jpad journal

AND option

OR option



T.P. Ng1,2, T.S. Lee3, W.S. Lim4, M.S. Chong2, P. Yap5, C.Y. Cheong5, K.B. Yap6, I. Rawtaer7, T.M. Liew8, Q. Gao1, X. Gwee1, M.P.E. Ng2, S.O. Nicholas2, S.L. Wee2,9


1. Gerontological Research Programme, Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore, Singapore; 2. Geriatric Education and Research Institute, Singapore; 3. Duke-NUS Medical School, Singapore; 4. Department of Geriatric Medicine, Tan Tock Seng Hospital, Singapore; 5. Department of Geriatric Medicine, Khoo Teck Puat Hospital, Singapore; 6. Department of Geriatric Medicine, Ng Teng Fong General Hospital, Singapore; 7. Department of Psychiatry, Sengkang General Hospital, Singapore; 8. Department of Psychiatry, Singapore General Hospital, Singapore; 9. Health and Social Sciences Cluster, Singapore Institute of Technology, Singapore

Corresponding Author: A/P Tze Pin Ng, Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore, Level 9, NUHS Tower Block, 1E Kent Ridge Road, Singapore 119228, Fax: 65-67772191, Email:



Background: Mild cognitive impairment (MCI) is a critical pre-dementia target for preventive interventions. There are few brief screening tools based on self-reported personal lifestyle and health-related information for predicting MCI that have been validated for their generalizability and utility in primary care and community settings.
Objective: To develop and validate a MCI risk prediction index, and evaluate its field application in a pilot community intervention trial project.
Design: Two independent population-based cohorts in the Singapore Longitudinal Ageing Study (SLAS). We used SLAS1 as a development cohort to construct the risk assessment instrument, and SLA2 as a validation cohort to verify its generalizability.
Setting: community-based screening and lifestyle intervention
Participants: (1) SLAS1 cognitively normal (CN) aged ≥55 years with average 3 years (N=1601); (2) SLAS2 cohort (N=3051) with average 4 years of follow up. (3) 437 participants in a pilot community intervention project.
Measurements: The risk index indicators included age, female sex, years of schooling, hearing loss, depression, life satisfaction, number of cardio-metabolic risk factors (wide waist circumference, pre-diabetes or diabetes, hypertension, dyslipidemia). Weighted summed scores predicted probabilities of MCI or dementia. A self-administered questionnaire field version of the risk index was deployed in the pilot community project and evaluated using pre-intervention baseline cognitive function of participants.
Results: Risk scores were associated with increasing probabilities of progression to MCI-or-dementia in the development cohort (AUC=0.73) and with increased prevalence and incidence of MCI-or-dementia in the validation cohort (AUC=0.74). The field questionnaire risk index identified high risk individuals with strong correlation with RBANS cognitive scores in the community program (p<0.001).
Conclusions: The SLAS risk index is accurate and replicable in predicting MCI, and is applicable in community interventions for dementia prevention.

Key words: Mild cognitive impairment, dementia, metabolic syndrome, diabetes, cardiovascular risk factors.



