jpad journal

AND option

OR option

A FAY-HERRIOT MODEL FOR ESTIMATING SUBJECTIVE COGNITIVE DECLINE AMONG MILITARY VETERANS

 

J.T. McDaniel, R.J. McDermott, T. Schneider

 

Southern Illinois University Carbondale, USA

Corresponding Author: Justin T McDaniel, Southern Illinois University Carbondale, USA, jtmcd@siu.edu

 


Abstract

BACKGROUND: Although studies have examined the geographic distribution of dementia among the general population in order to develop geographically targeted interventions, no studies have examined the geographic distribution of subjective cognitive decline (SCD) among military veterans specifically.
Objectives: To map the geographic distribution of subjective cognitive decline from 2011-2019 in the United States among military veterans.
Design: Cross-sectional.
Setting: United States.
Participants: Individuals reporting previous service in the United States Armed Forces.
Measurements: Using 2011 Behavioral Risk Factor Surveillance System (BRFSS) data, which is last year for which geocoded SCD data is publicly available, we estimated the survey-weighted county-level prevalence of veteran SCD for counties with >30 veterans (43 counties in 7 states). We then developed a Fay-Herriot small area estimation linear model using auxiliary data from the Census, with county-level veteran-specific covariates including % >65 years old, % female, % college educated, and median income. Following model validation, we created beta-weighted predictions of veteran SCD for all USA counties for 2011-2019 using relevant time-specific Census auxiliary data. We provide choropleth maps of our predictions.
Results: Results of our model on 43 counties showed that county-level rates of SCD were significantly associated with all auxiliary variables except annual income (F = 1.49, df = 4, 38). Direct survey-weighted rates were correlated with model-predicted rates in 43 counties (Pearson r = 0.32). Regarding predicted rates for the entire USA, the average county-level prevalence rate of veteran SCD in 2011 was 13.83% (SD = 7.35), but 29.13% in 2019 (SD = 14.71) – although variation in these rates were evident across counties.
Conclusions: SCD has increased since 2011 among veterans. Veterans Affairs hospitals should implement plans that include cognitive assessments, referral to resources, and monitoring patient progress, especially in rural areas.

Key words: Alzheimer’s disease, subjective cognitive decline, veterans, geographic distribution, country, maps.


 

Introduction

Subjective cognitive decline (SCD), which is a self-reported measure of increasing severity of memory loss or confusion, may be one of the earliest symptoms of Alzheimer’s disease (1). Undertaking an intervention as soon as possible following noticeable SCD may delay its progress, and thus, its subsequent deleterious health effects, including mortality (2). In the United States (U.S.) an estimated 6.2 million people have Alzheimer’s disease (3, 4). An estimated one in nine people in the U.S. general population aged 45 and older (11.1%) report SCD, a figure that rises to 11.6% for persons 65 and older (5). The U.S. Centers for Disease Control and Prevention (CDC) labels SCD as a public health issue of growing concern (6).
Taylor et al. (7) show that military veterans are more likely to report SCD than nonveterans (13.6% vs. 10.8%), possibly due to the increased risk of exposure to risk factors during military service (8). Although some studies have examined the distribution of cognitive decline in the general population to develop geographically targeted interventions (9, 10), no studies have examined the geographic distribution of SCD among military veterans specifically. Given the lack of a published choropleth map of military veteran SCD rates in the U.S. that would enable a focus on geographically targeted interventions, we sought identification of temporal and geographic trends in SCD rates among military veterans.

