Individual-Level Factors Associated with 10-Year Incidence of Alzheimer’s Disease and Related Dementias in the VA Million Veteran Program

Abstract

Background and Objectives:

Approximately 450,000 Veterans are living with Alzheimer’s disease and related dementias (ADRD), and the high prevalence of ADRD represents a major public health challenge to the Veterans Health Administration. While advancing age and genetic predisposition are well-established ADRD risk factors, growing evidence suggests that additional modifiable factors may also play an important role. This study leveraged data from the VA Million Veteran Program (MVP) to: (1) estimate 10-year incidence of ADRD and (2) evaluate associations between a broad range of individual-level risk and resilience factors and incident ADRD in a large, nationally representative sample of Veterans.

Methods:

This retrospective cohort study included Veterans aged ≥65 years at MVP enrollment who completed the MVP Baseline Survey and had VA electronic health record (EHR) data available. Individual-level variables including sociodemographic factors, military-specific characteristics, military environmental exposures (MEEs), health conditions, and health behaviors were characterized using MVP Baseline Survey data and supplemented with EHR data as available. The primary outcome was ADRD, which was determined using a validated algorithm based on International Classification of Diseases diagnosis codes extracted from the EHR. Associations between each risk/resilience factor and incident ADRD were examined using separate Cox regression models adjusted for age, sex, and education.

Results:

The sample included 245,949 Veterans (age: M=73.16, SD=6.84; 2.59% female). Approximately 4.56% (n=11,216) of the sample developed ADRD over 10 years. History of traumatic brain injury (TBI; HR=2.96, 95% CI [2.76–3.17]), depression (HR=2.93, 95% CI [2.82–3.04]), and alcohol use disorder (AUD; HR=2.35, 95% CI [2.19–2.53]) were the health factors most strongly associated with ADRD. ADRD risk was also elevated among Veterans with history of exposure to Agent Orange (HR=1.09, 95% CI [1.03–1.14]), chemical/biological warfare agents (HR=1.31, 95% CI [1.23–1.39]), and pyridostigmine bromide tablets (HR=1.67, 95% CI [1.44–1.93]).

Discussion:

Findings identified TBI, depression, AUD, and MEEs as key variables associated with ADRD in Veterans. These factors may represent important targets for prevention and intervention efforts aimed at improving the long-term health of aging Veterans. Additional work is needed to clarify the mechanisms through which these factors influence ADRD risk and to establish whether observed associations are causal.

INTRODUCTION

As the Veteran population ages, Alzheimer’s disease (AD) and AD-related dementias (ADRD) are becoming an increasingly urgent concern for the Veterans Health Administration (VHA). Recent estimates indicate that approximately 450,000 Veterans are currently impacted by ADRD, and this number is projected to increase 29% by 2033.¹ In 2013 alone, VHA healthcare expenditures for ADRD-related treatment exceeded $3.1 billion, and the associated economic burden is expected to increase substantially as novel biomarker-based diagnostic assessment and emerging AD therapeutics become more widely available.² The effective management of ADRD will require targeted healthcare prevention and intervention efforts, and the success of these initiatives within the VHA depends on a more nuanced understanding of risk and resilience factors that influence the manifestation and progression of ADRD in the Veteran population.

While age and genetic factors (e.g., APOE, APP, MAPT, PARK2) are strongly associated with the development of ADRD,^3,4 growing evidence suggests that non-genetic factors also play a substantial role in late-life dementia risk.⁵ A recent report from the Lancet Commission on Dementia estimated that nearly 45% of ADRD risk may be attributable to 14 modifiable environmental (e.g., air pollution), health (e.g., cardiovascular disease, obesity), and behavioral (e.g., alcohol use, physical activity) factors.^6,7 However, the prevalence and impact of these risk factors within the Veteran population remains incompletely characterized. Veterans represent a distinct group with unique exposures and experiences—including elevated rates of military environmental exposures (MEEs),⁸ as well as health conditions such as traumatic brain injury (TBI),⁹ posttraumatic stress disorder (PTSD),¹⁰ and cardiovascular disease¹¹—that may uniquely influence ADRD risk.^12–17 These distinct experiences highlight the need to characterize ADRD risk factors specifically within the Veteran population, rather than extrapolating from data in civilian samples. Furthermore, a Veteran-specific approach to examining risk factors for ADRD is critical for the development of effective prevention and intervention strategies within the VHA healthcare system.

While previous studies have examined risk for ADRD in Veteran samples,^{12,13,15–17} this work is limited in that it typically: (1) focuses on only one or a narrow subset of risk and resilience factors (e.g., TBI, PTSD only); (2) relies exclusively on electronic health record (EHR) data, which generally does not allow for the consideration of important psychosocial and lifestyle dimensions of risk and resilience (e.g., sleep, social support, physical activity); and (3) has not comprehensively characterized the impact of military-specific factors (e.g., deployment history, combat exposure) or MEEs as potential contributors to ADRD risk. To address these gaps, the present study leveraged data from the VA Million Veteran Program (MVP) to advance our understanding of the development of ADRD in Veterans. We combined MVP questionnaire data and EHR data for a more comprehensive assessment of a wide range of risk and resilience factors in a large, nationally representative cohort of Veterans. Using this enriched dataset, we examined the 10-year incidence of ADRD among MVP-enrolled Veterans and evaluated the association between individual-level sociodemographic factors, military-specific characteristics, MEEs, health conditions, and health behaviors on incident ADRD.