The number of people worldwide who are living with dementia is currently estimated at about 50 million, and doubles every 20 years (1). Currently, there is a lack of effective disease-modifying treatments for Alzheimer’s disease (AD) and dementia. Based on research evidence over the past decade of risk and protective factors for dementia, a population prevention approach currently offers a promising prospect for reducing the worldwide burden of dementia (2). A viable population-based strategy is the identification of individuals whose psychosocial, lifestyle and health characteristics put them at significant risks of developing dementia in its precursor and reversible stage of mild cognitive impairment (MCI). Such individuals represent population target for multi-domain lifestyle behavioral and health interventions. Accurate screening for risks of MCI and early dementia should also be validated for use at low cost in community and primary care settings. This is especially important in low- and middle-income countries where the majority of persons with dementia live.
There are a number of earlier reports of validated risk indexes that predict the onset of Alzheimer’s disease and dementia. However, few risk indexes have been evaluated for assessing the risk of the MCI precursor of dementia, using easily available variables in the clinical setting (3-9). The Mayo Clinic Study of Aging (MSCA) basic risk score comprises 12 general demographic and clinical variables (such as age, education, marital status, diabetes, hypertension, and body mass index [BMI 7 The Australian National University Alzheimer’s Disease Risk Index (ANU-ADRI), originally developed and validated for predicting AD and dementia, comprises up to 15 similar variables (including age, sex, education, social engagement, physical activity, cognitively stimulating activities, smoking, alcohol consumption, depression, traumatic brain injury BMI, diabetes, cholesterol, fish intake and pesticide exposure) (8). Although scores in both risk indexes were strongly associated with the progression from CN to MCI, their predictive ability was limited (with C-statistic of 0.60 for both). Not all the predictive variables in the ANU-ADRI are easily available even in the study cohort of cognitive normal persons. Both risk indexes have not been reported to be externally validated in an independent population sample. It is possible that the identified risk scores may perform differently outside of the original study population. Both risk indexes were developed and evaluated in Caucasian populations. It cannot be presumed that the risk indexes are generalizable to ethnically different populations. Furthermore, to date, there are no translational research reports of the deployment and evaluation of field versions of MCI risk prediction tools in pragmatic intervention trials in the community setting.
In this study, we used risk prediction model data from the Singapore Longitudinal Ageing Study (SLAS-1) population-based cohort to derive a risk index that predicted incident MCI or dementia from 3-5 (mean 4.5) years of follow up, using risk scores for age, sex, education, depression, life satisfaction, hearing loss, and the number of cardio-metabolic risk factors (waist circumference, pre-diabetes or diabetes, hypertension, triglycerides, and high-density lipoprotein levels). We evaluated the external validity of the SLAS risk index by assessing its predictive accuracy among participants in an independent validation cohort (SLAS-2). Finally, we designed a brief self-administered questionnaire field version of the risk index and evaluated its application outside of the SLAS study cohorts in a pilot community-based interventional trial. This was a National Innovation Challenge (NIC) community project which aimed to screen and enroll high-risk individuals for targeted multi-domain lifestyle interventions for dementia prevention.



Study design and participants

The Singapore Longitudinal Ageing Study (SLAS) is an ongoing population-based observational prospective cohort study of ageing and health transition among older adults, aged 55 and above in Singapore. The study was approved by the National University of Singapore Institutional Review Board. Written informed consent was obtained from all participants.
Two distinct population cohorts were recruited in separate waves of recruitment of community-dwelling older adults in geographically different locations. SLAS-1 cohort participants who were aged 55 years and above and residents in the South East region of Singapore (N=2804) were recruited in 2003-2004 and were followed up at two re-assessment visits approximately 3 years apart in 2005-2007 and 2007-2009 respectively. The participants in the SLAS-2 cohort (N=3246) were aged 55 and above and residents in the South West region who were recruited in 2009-2013, with follow up assessments 3 to 5 (mean 4.5) years apart conducted in 2013-2018. Trained nurses visited the participants at home to perform face-to face questionnaire interviews, and clinical and neurocognitive assessments and blood draws were performed in a local study site center. The background and methodology of the SLAS research are described in detail elsewhere (10, 11).

SLAS-1 development cohort study

We used data from the SLAS-1 development cohort to derive a prediction model and risk scores for incident MCI or dementia. The development cohort comprised cognitive normal participants (N=2611), after excluding small numbers of Malay and Indian participants, those with stroke, Parkinson’s disease, other brain disorders and injury (N=101), prevalent MCI (N=478) and dementia (N=42), and undetermined neurocognitive status (N=4) at baseline. During the follow-up period, 227 participants died and 296 were lost to follow up. The MCI risk prediction model was based on longitudinal data analysis of 1,610 individuals, who were followed up to incident MCI (N=149) or dementia (N=14).

