Comparing Veteran and Nonveteran Epigenetic Aging in a Representative Sample of United States Adults

ABSTRACT

Introduction

Military service can significantly impact human health, with research showing that veterans experience higher mortality rates than the general population. However, limited data exist on the relationships of veteran status with biomarkers of aging that may precede clinical illness and mortality.

Methods

Using survey-design weighted generalized linear regression models, we examined the cross-sectional relationship of self-reported veteran status with DNA methylation (DNAm)-based biomarkers of aging (epigenetic age) in a representative sample of 2344 U.S. adults participating in the 1999-2000 and 2001-2002 cycles of the National Health and Nutrition Examination Survey. We tested 7 epigenetic aging markers: HannumAge, HorvathAge, SkinBloodAge, PhenoAge, GrimAge2, DNAm Telomere Length (TL), and DunedinPoAm.

Results

After adjusting for basic demographics, veterans had marginally greater SkinBloodAge (β = 0.86 years, 95% CI: −0.10, 1.81, P = .08) and GrimAge2 (β = 0.71 years, 95% CI: −0.07, 1.49, P = .07) measures when compared to nonveterans. Similar SkinBloodAge (β = 1.00 years, 95% CI: −0.01, 2.00, P = .05) and GrimAge2 (β = 0.69 years, 95% CI: −0.14, 1.52, P = .09) relationships were observed in fully-adjusted models where missing health and lifestyle covariates were imputed. Compared to nonveterans, veterans also had higher DNAm-estimated blood levels of GrimAge2-components hemoglobin A1c (β = 0.006, 95% CI: 0.0005, 0.01, P = .03) and protein TIMP1 (β = 71.14, 95% CI: 8.28, 134.01, P = .03) in basic demographic-adjusted models. In fully-adjusted imputed models (β = 96.40, 95% CI: −15.05, 207.85, P = .08) and complete case models (β = 98.66, 95% CI: −25.24, 222.55, P = .099), the TIMP1 relationships remained marginally significant.

Conclusions

Our marginal results support existing veteran morbidity and mortality literature while suggesting a modest utility of epigenetic aging biomarkers for further understanding veteran health. As veterans represent an important subset of the population and are a priority in federal government budgets, future research in this area holds the potential for significant public health and policy impact.

INTRODUCTION

Research indicates that military service can have an important influence on an individual’s health.¹ Similar to what has been described in other employed populations, some studies have suggested that veterans have lower rates of mortality when compared to the general population given that service members may generally be healthier than the overall population—a “healthy worker/soldier effect.”² However, many studies have underscored the injuries and adverse exposures associated with military service and have provided compelling evidence of increased morbidity and mortality in veterans.³ For example, data from the National Center for Veterans Analysis and Statistics, with over 32 million veteran records, estimated that expected life-years for veterans from 2000 to 2014 were 0.8 to 1.2 life-years shorter when compared to the overall U.S. population in 2006.⁴ A different study of over 2 million post-September 11, 2001, U.S. veterans estimated higher adjusted, age-specific all-cause mortality rates in veterans (3858 estimated excess deaths) compared to the general U.S. population.⁵

Despite this evidence, there remains a notable lack of data exploring the relationship of veteran status with epigenetic age biomarkers, which are DNA methylation (DNAm)-based measures of healthspan and lifespan.^6–11 The broader literature has demonstrated that epigenetic aging is influenced by a variety of lifestyle and environmental factors.¹² Most studies in veterans have examined the relationships of stress, trauma, and substance-use disorders with epigenetic aging among veterans, but few studies have directly compared epigenetic aging in veterans and nonveterans.^13–15 For example, one study compared 112 U.S. veterans with post-traumatic stress disorder (PTSD) to 28 veteran controls without PTSD and 59 nontrauma exposed controls and reported GrimAge acceleration of approximately 2 years in veterans with PTSD.¹³ Since changes in epigenetic aging biomarkers may occur before clinical illness or death, these biomarkers create an opportunity for clinical interventions or lifestyle changes before an individual becomes ill or dies. These biomarkers may also be useful for monitoring the effectiveness of interventions after disease diagnosis. With over 18 million veterans in the United States and over $400 billion in 2024 U.S. Department of Veterans Affairs budgetary resources,^16,17 understanding the relationships of veteran status with epigenetic aging may have significant implications for population health and health policy. Epigenetic aging may be a useful biomarker for screening, early interventions, and ongoing health monitoring in this population.

In this study, we evaluate the cross-sectional relationship of veteran status with epigenetic age in the National Health and Nutrition Examination Survey (NHANES), a nationally representative sample of individuals in the United States, encompassing demographic information, questionnaire responses, and laboratory data.¹⁸ Drawing from the evidence demonstrating that veterans have greater morbidity and mortality risk, and utilizing the framework where adult epigenetic age acceleration (being biologically older than expected) is viewed as deleterious while deceleration (being biologically younger than expected) is considered to be beneficial, we hypothesize that being a veteran will be associated with greater epigenetic aging, relative to nonveterans among U.S. adults.