METHODS

Standard Protocol Approvals, Registrations, and Patient Consents

The VA MVP is a nationwide research program launched in 2011 with the goal of understanding how a wide range of factors influence health and wellness in Veterans.^18,19 Enrollment is open to any Veteran able to provide informed consent. Upon enrollment and following informed consent, Veterans are asked to complete two comprehensive surveys about their military experiences, health history, and well-being (i.e., the “MVP Baseline Survey” and “MVP Lifestyle Survey”); provide a blood sample for genetic analysis; and grant permission to MVP for ongoing access to their EHR. MVP was approved by the VA Central Institutional Review Board (cIRB) in 2010 and the current study (“MVP026”) under which this project was conducted received cIRB approval in 2019.

Present Study & Eligibility

Veterans were eligible for this retrospective cohort study if they had completed the MVP Baseline Survey and had EHR data available. Exclusion criteria were as follows: age at MVP enrollment <65; presence of a dementia diagnosis at MVP enrollment; missing or unavailable information pertaining to ADRD diagnostic status or pertinent sociodemographic characteristics (e.g., sex); administrative/data entry errors related to censoring (e.g., implausible dates); evidence of a severe mental health condition (e.g., schizophrenia, bipolar disorder); and evidence of a neurological disorder known to impact cognition (e.g., epilepsy, HIV, multiple sclerosis). Applying these criteria resulted in a final sample of 254,949 Veterans. A diagram outlining these criteria are displayed in Figure 1.

Figure 1.

Flowchart showing participant selection for final analytic sample of N=245,949.

Participants were followed from the date of MVP enrollment until the earliest documented ADRD diagnosis, death, or last visit date before the end of the observation period (January 2011 through September 2021). Median follow-up was 4.97 years (interquartile range: [2.90–7.24]). Administrative censoring (fixed study end date) occurred in 74.67% of participants.

Data Sources & Measures

Data sources for the present study included the MVP Baseline Survey, MVP Lifestyle Survey, and VHA EHR data using MVP data release v21.1. Details on the derivation of the primary outcome, ADRD, and each risk and resilience variable are provided below, and specifics on how each risk and resilience variable was defined are included in eTable 1.

Primary Outcome: ADRD

The primary outcome used in this study was a validated ADRD phenotype.²⁰ Briefly, ADRD status was defined using International Classification of Diseases (ICD)-9/10 codes derived from the EHR (see eTable 2 for the complete list of qualifying ICD codes). ADRD was defined as having two or more ADRD-related ICD codes (i.e., ICD codes for AD, non-specific dementia, or another related dementia^20,21) documented on different dates, with age of onset 65 or older. Veterans without ADRD were defined as having no dementia-related ICD-9/10 codes, no ICD codes for mild cognitive impairment, and no evidence of any prescription medication for AD.

Risk and Resilience Factors of Interest

MVP survey data were used to characterize risk and resilience factors of interest including sociodemographic factors, military-specific characteristics, MEEs, health conditions, and health behaviors; survey data were complemented with EHR data when necessary. With regard to health conditions specifically, a Veteran was classified as having the health condition present if there was evidence of the condition from the survey data. If there was no evidence of the condition from the survey data, ICD-9/10 codes were evaluated. If there was sufficient evidence from ICD codes that the condition was present, the health condition was classified as present; if there was not sufficient evidence from ICD codes that the condition was present, the health condition was classified as absent.

Sociodemographic factors included age at MVP enrollment, sex (Male or Female), education (High School or Less, College, or Advanced Degree), race (White, Black/African American, American Indian/Alaska Native, Asian or Pacific Islander, Multiracial, Unknown, or Other), and ethnicity (Hispanic/Latino, Non-Hispanic/Latino, or Unknown), and income (<$60K/year; $60K-%99.9K/year; >$100K/year). Military-specific characteristics included deployment (Yes or No), combat exposure (Yes or No), branch of service (Army, Air Force, Marines, Navy, or Other), and service era (Before Vietnam, Vietnam, After Vietnam, or Multiple). MEE variables included Agent Orange, chemical or biological warfare agents, and pyridostigmine bromide tablets (i.e., anti-nerve agent pills). Self-reported health conditions included TBI, PTSD, depression, hearing loss, vision loss, spinal cord injury, hypertension, heart attack/coronary artery disease, congestive heart failure, pulmonary embolism or deep vein thrombosis, high cholesterol, diabetes, obesity, and obstructive sleep apnea (OSA). Health behaviors included self-reported smoking status, alcohol use (i.e., the Alcohol Use Disorders Identification Test [AUDIT-C] total score), physical exercise (Regular, Less Frequent, Never), sleep disturbance (None, Insomnia, Poor Sleep Quality, Both), and social support (Medical Outcome Survey Social Support Index). In addition, history of alcohol use disorder was collected from EHR ICD-9/10 codes. See eTable 1 for more details on how each health condition was defined.