 

Methods

Data Source and Sample

We acquired data from the CDC’s 2011 Behavioral Risk Factor Surveillance System (BRFSS), the most recent year for which subjective cognitive decline data are publicly available with county-level geocodes (11). BRFSS is a telephone survey of non-institutionalized adults (e.g., persons who are not in nursing homes or on active duty in the military) in the U.S., aged > 17 years. Topics covered in the BRFSS include, but are not limited to, health behaviors, healthcare access, and chronic diseases. In 2011, the BRFSS had a response rate of 53% for landline telephone respondents and 28% for cell phone respondents.
During the 2011 BRFSS survey waves, seven states included the subjective cognitive decline survey questions: Hawaii, Illinois, New Hampshire, South Carolina, Tennessee, West Virginia, and Wisconsin. Consequently, the analytic sample in this study was limited to residents in those seven states. We delimited the dataset to individuals who answered in the affirmative to the following question about military service status (n = 6,108): “Have you ever served on active duty in the United States Armed Forces, either in the regular military or in a National Guard or military reserve unit?”

Measures

SCD was measured using the following question: “During the past 12 months, have you experienced confusion or memory loss that is happening more often or is getting worse?” Closed-ended response options to the aforementioned question included yes, no, don’t know, or refuse to answer. Of 6,108 military veteran respondents, 53 veterans indicated that they did not know whether they had experienced SCD and 93 refused to answer the question. Subsequently, we delimited the dataset only to military veterans who responded with an answer of yes or no to the SCD question (n = 5,962).
The BRFSS protocol also asked participants: “What county do you live in?” In response to this question, survey participants stated the name of the county in which they lived at the time of the survey. The CDC subsequently coded each person’s geographic location response with a federal information processing standard (FIPS) number.

Data Analysis

Phase 1

For the present study, we followed the area-level Fay-Herriot (12) approach to estimate the prevalence of SCD among military veterans in small areas (i.e., counties). As Li and Lahiri (13) note, “the Fay-Herriot model has been widely used in small area estimation and related problems for a variety of reasons, including its simplicity, its ability to protect confidentiality of microdata and its ability to produce design-consistent estimators” (p. 882). Given our analysis involves a subset of the general population (e.g., military veterans), we selected the Fay-Herriot approach especially for it’s ability to retain the confidentiality of microdata. In the first phase of our analysis, we calculated direct survey-weighted prevalence rates of SCD for each county in the BRFSS dataset. Only counties with ≥ 30 military veterans in the denominator of this rate were included in the analysis given that the National Center on Health Statistics has suggested that rates with denominators < 30 produce unstable estimates (14). In total, our analyses revealed that 43 counties met the criteria for inclusion in the study, with county samples ranging from 32 to 599 (M = 87, SD = 92).

Phase 2

Following procedures outlined in Fay and Herriot’s (12) paper, we obtained auxiliary data (i.e., county-level independent variables) for each of the 43 counties in our dataset. Specifically, we obtained 2011 county-level summary measures for each of these counties in the following areas from the U.S. Census Bureau’s American Community Survey (15): (a) the percent of military veterans aged ≥ 65 years, (b) percent of military veterans reporting female as their sex, (c) percent of military veterans with at least a bachelor’s degree, and (d) median annual income for military veterans. The previously listed auxiliary variables have been shown to influence cognitive decline (16). Our selection of 4 auxiliary variables also was determined, in part, by considerations of power and model stability. That is, with only 43 counties in our dataset we did not want to overextend our model with > 4 predictor variables (17). Subsequently, we regressed our county-level prevalence rates of SCD on these 4 county-level auxiliary variables in a multivariable regression model. Although it would have been useful to explore interaction effects in this model, we were unable to do so owing to the limited sample size. Model estimation was performed in Stata IC version 16.

Phase 3

After we fit a model to the data, we conducted an internal model validation procedure following the guidelines of Zhang et al. (18). Specifically, we calculated a Pearson correlation coefficient for the relationship between our modeled county-level prevalence rates and the county-level prevalence rates based on the raw survey data. We also created a scatterplot for the relationship between these two variables using the “ggplot2” package in R Studio version 3.6.1. As in Zhang et al.’s (18) study, a statistically significant and positive correlation coefficient was indicative of consistency between the modeled rates and the direct survey estimates.