SLAS-2 validation cohort study

We evaluated the external validity of the SLAS-1 risk model by applying the prediction algorithm on individual participants in the SLAS-2 validation cohort. We used the calculated risk scores to assess its accuracy in predicting prevalent and incident cases of MCI or dementia. Both prevalent and incident cognitive outcomes were evaluated because the practical application of the risk prediction tool should include the identification of prevalent cases of MCI and interventions to prevent MCI progression to dementia. At baseline, after excluding 219 participants with missing data on neurocognitive status, there were a total of 3051 participants with known neurocognitive status: cognitive normal (CN)=2700, MCI =265, dementia=86) and had complete data on risk factors. A total of 1323 cognitive normal participants followed up for an average of 4.5 years were re-assessed on their neurocognitive status (CN=1157, MCI=69, dementia=6), after excluding 45 participants with missing data.

Mild Cognitive Impairment (MCI) and Dementia

As described in detail in a previous publication (12), neurocognitive disorder (MCI and dementia) among study participants was ascertained from initial screening using MMSE and self or informant reports of subjective cognitive decline, using the IQCODE (13) followed by subsequent assessment using Clinical Dementia Rating Scale (14) and neuropsychological evaluation for cognitive domains: memory (RAVLT immediate recall, RAVLT delayed recall, Visual Reproduction immediate recall, Visual Reproduction delayed); executive function (Symbol Digit Modality Test; Design Fluency; Trail Making Test B, language (Categorical Verbal Fluency); visuospatial skills (Block Design); attention (Digit Span Forwards, Digit Span Backwards, Spatial Span Forwards and Backwards), before final clinical assessments including MRI and additional blood tests and consensus diagnosis by a panel of geriatricians and psychiatrists.
MCI was defined according to published criteria (15, 16): cognitive concern expressed by the participant or informant; cognitive impairment in one or more domains (executive function, memory, language, or visuospatial); normal functional activities; and absence of dementia, and operationalized as follows:
1) Subjective cognitive complaints from a single question asking whether the subject had more problems with memory than most, or an informant report of memory decline, “Do you think your family member’s memory or other mental abilities have declined?” or IQCODE ≥3.0.
2) Cognitive deficits defined as Modified Chinese MMSE score of 24-27, or decline of 2 more points from baseline, a neurocognitive domain score that was 1 to 2 SD below the age-and-education adjusted means, or decline from baseline of 0.5 SD from serial measurements;
3) No functional dependency in performing instrumental daily living activities (Lawton et al);
4) Clinical Dementia Rating Scale score of 0 or 0.5;
5) No dementia.

Dementia was diagnosed based on the Diagnostic and Statistical Manual of Mental Disorders, DSM-4R (American Psychiatric Association, 1994) (17), with evidence of cognitive deficit from objective assessment (MMSE ≤23 or neuropsychological domain scores below 2 SD of age-education-adjusted mean), and evidence of social or occupational function impairment (dependency in one or more Instrumental ADL or Clinical Dementia Rating score ≥1).
Participants who did not meet these criteria for MCI or dementia were classified as cognitive normal.