Data Analysis

All analyses were performed with R version 4.4.1.²² Independent samples t-tests, Wilcoxon rank sum tests, or chi-square tests were used to examine differences in characteristics between Veterans with and without ADRD. Separate Cox proportional hazards regressions were used to assess the association between each variable of interest and the hazard of ADRD, adjusting for age at MVP enrollment, sex, and education. Missing data on covariates were minimal (1.60%) and were handled using complete-case analysis. The proportional hazards assumption was assessed using Schoenfeld residuals. Adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) were reported for each risk/resilience variable. Given the large number of statistical comparisons, a false discovery rate (FDR) correction was used to control for Type I error (using the Benjamini-Hochberg procedure); FDR-adjusted p-values are reported. Our interpretation of the findings also focused on the direction and magnitude of effect sizes (HRs) rather than significance alone. For increased risk, HRs of approximately 1.2, 1.5, and ≥2.0 were interpreted as small, medium, and large effects. For decreased risk (i.e., protective effects), HRs of approximately 0.83, 0.67, and ≤.50 were interpreted as small, medium, and large, respectively.²³

Data Availability

The data and code used to generate MVP results are accessible to researchers with MVP data access. Under current VA policy, MVP is only accessible to researchers with a funded MVP project (e.g., VA Merit Award, Career Development Award, NIH R01). Thus, the datasets generated and/or analyzed during the current study are not publicly available. However, VA-affiliated researchers can apply for MVP access through the funding opportunities in the Office of Research and Development (see: https://www.research.va.gov/funding/electronic-submission.cfm).

RESULTS

10-Year ADRD Incidence & Sample Characteristics

Participant characteristics and differences between Veterans with versus without ADRD are detailed in Table 1. Of the 245,949 Veterans who were included in the study, 4.56% (n=11,216) of the sample developed ADRD over 10 years of follow-up (1,240,135.26 total person-years). The corresponding incidence rate was 904.4 cases (95% CI [887.8, 921.3]) per 100,000 person-years.

Table 1.

Sample characteristics by ADRD status.