Phase 4

Following model validation, as described in phase 3, we obtained comprehensive and nationally representative county-level summary data (i.e., data for 3,220 counties in the U.S.) on our 4 previously described military veteran-specific auxiliary variables from the 2011-2019 U.S. Census Bureau’s American Community Surveys. Subsequently, we applied the beta coefficients calculated in our linear regression model (i.e., from phase 2) to the U.S. Census Bureau county-level summary data to estimate predicted prevalence rates of SCD for all U.S. counties from 2011 to 2019. Using ESRI ArcGIS version 10.5.1., we developed county-level choropleth maps of our predictions by year, as well as by percentage change from 2011-2019.

 

Results

Phase 1

Table 1 shows the characteristics of the military veterans included in our 43 counties. All 7 states with available data on SCD also included at least one county with a military veteran sample size ≥ 30. The aggregated prevalence rate of SCD in the sample, Table 1 shows, was 11.96% (SD = 32.50). County-level prevalence rates of SCD ranged from 2.38% (i.e., Shelby County, Tennessee) to 22.02% (i.e., Horry County, South Carolina), with an average of 12.24% (SD = 4.91).

Table 1. Demographic characteristics of the military veteran sample with available data on subjective cognitive decline, BRFSS 2011 (n = 5,962)

 

Phase 2

Table 2 shows the results of our multivariable linear regression model, where the BRFSS survey-weighted rates of SCD (n = 43) were regressed on 4 auxiliary variables obtained from the U.S. Census Bureau. Although we fit models with 1-3 auxiliary variables, the model with all 4 auxiliary variables explained the greatest amount of variance in SCD among veterans. Model 4 indicates that rates of SCD at the county level were positively associated with the percent of veterans aged ≥ 65 years, positively associated with the percent of veterans reporting female as their sex, and negatively associated with the percent of veterans with a college degree.

Table 2. Fay Herriot models for the prediction of subjective cognitive decline among military veterans

a. Unstandardized beta and robust standard error; b. e-5, scientific notation

 

Phase 3

We obtained predictions from the model described in phase 2 for all 43 counties for which military veteran data was available. We calculated a Pearson correlation coefficient between the county-level predictions and the direct survey estimates. As Figure 1 shows, our internal model validation procedure revealed that SCD rates modeled with our linear regression model were consistent with the direct survey estimates (r = 0.32, p = 0.03).

Figure 1. Scatterplot illustrating the relationship between the predicted prevalence rates of SCD based on our linear regression model and the direct survey-weighted prevalence estimates (n = 43)

 

Phase 4

Given the results of our internal model validation procedure, we used the model coefficients [y = –0.19 + (0.02*X1) + (0.04*X2) + (0.000016*X3) – (0.02*X4)] to estimate the prevalence of SCD among military veterans in 3,220 counties in the U.S. from 2011 to 2019. Figure 2 shows the results of our predictions across the 9-year study period. In 2011, the average county-level prevalence rate of veteran SCD, across all counties in the U.S. (i.e., not just the original 43 counties), was 13.83% (SD = 7.35), but in 2019, the estimated average county-level prevalence rate of veteran SCD was 29.13% (SD = 14.71) – although variation in these rates were evident across U.S. counties.
Using the formula for percent change (i.e., Y2 – Y1/Y1), we examined changes in SCD rates among military veterans by county for 2011-2019. Our analysis revealed that 2,948 counties (91.55%) had increases in predicted military veteran SCD for 2011-2019, with counties in Texas, Georgia, Kentucky, Kansas, West Virginia, Montana, and Puerto Rico having the highest increases (i.e., > 1000% increase).