METHODS

Study Population

The National Center for Health Statistics conducts NHANES to evaluate the health of the noninstitutionalized U.S. population.¹⁸ The NHANES provides a representative sample of the U.S. population, gathering information through interviews, physical examinations, and laboratory tests. Our study investigating the relationship of veteran status with DNAm-based epigenetic age measures used data from the 1999-2000 and 2001-2002 NHANES cycles, where epigenetic aging data were publicly available. This subsample included data from 2532 adult study participants, all at least 50 years of age. However, to protect participant privacy, individuals older than 85 years were top coded as 85 years, making their exact ages unknown (n = 130). To prevent potential bias in epigenetic age measures, these participants were removed before analyses. Participants with a mismatch between sex predicted from DNAm data and self-reported sex (n = 56) were also removed before analyses resulting in 2346 remaining study participants. Among these, 2344 participants had data on veteran status (n = 2 missing veteran status data) and 1829 had information on all covariates. All NHANES participants gave written informed consent, and the study protocols received approval from the NCHS Research Ethics Review Board (protocol #98-12).

Veteran Status

Information on participants’ veteran status was collected via self-report. Participants answered the question, “Did you ever serve in the Armed Forces of the United States?” Responses were coded as “yes,” “no,” or “missing.” Participants with missing responses were not included in the analyses.

DNA Methylation and Epigenetic Age

We obtained epigenetic age measures and DNA methylation-based leukocyte proportion estimates from the NHANES website (https://www.cdc.gov/nchs/nhanes/index.htm), where additional information on DNA methylation analysis and processing is available. Briefly, DNA was extracted from whole blood in a subset of NHANES 1999-2000 and 2001-2002 participants ≥50 years of age, and genome-wide DNA methylation was assessed using the Illumina EPIC BeadChip array. Our analysis included several epigenetic age measures, including the HannumAge, HorvathAge, SkinBloodAge, PhenoAge, GrimAge2, and DunedinPoAm epigenetic age measures as well as DNA methylation telomere length (DNAmTL), providing for a comprehensive assessment of the most robust DNA methylation-based biological aging measures in the literature.^6–11,19,20 HannumAge, HorvathAge, and SkinBloodAge epigenetic age measures are primarily regarded as DNA methylation-based predictors of chronological age. However, research has also linked these biomarkers to overall health status.^7,8,20 PhenoAge is a leading biomarker of healthspan, developed using 9 clinical variables: albumin, creatinine, glucose, C-reactive protein (CRP), lymphocyte percentage, mean cell volume, red cell distribution width, alkaline phosphatase, and white blood cell count.¹¹ GrimAge2, a biomarker of lifespan, integrates chronological age, gender, and 10 DNA methylation surrogates for cigarette pack-years and plasma protein markers. These markers include adrenomedullin (ADM), beta-2-microglobulin (B2M), CRP, cystatin C, growth differentiation factor-15 (GDF15), hemoglobin A1c (A1c), leptin, plasminogen activator inhibitor-1 (PAI1), and tissue inhibitor metalloproteinase-1 (TIMP1).⁹ The DNAmTL is a DNA methylation-based estimator of telomere length.¹⁰ DunedinPoAm measures the pace of biological aging associated with morbidity. It was developed by comparing longitudinal changes in 18 biomarkers of organ system function among individuals of the same chronological age, making it a robust indicator of how rapidly aging is occurring.⁶

Statistical Analysis

Because of the NHANES survey design, we used the R “Survey” package to perform generalized linear regression models incorporating participant sample weights created by NHANES for the epigenetic clock subsample.²¹ The associations of veteran status with each epigenetic age measure were evaluated using the svyglm R function accounting for the survey design using several levels of covariate adjustment. Model covariates were identified a priori. Minimally-adjusted models included adjustments for chronological age in years (continuous), chronological age in years squared (continuous), sex (dichotomous: female vs. male), and self-identified ethnicity/race (categorical; Non-Hispanic White, Mexican American, Other Hispanic, Non-Hispanic Black, Other Race).