Variablesa	NoADRD (n = 234,733)	ADRD (n = 11,216)	p b
Sociodemographic Factors
Age at MVP Enrollment (mean, SD)b	72.97 (6.75)	77.09 (7.56)	<.001
Sex			.955
Male	228,663 (97.41)	10,925 (97.41)
Female	6,070 (2.59)	291 (2.59)
Education			<.001
Advanced Degree	31,111 (13.47)	1,293 (11.79)
College	135,010 (58.44)	5,768 (52.58)
High School or less	64,917 (28.10)	3,908 (35.63)
Missing (n)	3, 695	247
Race			<.001
White	198,596 (84.61)	9,108 (81.21)
African American/Black	17,794 (7.58)	1,210 (10.79)
American Indian	1,407 (0.60)	59 (0.53)
Asian or Pacific Islander	1, 873 (0.80)	70 (0.62)
Multiracial	6,450 (2.75)	321 (2.86)
Unknown	5,406 (2.30)	293 (2.61)
Other	3,207 (1.37)	155 (1.38)
Ethnicity			<.001
Non-Hispanic/Non-Latino	221,487 (94.36)	10,7435 (93.04)
Hispanic/Latino	12,425 (5.29)	739 (6.59)
Unknown	821 (0.35)	42 (0.37)
Income			<.001
<$60,000	147,106 (71.27)	8,023 (82.39)
$60,000-$99,999	40,622 (19.68)	1,299 (13.34)
>=$100,000	18,666 (9.04)	416 (4.27)
Missing (n)	28,339	1,478
Military-Specific Characteristics
Deploymentc	115,548 (49.89)	4,588 (42.17)	<.001
Missing (n)	3,134	336
Combat Exposurec	86,246 (37.99)	3,286 (30.80)	<.001
Missing (n)	7,715	548
Branch			<.001
Army	106,071 (45.19)	5,030 (44.85)
Air Force	37,618 (16.03)	1,811 (16.15)
Marines	21,550 (9.18)	944 (8.42)
Navy	46,867 (19.97)	2,125 (18.95)
Other	22,627 (9.64)	1,306 (11.64)
Service Era			<.001
Vietnam	120,780 (52.23)	3,194 (29.38)
After Vietnam	4,631 (2.00)	143 (1.32)
Before Vietnam	64,547 (27.91)	6,023 (55.41)
Multiple	41,286 (17.85)	1,510 (13.89)
Missing (n)	3,489	346
Military Environmental Exposures d
Agent Orange	72,124 (31.08)	2,407 (21.99)	<.001
Missing (n)	2,672	269
Chemical or Biological Warfare Agents	23,150 (10.01)	1,130 (10.34)	.259
Missing (n)	3,405	287
Pyridostigmine Bromide	3,297 (1.73)	182 (2.02)	.038
Missing (n)	44,328	2,228
Health Conditions c
Traumatic Brain Injury	6,687 (2.85)	926 (8.26)	<.001
Posttraumatic Stress Disorder	42,733 (18.20)	2,353 (20.98)	<.001
Depression	69,451 (29.59)	5,483 (48.89)	<.001
Hearing Loss	141,147 (60.13)	7,854 (70.02)	<.001
Vision Loss	12,460 (5.31)	1,328 (11.84)	<.001
Spinal Cord Injury	13,380 (5.70)	784 (6.99)	<.001
Hypertension	200,725 (85.51)	10,289 (91.74)	<.001
Heart Attack/Coronary Artery Disease	93,896 (40.00)	6,193 (55.22)	<.001
Congestive Heart Failure	41,093 (17.51)	3,575 (31.87)	<.001
Pulmonary Embolism or Deep Vein Thrombosis	18,604 (7.93)	1,422 (12.68)	<.001
High Cholesterol	200,345 (85.35)	9,989 (89.06)	<.001
Diabetes	104,562 (44.55)	5,746 (51.23)	<.001
Obesity*	114,462 (49.99)	5,301 (49.12)	.076
Obstructive Sleep Apnea	55,439 (23.62)	2,652 (23.64)	.948
Health Behaviors
Smoking			.010
Never	56,146 (24.19)	2,784 (25.28)
Current or Former	175,924 (75.81)	8,230 (74.72)
Missing (n)	2, 663	202
Alcohol			<.001
Low Risk	180,039 (76.70)	9,424 (84.02)
Medium Risk	38,565 (16.43)	1,392 (12.41)
High Risk	16,129 (6.87)	400 (3.57)
Alcohol Use Disorderc	10,306 (4.49)	825 (7.39)	<.001
Missing (n)	5,433	47
Physical Exercise			<.001
Regular Exercise	100,410 (43.51)	4,128 (37.92)
Less Frequent Exercise	57,232 (24.80)	2,432 (22.34)
Never	73,139 (31.69)	4,326 (39.74)
Missing (n)	3, 952	330
Sleep			<.001
None	46,357 (26.36)	2,013 (25.23)
Insomnia	67,113 (38.16)	3,101 (38.87)
Poor Sleep Quality	17,599 (10.01)	964 (12.08)
Both	44,819 (25.48)	1,900 (23.82)
Missing (n)	58,845	3,238
Social Support (Mdn; Q1, Q3)b	3. 90 (2.80, 4.70)	3.80 (2.70, 4.60)	<.001
Missing (n)	62,722	3,554

Notes: Percentages are calculated based on the available n for each variable.

^an (%) unless otherwise specified.

^bIndependent samples t-test or Wilcoxon rank sum test; Pearson’s chi-squared.

^cReference is “No”;

^dReference is “No or Don’t Know.”

^*Only survey data was used to define obesity (see eTable 1 for definition); thus, it was possible to have missing data for obesity; missing values were as follows: n=5,775 for “No ADRD” and n=424 for “ADRD.”

Abbreviations: ADRD = Alzheimer’s disease and related dementias.

Veterans with ADRD were, on average, older at enrollment compared to Veterans without ADRD (mean difference of about 4 years), and education level and annual income were overall lower among Veterans with ADRD. A higher proportion of Veterans with ADRD self-identified as African American/Black (henceforth Black) and Hispanic/Latino, whereas Veterans without ADRD self-identified more frequently as White and Non-Hispanic/Latino. Regarding military-specific characteristics, rates of deployment and combat exposure were lower among Veterans with ADRD compared to Veterans without ADRD. Furthermore, we observed differences in service era: a greater number of Veterans with ADRD had served before the Vietnam War whereas a greater number of Veterans without ADRD had served during the Vietnam War. Veterans without ADRD also endorsed higher rates of exposure to Agent Orange, which is likely attributable to the greater representation of Veterans without ADRD in the Vietnam era, given Agent Orange exposure was very common to this period.

Other notable differences were that Veterans with ADRD endorsed nearly all of the health conditions at higher rates than Veterans without ADRD. Regarding health behaviors, similar smoking patterns were observed among Veterans with versus without ADRD, but risky alcohol use (assessed as “High Risk” and “Medium Risk” using an alcohol screening measure) was more frequent among Veterans without ADRD. In contrast, a greater number of Veterans with ADRD had a documented history of an alcohol use disorder compared to Veterans without ADRD. Finally, Veterans with ADRD reported engaging in regular physical exercise less frequently and endorsed poor sleep quality more often when compared to Veterans without ADRD. Likewise, Veterans with ADRD endorsed slightly lower levels of social support compared to those without ADRD.