Predictor variables

We developed the risk prediction model from stepwise selection of a parsimonious set of significant predictor variables from among 20 known variables that represent known psychosocial, lifestyle and health risk factors for the onset and development of MCI and dementia.
1. Age, gender, education (none, 6 or less years, and more than 6 years), marital status (single, widowed, divorced), APOE- ε4 genotype carrier status;
2. Physical, social or productive activities based on the number and frequencies on a 3-point Likert scale (0=never or less than once a month; 1=sometimes (once a month or more but less than once a week); 2=often (at least once a week) of usual participation in 18 different categories of physical, social and productive activities; level of physical, social or productive activities was scored as the sum of frequencies across all activities, with higher score denoting higher level of participation; hearing loss (self-reported or whisper test);
3. Psychosocial variables: living alone, loneliness, life satisfaction as reported in a previous publication, depression (Geriatric Depression Scale ≥5) (18);
4. Smoking (current or past smoking versus never smoking); alcohol consumption (one or more drinks daily);
5. Cardio-metabolic and vascular risk factors: body mass index (BMI)≥27.5kg/m2, waist circumference ≥90cm for male and ≥80cm for female; pre-diabetes and diabetes (raised fasting plasma glucose (FPG) (FPG ≥ 100 mg/dL (5.6 mmol/L), or on treatment for previously diagnosed type 2 diabetes); raised triglyceride (TG) level ( ≥ 150 mg/dL (1.7 mmol/L) or on anti-lipid treatment); reduced high density lipoprotein (HDL) cholesterol (< 40 mg/dL (1.03 mmol/L) in males and < 50mg/dL (1.29 mmol/L) in females or on anti-lipids treatment; hypertension (raised blood pressure(BP) (systolic BP ≥ 130 or diastolic BP ≥ 85 mm Hg, or on treatment of previously diagnosed hypertension).

Field version questionnaire and validation

We created a field version of the risk prediction tool by designing a simple self-administered checklist questionnaire of the 11 risk factor items that respondents endorse, and derive a summary risk score for each individual. (Table 1 and Figure 1) We deployed the risk screening questionnaire in a community-based National Innovation Challenge trial project in which individuals with high risk scores were identified and enrolled for multi-domain lifestyle interventions for dementia prevention. A total of 437 community-living older persons aged ≥55 years without ADL disability, visual impairment, or known neurodegenerative disorders (dementia, Parkinson’s and other diseases) were administered the risk screening questionnaire at neighourhood senior activity centres. A total of 194 participants who scored ≥6 points on the risk index were enrolled into the 6-month multi-domain intervention or control programmes. At the same time, they were assessed at baseline (pre-intervention) with the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) (19, 20). Raw scores were standardized to T scores with a mean of 100 and standard deviation of 15 in the global index score, with higher scores indicating better performance.

Table 1. SLAS Index of Risk Prediction for Pre-dementia and Dementia

Figure 1. Predicted risk of MCI/Dementia and risk prediction score


Statistical Analyses

In the SLAS-1 developmental cohort, we examined a total of 20 available demographic, psychosocial, lifestyle behavior and health risk variables known from the literature review or demonstrated in our previous published reports to be risk or protective factors for the risk of MCI or dementia. We modeled the cumulative incidence (rather than time-to-event) of new cases of MCI or dementia that develop within the 4.5-year period in logistic regression as it was considered to be more informative and relevant for the purpose of predicting the risk or probability of MCI, and it would be more interpretable for lay users of the information. We used stepwise procedures to select a parsimonious set of 7 surrogate variables in the final logistic model that predict incident MCI or dementia. The outcome variable combined MCI and dementia since the follow up of these cognitive normal participants identified incident cases of MCI as well as a small number of dementia cases (N=14, diagnosed without known progression from MCI). Results are presented using complete case analysis as they were similar to results using imputation methods for missing data. Estimated regression coefficients in the logistic model were used as weights to derive simple point scores (from 0 to 3) associated with each risk indicator, and a summed risk score. In the SLAS-2 validation cohort, we used the same prediction algorithm to calculate the summary risk score for each participant and estimated the risks of prevalent and incident MCI-dementia associated with each risk score category. The predictive accuracy of the risk index was assessed by the area under the curve (AUC) using receiving operating curve techniques. In the NIC community –based interventional trial, the risk scores (ranging from 6 to 11) of 194 participants were evaluated on the trend of association with mean RBANS T-score of global cognitive performance using ANOVA with significance tests for linear trend.