Figure 2. Predicted county-level prevalence rates of SCD among military veterans from 2011-2019 in the United States

 

Discussion

Using publicly available, de-identified data from a nationally representative health indicator survey and auxiliary data from the U.S. Census Bureau, we were able to estimate a predictive model of SCD among military veterans. We subsequently used our predictive model to estimate the prevalence of SCD among military veterans in 3,220 counties across the U.S. Results showed that the prevalence of military veteran SCD increased in 91.55% of counties, with counties in Texas, Georgia, Kentucky, Kansas, West Virginia, Montana, and Puerto Rico having the greatest increases.
Research demonstrates that early intervention in Alzheimer’s disease can have quality of life and economic benefits (19). The prevention of Alzheimer’s disease may be carried out with pharmacological (20) or behavioral approaches, especially those that target cardiovascular problems, which are risk factors for cognitive decline (21). For military veterans, specifically, greater enrollment in the U.S. Department of Veterans Affairs (VA) programs focused on physical activity and nutrition, such as the “MOVE!” weight management program (22), may be particularly advantageous in preventing cognitive decline. Complementary and integrative medicine approaches, such as the VA THRIVE program (23), also have been shown to improve mental functioning in military veterans and may be applicable to the prevention of Alzheimer’s disease.

Limitations

Several limitations accompany the methods and results of the present study. First, although the measure of SCD used in this study has been used widely, it only includes one self-report item and, therefore, may be subject to recall or social desirability bias. Second, several counties in the present study were excluded due to unstable or unavailable data, which prevented a more comprehensive picture of the geographic distribution of SCD in military veterans. Third, although other county level predictors may have been beneficial for inclusion in our predictive model, we were limited due to the small sample size of counties obtained from the BRFSS dataset.
Those limitations notwithstanding, this is the first study to map the prevalence of SCD among military veterans in U.S. counties. Our analysis revealed geographic variation in the distribution of SCD prevalence in the U.S., which implies a need for geographically targeted interventions that may prevent or moderate the progression of cognitive decline in veterans.

 

Funding: No funding was received for this study.

Conflicts of interest/Competing interests: The authors have no conflicts of interest to disclose.

Availability of data and material: https://www.cdc.gov/brfss/annual_data/annual_2011.htm

Code availability: Not applicable.

Ethical standards: Because this study used de-identiied data freely available on the web, this study was considered exempt from IRB review.

 