Associations Between Risk and Resilience Factors and ADRD

Sociodemographic Factors, Military-Specific Characteristics, Military Environment Exposures (MEEs), and ADRD Risk

Adjusting for age, sex, and education, Cox models indicated that Black (HR=1.65, 95% CI [1.55–1.75]), Multiracial (HR=1.29, 95% CI [1.16–1.45]), and Veterans who endorsed belonging to the “Other” race category (HR=1.45, 95% CI [1.24–1.71]) were at increased risk for ADRD when compared to White Veterans. Additionally, Veterans identifying as Hispanic/Latino (HR=1.33, 95% CI [1.23–1.43]) were at increased for ADRD when compared to Veterans identifying as Non-Hispanic/Latino. In contrast, Veterans with an annual income ranging from $60,000 to $99,9999 (HR=0.73, 95% CI [0.68–0.77]) and above $100,000 (HR=0.53, 95% CI [0.48–0.59]) had a decreased risk for ADRD when compared to Veterans with an annual income below $60,000.

When compared to Veterans who had served during Vietnam, those who served before Vietnam (HR=1.25, 95% CI [1.17–1.33]) and after Vietnam (HR=1.40, 95% CI [1.18–1.67]) were at increased risk for ADRD. In contrast, Veterans who had been deployed (HR=0.76, 95% CI [0.73–0.78]) and experienced combat (HR=0.78, 95% CI [0.75–0.81]) were at decreased risk for ADRD. When compared to Veterans who served in the Army, Veterans who served in the Navy (HR=0.89, 95% CI [0.85–0.93]) were at decreased risk for ADRD. Finally, exposure to pyridostigmine bromide tablets (HR=1.67, 95% CI [1.44–1.93]), chemical or biological warfare agents (HR=1.31, 95% CI [1.23–1.39]), and Agent Orange (HR=1.09, 95% CI [1.03–1.14]) were associated with an increased risk for ADRD. Adjusted HRs and 95% CIs for each risk and resilience variable of interest are detailed in Table 2 and are displayed in order of magnitude in Figure 2.

Table 2.

Results of separate Cox proportional hazards regression models adjusted for age, sex, and education.

	AHR (95 % CI)	p
Sociodemographic Factors
Race
White	Reference
African American/Black	1.65 (1.55, 1.75)	<.001
American Indian	1.25 (0.97, 1.62)	.096
Asian or Pacific Islander	0.97 (0.77, 1.23)	.800
Multiracial	1.29 (1.16, 1.45)	<.001
Unknown	1. 10 (0.91, 1.31)	.343
Other	1.45 (1.24, 1.71)	<.001
Ethnicity
Non-Hispanic/Non-Latino	Reference
Hispanic/Latino	1. 33 (1.23, 1.43)	<.001
Unknown	0.86 (0.62, 1.20)	0.393
Income
<$60,000	Reference
$60,000-$99,999	0. 73 (0.68, 0.77)	<.001
>=$100,000	0. 53 (0.48, 0.59)	<.001
Military-Specific Characteristics
Deploymenta	0.76 (0.73, 0.78)	<.001
Combat Exposurea	0.78 (0.75, 0.81)	<.001
Branch
Army	Reference
Air Force	0.99 (0.94, 1.05)	.838
Marines	1.02 (0.95, 1.09)	.637
Navy	0. 89 (0.85, 0.94)	<.001
Other	1. 08 (1.01, 1.14)	.029
Service Era
Vietnam	Reference
After Vietnam	1.40 (1.18, 1.67)	<.001
Before Vietnam	1.25 (1.17, 1.33)	<.001
Multiple	0.97 (0.90, 1.03)	.308
Military Environmental Exposures b
Agent Orange	1.09 (1.03, 1.14)	.001
Chemical or Biological Warfare Agents	1.31 (1.23, 1.39)	<.001
Pyridostigmine Bromide	1.67 (1.44, 1.93)	<.001
Health Conditions a
Traumatic Brain Injury	2.96 (2.76, 3.17)	<.001
Posttraumatic Stress Disorder	1.80 (1.72, 1.89)	<.001
Depression	2.93 (2.82, 3.04)	<.001
Hearing Loss	1.12 (1.07, 1.16)	<.001
Vision Loss	1.53 (1.44, 1.62)	<.001
Spinal Cord Injury	1.32 (1.23, 1.42)	<.001
Hypertension	1.58 (1.48, 1.69)	<.001
Heart Attack/Coronary Artery Disease	1.55 (1.49, 1.61)	<.001
Congestive Heart Failure	1.88 (1.81, 1.96)	<.001
Pulmonary Embolism or DVT	1.63 (1.54, 1.72)	<.001
High Cholesterol	1.36 (1.28, 1.44)	<.001
Diabetes	1.42 (1.37, 1.48)	<.001
Obesity	1.23 (1.19, 1.28)	<.001
Obstructive Sleep Apnea	1.26 (1.21, 1.32)	<.001
Health Behaviors
Smoking
Never	Reference
Current or Former	0.96 (0.92, 1.00)	.085
Alcohol
Low Risk	Reference
Medium Risk	0.68 (0.64, 0.72)	<.001
High Risk	0.67 (0.60, 0.74)	<.001
Alcohol Use Disordera	2.35 (2.19, 2.53)	<.001
Physical Exercise
Regular Exercise	Reference
Less Frequent Exercise	1. 10 (1.04, 1.16)	<.001
Never	1.42 (1.35, 1.48)	<.001
Sleep
None	Reference
Insomnia	1. 07 (1.01, 1.13)	.034
Poor Sleep Quality	1.39 (1.29, 1.51)	<.001
Both	1.17 (1.10, 1.25)	<.001
Social Support	0.90 (0.88, 0.91)	<.001