SLAS1 development cohort

The baseline characteristics and risk factors of cognitively normal participants who presented with incident MCI or dementia (N=163) at follow up and their counterparts (N=1438) are shown in Table 2. Older age, female sex, low education, hearing loss, depression and low life satisfaction were significant predictors of incident MCI-dementia. Diabetes was the only cardio-metabolic risk factor among the metabolic syndrome components that was significantly associated with MCI-dementia, but the number of metabolic syndrome components was more strongly predictive of MCI-dementia.

Table 2. Baseline characteristics and risk factors of SLAS-1 development cohort of cognitive normal participants by MCI-dementia follow-up status


Table 3 shows the estimates of the strength of association of individual risk factors with incident MCI-dementia in the final risk prediction model, and associated weighted risk scores. The summed risk scores for individual participants ranged from 0 to 11, and was associated respectively with cumulative incidence of MCI-dementia ranging from 0% to 50%, (Figure 2) or 45% increased odds of incident MCI-dementia per point score increase. The AUC was 0.73, 95% confidence interval (95%CI): 0.70-0.75. Using an optimum risk score of ≥6, the sensitivity was 0.748 (95%CI: 0.677-080), specificity was 0.672 (0.647-0.695) and positive predictive value was 0.169 (95%CI: 0.144-0.199), based on the observed cumulative incidence of MCI-dementia of 10.2%
SLAS-2 validation cohort

Figure 2. Predicted cumulative probabilities of incident MCI-dementia by risk index score in SLAS-1 developmental cohort

Table 3. Estimates of association of SLAS risk factors with and risk score predictors of MCI-dementia and ROC measure of discriminant accuracy


The baseline characteristic and risk profile of participants overall and among cognitive normal participants who were followed up are shown in Table 4. In cross-sectional cohort, the risk index ranged from 0 to 12 (mean of 5.7 and SD of 2.4) and were associated with increasing prevalence of MCI-dementia ranging from 0% to 45% (Figure 3A). The AUC was 0.74 (95%CI: 0.72-0.77), sensitivity was 0.755 (95%CI: 0.707-0.797), specificity was 0.675 (95%CI: 0.576-0.613) and PPV was 0.209, 95%CI: 0.175-0.217) based on the overall prevalence of 11.5%.
In the follow up cohort of cognitive normal participants, risk scores from 0 to 12 were associated with increasing cumulative incidence of MCI-dementia from 0% to 36%. (Figure 3B) The AUC was 0.74 (95%CI: 0.69-0.79); sensitivity=0.720 (95%CI: 0.609-0.809), specificity=0.672 (95%CI: 0.644-0.698), PPV=0.124 (95%CI: 0.097-0.159), based on the observed overall cumulative incidence of 6.1%.

Figure 3a. Predicted probabilities of prevalent MCI-dementia by risk index score in SLAS-2 validation cohort


Figure 3b. Predicted cumulative probabilities of incident MCI-dementia by risk index score in SLAS-2 validation cohort

Table 4. SLAS-2 validation cohort baseline characteristics


Field version questionnaire risk score validation

The risk scores range in values from 6 to 11 (mean of 8.0 and SD of 1.1), and were significantly associated with decreasing trend of mean RBANS T-score from 105 to 95 (p for linear trend <0.001) (Figure 4).

Figure 4. RBANS T-scores by risk score among persons with risk scores of six and above who are selected for targeted intervention



