WE, Melchor J, Wagner M, Davis J. Cerebrovascular smooth muscle cell surface fibrillar A beta. Alteration of the proteolytic environment in the cerebral vessel wall. Ann N Y Acad Sci 2000;903:89–96
30. Vinutha N, Pattar S, Sharma S, Shenoy PD, Venugopal KR. A Machine Learning Framework for Assessment of Cognitive and Functional Impairments in Alzheimer’s Disease: Data Preprocessing and Analysis. J Prev Alzheimers Dis 2020;7:87–94