Notes: Separate Cox proportional hazards regressions assessed the association between each variable of interest and ADRD. All models are adjusted for age, sex, and education. p-values are false discovery rate (FDR) corrected.

^aReference is “No”.

^bReference is “No or Don’t Know”.

Abbreviations: ADRD = Alzheimer’s disease and related dementias; AHR = Adjusted Hazard Ratio; DVT = deep vein thrombosis.

Figure 2.

Adjusted hazard ratios with 95% confidence intervals of military environmental exposures (MEEs), health conditions, and health behaviors associated with ADRD. Adjusted hazard ratios are displayed on the x-axis on a logarithmic scale.

Health Conditions, Health Behaviors, and ADRD Risk

Among the health conditions evaluated, TBI history (HR=2.96, 95% CI [2.76–3.17]) and depression (HR=2.93, 95% CI [2.82–3.04]) had the largest effect on ADRD risk. Additionally, the risk of ADRD was significantly increased in Veterans with heart failure (HR=1.88, 95% CI [1.81–1.98]), hypertension (HR=1.82, 95% CI [1.70–1.95]), PTSD (HR=1.80, 95% CI [1.72–1.89]), pulmonary embolism/deep vein thrombosis (DVT) (HR=1.63, 95% CI [1.54–1.72]), heart attack/coronary artery disease (HR=1.55, 95% CI [1.40–1.59]), vision loss (HR=1.53, 95% CI [1.44–1.62]), diabetes (HR=1.42, 95% CI [1.37–1.48]), high cholesterol (HR=1.36, 95% CI [1.28–1.44]), spinal cord injury (HR=1.32, 95% CI [1.23–1.42]), OSA (HR=1.26, 95% CI [1.21–1.32], obesity (HR=1.23, 95% CI [1.19–1.28]), and hearing loss (HR=1.12, 95% CI [1.07–1.16]).

With regard to health behaviors, higher scores of risky alcohol use (‘Medium Risk’ HR=0.68, 95% CI [0.64–0.72]; ‘High Risk’ HR=0.67, 95% CI [0.60–0.74]) were associated with a decreased risk for ADRD, whereas a history of alcohol use disorder was associated with an increased risk for ADRD (HR=2.53, 95% CI [2.19–2.53]). Lack of exercise (‘Never’ Exercise HR=1.42, 95% CI [1.35–1.48]), poor sleep quality (HR=1.39, 95% CI [1.29–1.51]), and a combination of poor sleep quality and insomnia (HR=1.17, 95% CI [1.10–1.25]) were also associated with increased risk for ADRD. Finally, higher levels of social support were associated with a lower risk for ADRD (HR=0.90, 95% CI [0.88–0.91]). Adjusted HRs and 95% CIs for each risk and resilience variable of interest are detailed in Table 2 and Figure 2.

Follow-up Analyses

Given that some of the risk and resilience factors of interest may be specific to or dependent upon military service era, follow-up analyses adjusting for age, sex, and education were conducted examining the full set of risk/resilience factors and ADRD stratified by service era. Results of the stratified analyses are presented in eTable 3 (sample characteristics by ADRD status) and eTable 4 (Cox regression analyses). Overall, patterns were generally consistent with those observed in the overall sample, with several sociodemographic, military, MEE, and health conditions and behaviors showing significant associations with ADRD. However, there was some variability in the specific risk factors identified and the magnitude of these associations across service eras. For example, TBI history and depression still had the largest effect on ADRD risk, but with notable variations in the observed hazard ratios by service era. For example, the strength of association of TBI with ADRD ranged from a HR=2.74 (95% CI [2.48–3.01]) for Veterans serving before Vietnam to a HR=5.30 (95% CI [3.25–8.62]) for Veterans serving after Vietnam.

We also conducted a sensitivity analysis that used multiple imputation with predictive mean matching (using the Multivariate Imputation by Chained Equations [mice] package in R) to evaluate the impact of missing data across risk and resilience factors. Specifically, imputation was performed on those variables with >10% missingness (i.e., income, exposure to pyridostigmine bromide tables, sleep, and social support). Results of the Cox regression analyses with imputation yielded similar results as to the original analyses when no imputation was applied (see eTable 5).