We describe in this report the development of a risk index that predicts average 4.5-years risk of MCI or dementia in an Asian population. It comprises established personal, lifestyle and health factors that are easily measured and commonly used in primary care. The risk index showed good predictive accuracy both in the development cohort (AUC=0.73) and the external validation cohort (AUC=0.74). Equally important is that the scores predict a wide range of cumulative risk estimates of MCI-dementia with the highest risk score predicting a probability of 50% in the development cohort. In the validation cohort, the highest risk score predicted 45% probability of prevalent MCI-dementia and 36% probability of incident MCI-dementia. A lower prediction performance is usually demonstrated in external validation samples compared to original development samples (21, 22).
Several other risk prediction tools for MCI have been previously described. They include the Mayo Clinic Study of Aging (MCSA) basic risk score comprising age, education, marital status, diabetes, hypertension, body mass index (BMI), which has a AUC of 0.60 (7). The Australian National University-Alzheimer’s Disease Risk Index (ANU-ADRI) (8) is computed from 11 to 15 predictive variables, including self-reports of age, education, BMI, diabetes, depressive symp¬toms, high cholesterol, head trauma, smoking, alcohol consumption, social engagement (marital status, size of social network, quality of social network, level of social activities), physical activity (number of hours performing mild, moderate and vigorous activities), cognitively stimulating activities (number of cognitive activities undertaken), BMI, and (as available) cholesterol, fish intake and pesticide exposure. It also showed a AUC of 0.61 for incident MCI/dementia (5). In comparison, the SLAS risk index has higher predictive accuracy. While these studies have evaluated the predictive accuracy of their risk scoring tools within the development cohort, there are no reports of calibrating the performance of these tools in external validation cohorts. In our study, the external validity and generalizability of the SLAS risk prediction tool was demonstrated in an independent population sample (SLAS-2). This shows that the risk index is transportable to another population that differed in geographical location and other characteristics.

Furthermore, we re-designed and created a brief questionnaire version of the risk prediction index for field application in a community interventional trial. The questionnaire was self-administered or administered by a healthcare or community service provider within 2 minutes. In Singapore, basic health screening is easily accessible in primary care clinics at no or minimal costs for older residents, and include blood tests for fasting glucose and lipids, blood pressure and body size measurements. Given the advantage of ease of collection of self-reported data, there was no material loss of accuracy of information ascertainment, as shown by the RBANS cognition data. We demonstrated that the risk index questionnaire was able to discriminate individuals with varying levels of cognitive function. Using a pre-determined cutoff of 6 and above (associated with less than 10% predicted risk of MCI, or equivalent to the combined scores for non-modifiable risk factors of age, sex and education), the RBANS T-score stands at about 105 for the threshold risk scores of 6 and 7 and are close to local population norms (17), but dropped to 97 and 95 for higher score of 8 and above. The 10-point difference is more than the minimum clinically important difference of 8 for RBANS total score (18). Although the data empirically suggests a critical threshold risk score of 8 and above, practically a lower screening threshold of 6 or 7 is optimal. As yet, we found no other studies that have validated the field performance of a risk index for MCI in a pragmatic intervention trial in the community setting. Only a Portuguese questionnaire version of the ANU-ADRI was assessed in Brazil in a primary care study to be reliable and valid for assessing risk for AD (not MCI), by showing a moderate negative linear relation between the ANU-ADRI and MMSE scores (23). Our data thus show the feasibility of using the risk index to screen and identify individuals with prevalent MCI or at risk of MCI who are likely to benefit from early targeted multi-domain lifestyle intervention in a randomized controlled community trial which is ongoing.