References

1. Alzheimer’s Association. 2018 Alzheimer’s disease facts and figures. Alzheimer’s and Dementia. 2018;14:367-429 1. https://doi.org/10.1016/j.jalz.2018.02.001.
2. Crous-Bou M, Minguillón C, Gramunt N, et al. Alzheimer’s disease prevention: From risk factors to early intervention. Alzheimer’s Research and Therapy. 2017; 9:71 2. https://doi.org/10.1186/s13195-017-0297-z.
3. Alzheimer’s Association. 2021 Alzheimer’s disease facts and figures. Special report. Race, ethnicity and Alzheimer’s in America. Chicago, IL: Alzheimer’s Association. Retrieved 03/292021 from https://www.alz.org/media/Documents/alzheimers-facts-and-figures.pdf.
4. Rajan KB, Weuve J, Barnes LL, McAninch EA, Wilson RS, Evans DA. Population estimate of people with clinical AD and mild cognitive impairment in the United States (2020-2060). Alzheimer’s & Dementia. In press.
5. Alzheimer’s Association. Fact sheet. 2020. Retrieved 03/29/2021 from https://www.alz.org/media/Documents/subjective-cognitive-decline-in-us.pdf
6. Centers for Disease Control and Prevention. Subjective cognitive decline – a public health issue. 2019. Retrieved 03/29/2021 from https://www.cdc.gov/aging/agingdata/docs/subjective-cognitive-decline-508.pdf
7. Taylor CA, Bouldin ED, McGuire LC. Subjective cognitive decline among adults aged >45 years – United States, 2015-2016. Morbidity and Mortality Weekly Report. 2018;67(27):753-757 7. http://dx.doi.org/10.15585/mmwr.mm6727a1.
8. Weiner MW, Friedl KE, Pacifico A, et al. Military risk factors for Alzheimer’s disease. Alzheimer’s and Dementia. 2013;9(4):445-451 8. 10.1016/j.jalz.2013.03.005.
9. Kirson NY, Meadows ES, Desai U, Smith BP, Cheung HC, Zuckerman P, Matthews BR. Temporal and geographic variation in the incidence of Alzheimer’s disease diagnosis in the US between 2007-2014. Journal of the American Geriatrics Society. 2019;68(2): 346-353 9. 10.1111/jgs.16262.
10. Fowler M, DiBlasio C, Crowe M, Kennedy RE. Geographic disparities in county-level prevalence of Alzheimer’s disease across the United States. Innovations in Aging. 2019;3(1):S782.
11. Centers for Disease Control and Prevention. BRFSS 2011 survey data and documentation. 2011. Retrieved 1/2/2021 from https://www.cdc.gov/brfss/annual_data/annual_2011.htm.
12. Fay RE, Herriot RA. Estimates of income for small places: An application of James-Stein procedures to census data. Journal of the American Statistical Association. 1979;74:269-277 12. https://doi.org/10.2307/2286322.
13. Li H, Lahiri P. An adjusted maximum likelihood method for solving small area estimation problems. J Multivariate Analysis. 2010;101(4):882-982 13. https://doi.org/10.1016/j.jmva.2009.10.009.
14. Parker JD, Talih M, Malec DJ, et al. National Center for Health Statistics data -presentation standards for proportions. National Center for Health Statistics. Vital Health Statistics. 2017;2(175):1-22 14. https://www.cdc.gov/nchs/data/series/sr_02/sr02_175.pdf.
15. United States Census Bureau. Advanced search. 2021. Retrieved 1/5/2021 from https://data.census.gov/cedsci/advanced.
16. Evans DA, Hebert LE, Beckett LA, Scherr PA, Albert MS, Chown MJ, Pilgrim DM, Taylor JO. Education and other measures of socioeconomic status and risk of incident Alzheimer disease in a defined population of older persons. Archives of Neurology. 1997;54:1399-1405. 16. 10.1001/archneur.1997.00550230066019
17. VanHoorhis CRW, Morgan BL. Understanding power and rules of thumb for determining sample sizes. Tutorials in Quantitative Methods for Psychology. 2017;3(2):43-50. 17. 10.20982/tqmp.03.2.p043
18. Zhang X, Holt JB, Yun S, Lu H, Greenlund KJ, Croft JB. Validation of multilevel regression and poststratification methodology for small area estimation of health indicators from the Behavioral Risk Factor Surveillance System. American Journal of Epidemiology. 2015;182(2):127-137. 18. 10.1093/aje/kwv002
19. 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 Neurology. 2014;14,101. 19. 10.1186/1471-2377-14-101
20. Lopez C, Caputo A, Liu F, Riviere ME, Rouzade-Dominguez ML, Thomas RG, Langbaum JB, Lenz R, Reiman EM, Graf A, Tariot PN. The Alzheimer’s prevention initiative generation program: Evaluating CNP520 efficacy in the prevention of Alzheimer’s disease. Journal of the Prevention of Alzheimer’s Disease. 2017;4(4):242-246. 20. 10.14283/jpad.2017.37
21. Norton S, Matthews FE, Barnes DE, Yaffe K, Brayne C. Potential for primary prevention of Alzheimer’s disease: An analysis of population based data. Lancet Neurology. 2014;13:788-794. 21. 10.1016/S1474-4422(14)70136-X
22. United States Department of Veterans Affairs. MOVE! Weight management program. 2021. Retrieved 3/9/2021 from https://www.move.va.gov/.
23. Haun JN, Paykel J, Alman AC, Patel N, Melillo C. A complementary and integrative health group-based program pilot demonstrates positive health outcomes with female veterans. EXPLORE. 2020;16(2):85-89. 23. https://doi.org/10.1016/j.explore.2019.08.001