Finally, regarding the proportional hazards assumption, the majority of the Schoenfeld residual plots did not show a substantial departure from linearity. However, a small violation was noted on some variables at roughly 2,700–2,800 days of follow-up. Veterans with a longer follow-up time were a slightly younger, more educated, and somewhat healthier, and had a lower incidence of ADRD (see eTable 6). Adjusted hazard ratios for stratified analyses by follow-up time (<2,700 days vs. 2,700+ days) are reported in eTable 7; results revealed no appreciable differences in the hazard ratios originally reported in the main analyses, with the strength of associations fluctuating but maintaining the same direction.

DISCUSSION

The present study leveraged data from MVP to examine the association between individual-level risk and resilience factors and 10-year incidence of ADRD in a large sample of Veterans. We combined survey and EHR data to comprehensively characterize these risk and resilience factors and used a previously validated ADRD algorithm that has been shown to reliably identify Veterans with and without ADRD. We observed nuanced differences in ADRD risk as a function of sociodemographic factors, military-specific characteristics, MEEs, and health conditions and behaviors. In particular, results showed that TBI, depression, and alcohol use disorder had the largest effect sizes and likely represent important factors contributing to the development of ADRD in older Veterans. Moreover, history of MEEs was associated with increased ADRD risk, highlighting potentially novel risk factors for the Veteran population. Targeted screening and treatment efforts for these conditions could be helpful for improving long-term health outcomes of Veterans. However, additional research is needed to clarify the temporal and mechanistic pathways linking these factors to ADRD, and to determine whether these observed associations reflect causal relationships.

Regarding ADRD incidence, 4.56% of MVP-enrolled Veterans developed ADRD across a 10-year period. This incidence rate falls within the range of reported incidence rates observed in other EHR-based studies of Veterans, which have ranged between 0.31% up to 16%.^17,24 Importantly, variations in incidence rates may be a function of critical differences in sample sociodemographic characteristics (e.g., age, sex), time periods of examination, measurement and definition of the outcome of interest, as well as who is ultimately selected into these study samples. While we leveraged EHR data in a similar manner to these prior studies, it is important to acknowledge that Veterans who self-selected into MVP and completed surveys are likely somewhat healthier than those who may be purely seeking clinical treatment within the VHA, which could explain the lower incidence rate we observed.

When examining an array of risk and resilience factors for ADRD among MVP-enrolled Veterans, we found variations in risk for ADRD across sociodemographic and military characteristics. Specifically, Veterans who identified as Black and Hispanic were at increased risk for ADRD relative to White Veterans, consistent with prior research in both Veteran and non-Veteran cohorts.^25,26 This increase in ADRD risk is likely multifactorial and has been discussed in detail in the larger aging literature,²⁷ which highlights that racial disparities in ADRD are a consequence of greater exposure to adverse conditions across the life course (e.g., chronic stress) and decreased access to important health-promoting resources.^28,29 Specific to the Veteran population, there is some evidence to suggest that racial/ethnic minorities are more likely to report combat exposure, are underrepresented at officer positions and senior level positions due structural barriers, and are less likely to received timely appointments or person-centered, coordinated care.^30,31

As for military characteristics, Veterans who served in the Navy were at decreased risk for ADRD compared to Veterans who served in the Army. In general, the Army is more involved in war-time conflict and defense when compared to the other branches of service, and critical differences in combat experiences and engagement in defense duties may explain this pattern of results.³² That said, we also observed that both deployment and combat exposure were associated with a decreased risk for ADRD. This may be a consequence of the “healthy soldier effect”³³—a phenomenon that suggests military personnel who are most likely to deploy are generally fitter and healthier compared to those who do not deploy. Relatedly, these paradoxical associations may be partially attributable to residual confounding, as unmeasured pre-service health status or socioeconomic background could influence both the likelihood of deployment/combat exposure and long-term cognitive health outcomes. Finally, recall and survival bias may play a role in the paradoxical findings. For example, some Veterans may be underrepresented in the current sample due to premature mortality or differential participation in the healthcare system.

Another notable finding from this study was that MEEs (i.e., Agent Orange, chemical or biological warfare agents, and pyridostigmine bromide tablets) were associated with an increased risk for ADRD, although we observed nuanced differences across service eras that are likely attributable to variations in toxic exposures across these war eras. These findings are consistent with prior literature that has similarly shown elevated ADRD risk with Agent Orange exposure.¹³ Studies examining the association between other exposures and ADRD are more limited, though there is evidence for a relationship between various neurotoxins (e.g., chemical weapons, pyridostigmine bromide tablets, herbicides) and cognitive functioning among Gulf War Veterans.³⁴ Nevertheless, our findings add further evidence to suggest that a relationship does exist between several MEEs and ADRD. Although more research is needed to understand the underlying mechanisms responsible for this relationship, it is possible that exposure to various chemicals and toxins may promote inflammatory processes and oxidative stress, which may result in neuronal damage and ultimately increase risk for ADRD. Taken together, our results support that the toxic exposure screenings taking place within the VA³⁵ may also facilitate identification of Veterans who may be at increased risk for ADRD.