Strengths and limitations

Our risk index for predicting MCI was based on risk factors from a single study. Other approaches of modelling with data from meta-analyses of risk factors from multiple studies may produce more predictive accuracy. A Shanghai risk model of 10 risk predictors was derived from a meta-analysis of 38 Chinese population studies and reported higher AUC from multivariate logistic regression model of 0.77. Notably, the predictive performance of the risk index was verified on a cross-sectional study predicting prevalent cases of MCI (6). Interestingly, the Shanghai study showed that better AUC of 0.86 was obtained using the same data using the Rothman-Keller model with additional model parameterizations. Further work should explore this new approach in developing risk prediction models of MCI and dementia.
Our risk index was derived from available data on 20 candidate risk and protective factors in a single population study. However, they are not much less wide and diverse in coverage as those reported in published meta-analyses. In general, risk prediction is improved with more independent predictors that contribute additionally to the model, but at greater costs of data collection. This may be done by adding information from cognitive, MRI or other testing, which may be desirable in clinical settings but does not suit the purpose of a risk index for population-based prevention interventions in the community. Our risk index was intentionally designed to be brief, inexpensive and non-invasive. It can be self-administered or administered by non-highly trained staff, hence based on a minimalist set of risk factors predicting MCI or dementia. It turns out to compare quite favorably or outperform relatively lengthy risk prediction instruments. It is possible that the weights derived for the SLAS risk index are study specific. Studies in other heterogeneous populations using the same risk prediction model may have optimal weights that differ for specific populations. More studies in ethnically similar or different populations elsewhere are needed. This risk prediction model appears to be successful in predicting risks of MCI-dementia in a Chinese population in Singapore. Whether it will be equally successful in ascertaining risks in Chinese populations in other places such as Taiwan or China, or to non-Chinese populations elsewhere needs to be verified independently with further validation studies.
The relatively short length of follow up of this cohort is notable, but is not a limitation. Risk prediction requires a specification of the time span, and in this case, a 3 to 5-year time span for predicting the risk of MCI or early dementia appears to be appropriate. The relatively young age of the study cohorts is also notable. A diversity of risk factors has their own age-developmental trajectories in predicting future risk of dementia, hence risk prediction instruments are of necessity applicable at different ages. Hypertension, obesity and cholesterol are known to be risk factors for late-life dementia when they are measured at middle age. The SLAS risk index is therefore optimal in identifying individuals at increased risk of dementia at a younger age and suits the primary purpose of early population-level interventions to prevent late-life dementia.



We developed and validated the SLAS risk index for predicting the risk of MCI or dementia over 4.5 years. We observed a high level of predictive accuracy which was replicated in an external validation population, and demonstrated the feasibility of applying a field version self-reported questionnaire of the risk index in a pilot community intervention project for dementia prevention.


Declarations: The study was approved by the National University of Singapore Institutional Review Board. Written informed consent was obtained from all participants.

Potential Conflict of Interest: The authors declare that they have no competing interests.

Availability of data: The datasets generated and analysed during the current study are not publicly available due to local data regulations and institutional policies but may be available from the corresponding author on reasonable request with permission from relevant authorities.

Funding: The study was supported by the Agency for Science Technology and Research (A*STAR) Biomedical Research Council (BMRC) [grant number 08/1/21/19/567] and the National Medical Research Council [grant number: NMRC/1108/2007].

Authors’ contributions: TPN reviewed the literature, designed the study, analyzed the data, drafted and revised the manuscript. SON provided additional data analysis. TSL, WSL, MSC, KBY, PY, CYC, IR, TML, QG, XYG, and NPEM contributed to the study design and data collection. All authors reviewed the results and drafts, and approved the final manuscript.

Acknowledgments: We thank the following voluntary welfare organizations for their support: Geylang East Home for the Aged, Presbyterian Community Services, St Luke’s Eldercare Services, Thye Hua Kwan Moral Society (Moral Neighbourhood Links), Yuhua Neighbourhood Link, Henderson Senior Citizens’ Home, NTUC Eldercare Co-op Ltd, Thong Kheng Seniors Activity Centre (Queenstown Centre) and Redhill Moral Seniors Activity Centre.

Ethical standards: The National University of Singapore (NUS) Institutional Review Board (IRB) approved the study, and all participants gave informed consent before participating.



1. Alzheimer’s Disease International. World Alzheimer Report 2010: The Global Economic Impact of Dementia. London: Alzheimer’s Disease International, 2010.
2. Norton S, Matthews FE, Barnes D, et al. Potential for primary prevention of Alzheimer’s disease: an analysis of population-based data. Lancet Neurol 2014;13:788–794.
3. Kivipelto M, Ngand T, Laatikainen T, et al. Risk score for the prediction of dementia risk in 20 years among middle aged people: a longitudinal, population-based study. Lancet Neurol 2006;5:735-741.
4. Reitz C, Tang MX, Schupf N, et al. A summary risk score for the prediction of Alzheimer disease in elderly persons. Arch Neurol 2010;67:835–841.
5. Anstey KJ, Cherbuin N, Herath PM, et al. A self-report risk index to predict occurrence of dementia in three independent cohorts of older adults: the ANU-ADRI. PLoS One 2014;DOI: 10.1371/journal.pone.0086141.
6. Vos SJB, van Boxtel MPJ, Schiepers OJG, et al. Modifiable Risk Factors for Prevention of Dementia in Midlife, Late Life and the Oldest-Old: Validation of the LIBRA Index. J Alzheimers Dis 2017;58:537-547.
7. Pankratz VS, Roberts RO, Mielke MM, et al. Predicting the risk of mild cognitive impairment in the Mayo Clinic Study of Aging. Neurology 2015;84:1433-1442.
8. ndrews SJ, Eramudugolla R, Velez JI, et al. Validating the role of the Australian National University Alzheimer’s Disease Risk Index (ANU-ADRI) and a genetic risk score in progression to cognitive impairment in a population-based cohort of older adults followed for 12 years. Alzheimers Res Ther 2017;9:16.
9. Wang B, Shen T, Mao L, et al. Establishment of a Risk Prediction Model for Mild Cognitive Impairment among Elderly Chinese. J Nutr Health Aging 2020;24:255-261.
10. Niti M, Yap KB, Kua EH, et al. Physical, social and productive leisure activities, cognitive decline and interaction with APOE-epsilon4 genotype in Chinese older adults. Int Psychogeriatr 2008;20:237-251.
11. Wei K, Nyunt MSZ, Gao Q, et al. Frailty and Malnutrition: Related and Distinct Syndrome Prevalence and Association among Community-Dwelling Older Adults: Singapore Longitudinal Ageing Studies. J Am Med Dir Assoc 2017;18:1019-1028.
12. Ng TP, Feng L, Nyunt MS, et al. Metabolic Syndrome and the Risk of Mild Cognitive Impairment and Progression to Dementia: Follow-up of the Singapore Longitudinal Ageing Study Cohort. JAMA Neurol 2016;73:456-463.
13. Jorm AF. A short form of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE): development and cross-validation. Psychol Med. 1994 Feb 24(1):145-53
14. Morris JC. The Clinical Dementia Rating (CDR): current version and scoring rules. Neurology. 1993 Nov;43(11):2412-4. doi: 10.1212/wnl.43.11.2412-a. PMID: 8232972.
15. Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Med 2004; 256:183–194.
16. 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;256:240–246.
17. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders: DSM-IV. Washington, DC: American Psychiatric Association, 1994.
18. Nyunt MSZ, Jin AZ, Fones CSL, et al. Criterion-based validity and reliability of the Geriatric Depression Screening Scale (GDS-15) in a large validation sample of community-living Asian older adults. Aging Ment Health 2009;13:376-382.
19. Collinson SL, Fang SH, Lim ML, et al. Normative Data for the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) in Elderly Chinese. Arch Clin Neuropsychol 2014;29:442-455.
20. Phillips R, Qi G, Collinson SL, et al. The Minimum Clinically Important Difference in the Repeatable Battery for the Assessment of Neuropsychological Status. Clin Neuropsychol 2015;29:905-923.
21. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules: A review and suggested modifications of methodological standards. JAMA 1997;277:488–494.
22. Schonberg MA, Davis RB, McCarthy EP, et al. External validation of an index to predict up to 9-year mortality of community-dwelling adults aged 65 and older. J Am Geriatr Soc 2011;59:1444-1451.
23. Borges MK, Jacinto AF, Citero VA. Validity and reliability of the Brazilian Portuguese version of the Australian National University – Alzheimer’s Disease Risk Index (ANU-ADRI). Dement Neuropsychol, 2018;12:235-243.