Our study also showed significant associations between a wide range of health conditions and health behaviors and ADRD risk, with the largest effects observed for TBI, depression, and alcohol use disorder. These findings underscore the importance of ongoing VHA screening initiatives that are designed to detect and treat these conditions as soon as possible, including the VHA’s TBI Screening and Evaluation Program that tracks and monitors deployment-related TBI’s sustained by Iraq/Afghanistan-era Veterans.³⁶ Ongoing screening efforts such as these are important so that these “at risk” individuals can be identified early and promptly referred for appropriate clinical services. Finally, results revealed that several positive health behaviors were associated with decreased risk for ADRD, including physical activity, sleep, and social support. Some of these resilience factors have also been highlighted in the Lancet Commission report^6,7 and are important modifiable factors that can be leveraged across the life course to reduce ADRD risk. Furthermore, assessment of these health behaviors can easily be implemented into VHA clinical care initiatives and could be important psychoeducational prevention targets for VHA users.

The present study has a number of notable strengths. This is one of the first large scale studies of ADRD in Veterans to leverage a combination of survey and EHR data, allowing for a more comprehensive characterization of risk and resilience factors for ADRD. Importantly, several of these modifiable factors (e.g., physical activity, sleep, social support) have oftentimes been overlooked in studies focusing solely on EHR data. We also utilized a robust and well-validated algorithm that appropriately captures the spectrum of dementia cases observed within VA clinical settings²⁰ to identify Veterans with and without ADRD. Furthermore, we employed Cox proportional hazards regression models to assess associations with ADRD over a 10-year period, which allowed us to account for time-to-event censoring. Additionally, we examined and reported both full-sample and stratified analyses by military service era in an effort to characterize unique patterns that reflect distinct exposure contexts across different war periods. Lastly, the MVP cohort represents a large, diverse, and nationally representative sample of U.S. Veterans, supporting the generalizability of our findings to the broader VHA population.

There are also limitations that should be considered when interpreting these results. First, the retrospective cohort study design does not allow for any causal inferences to be made. Future studies are needed employing time-varying models and causal inference approaches to better disentangle temporal relationships and assess the effects of risk/resilience variables on ADRD over time. Second, potential sources of bias that could influence the findings should be considered. For instance, some of the MVP survey variables could not be backfilled with EHR data and may therefore be subject to retrospective recall bias; this was particularly evident for MEE variables, where a sizable proportion of Veterans endorsed uncertainty (i.e., checked “Don’t Know”) surrounding their exposures, and underscores the need for additional verification of exposure status through Department of Defense records or other data sources in future studies. There was also a high rate of missingness for select variables, including exposure to pyridostigmine bromide tablets. However, multiple imputation analysis was conducted to address missingness, and results indicated no significant differences in reported hazards ratios. Moreover, reverse causality may be a concern for certain variables, as apathy or cognitive decline may lead to reduced physical activity or social isolation, for example, and these may not be true risk factors for ADRD. We also observed a paradoxical finding in which risky alcohol use, as assessed via the AUDIT-C, was associated with a lower risk of developing ADRD. This finding has been previously reported in other Veteran-based studies⁴², and may reflect survival bias, as those with riskier behaviors may have died earlier (i.e., did not live long enough to develop ADRD), or could be the consequence of residual confounding related to overall health status or drinking patterns. Relatedly, we did not include genetic risk (e.g., APOE genotype) or medication use in our analyses and can therefore not rule out the possibility that some of the observed associations are partially explained by these or other unmeasured confounders. Other important considerations relate to the number of risk/resilience factors examined (n=30), which introduces the potential risk of spurious associations, and could also explain some of the observed paradoxical findings (e.g., risky alcohol use, service era). Finally, there are limitations associated with generalizability. Non-Hispanic White Veterans were overrepresented in the sample, with lower proportions of other racial and ethnic groups than typically observed in the VHA clinical population,³¹ which may constrain external validity of these results to the broader and increasingly diverse Veteran population. Additionally, females made up less than 3% of the overall sample, limiting the generalizability of findings to female Veterans.

CONCLUSIONS

The study revealed that an array of sociodemographic and military characteristics, MEEs, and health conditions and behaviors are associated with incident ADRD among Veterans and ultimately highlights that ADRD risk may be somewhat modifiable. Importantly, preventing TBI and increasing access to evidence-based treatments for depressive symptoms and alcohol use disorder may help mitigate risk for developing ADRD in late life. Furthermore, encouraging MEE screenings may also help with earlier identification of Veterans at risk for ADRD. From a behavioral health perspective, promoting physical activity, healthy sleep behaviors, and social support can also be leveraged across the life course to potentially reduce ADRD risk. Nevertheless, findings from this study are correlational and additional longitudinal and mechanistic studies are needed to clarify pathways linking these risk factors to ADRD and to inform and strengthen future prevention efforts.