Given the degree of missingness of covariates for some study participants, we performed our fully-adjusted analyses with imputed missing covariates. Imputation was achieved using the MICE function in R (10 imputations). Estimates from each imputed dataset were pooled using the pool function in R.²² Fully-adjusted models adjusted for estimated leukocyte proportions by using epigenetic age measures from the residuals of regressions of chronological age and estimated leukocyte proportions (B cells, CD4 cells, CD8 cells, NK cells, monocytes, and neutrophils) on epigenetic age. Fully-adjusted models also included additional adjustments for other demographic and health variables including: education (categorical; less than high school, high school diploma or general educational development test, greater than high school education, n = 1 missing), occupation (categorical; white-collar and professional work, white-collar and semi-routine work, blue-collar and high-skill work, blue-collar and semi-routine work, or no work, n = 136 missing), poverty-to-income ratio (continuous, n = 266 missing), alcohol intake (categorical; abstainer, moderate drinker, heavy drinker, n = 117 missing), body mass index (BMI [kg/m²]) (continuous, n = 84 missing), general health condition (categorical; good, fair, poor, n = 3 missing), smoking status (categorical; never, former, current, n = 5 missing), and physical activity (dichotomous; moderate/vigorous activity in the last 30 days: yes vs. no, n = 1 missing).

We repeated the fully-adjusted analyses in a complete case sensitivity analysis with only participants with all covariates (n = 1829). Given our main epigenetic age results, we next used the same leveled covariate adjustment strategy to evaluate associations of veteran status with the DNAm-predicted blood biomarker components comprising GrimAge2: A1c, ADM, B2M, CRP, cystatin C, GDF15, leptin, smoking pack-years, PAI1, and TIMP1. Lastly, given a previously published study that suggested that veteran mortality increases specifically in the 55-74 year age window,²³ we performed a sensitivity analysis stratified by this age range. All statistical analyses were performed using R Version 4.4.1 (R Core Team, Vienna, Austria), and P-values <.05 were considered statistically significant. P-values <.10 are discussed as marginal.

RESULTS

Study Sample Characteristics

Supplementary Table S1 describes the study sample characteristics before the application of survey weights. A total of 1829 of the participants (78%) had all covariate data. Participants missing covariates were generally chronologically and epigenetically older than participants with all covariates. Nevertheless, apart from GrimAge2 and DunedinPoAm, these differences were not statistically significant. A greater proportion of participants not missing data were veterans (P = .04). Supplementary Table S2 describes study sample characteristics for complete cases by veteran status before the application of survey weights. Compared to nonveterans, veterans were chronologically older with a mean (SD) age of 67.4 (9.5) years vs. 64.1 (9.0) years (P < .001). Veterans were also epigenetically older than nonveterans with the largest observed difference in GrimAge2 74.3 (8.8) years vs. 70.3 (8.0) years. In all, 24% of complete case study participants were veterans. Most veterans were male (98%), but most nonveterans were female (61%). Among both veterans and nonveterans, most participants were Non-Hispanic White, reported being in good general health. Most veterans had more than a high school education (50%) and were physically active (65%) while 47% of nonveterans had less than a high school education and most nonveterans were not physically active (52%). Most veterans were former smokers (56%) while 49% of nonveterans were never smokers. Other important differences in covariates between veterans and nonveterans are also highlighted in Supplementary Table S2. Figure 1 describes the correlations between the epigenetic biomarkers and chronological age in the entire study sample (n = 2344). We observed strong correlations between epigenetic age and chronological age with the strongest correlation being with SkinBloodAge (r = 0.87, Median Absolute Error = 3.44 years, P < .001). DNA methylation telomere length was negatively correlated with chronological age (r = −0.58, P < .001). Supplementary Table S3 presents study sample characteristics for complete cases by veteran status after the application of survey weights. We observed similar chronological age, epigenetic age, and health variable trends after the application of survey weights.

Figure 1.

Pearson correlations (r) and Median Absolute Error (MAE) of epigenetic age with chronological age. The figure presents the chronological age and epigenetic age correlation coefficients and median absolute errors for the study sample (n = 2344) for HannumAge [A], HorvathAge [B], SkinBloodAge [C], PhenoAge [D], GrimAge? [E], DNAmTL [F], and DunedinPoAm [G]. Values for veterans are in blue and values for nonveterans are in rust.

Epigenetic Age Relationships

Table 1 presents the results from our veteran status and epigenetic aging analyses. In our minimally-adjusted models (adjusted for chronological age, chronological age,² sex, and race/ethnicity) using the entire study sample, veterans had marginally greater SkinBloodAge (β = 0.86 years, 95% CI: −0.10, 1.81, P = .08) and GrimAge2 measures (β = 0.71 years, 95% CI: −0.07, 1.49, P = .07) compared to nonveterans. Both the SkinBloodAge (β = 1.00 years, 95% CI: −0.01, 2.00, P = .05) and GrimAge2 (β = 0.69 years, 95% CI: −0.14, 1.52, P = .09) relationships were marginally significant in fully-adjusted imputed models using the entire study sample. Veteran status had similar relationships with SkinBloodAge (β = 0.75 years, 95% CI: −0.22, 1.71, P = .11) and GrimAge2 (β = 0.64 years, 95% CI: −0.33, 1.60, P = .16) in fully-adjusted complete case models, but the relationships were not marginally or statistically significant in this smaller subsample. Although they did not meet the threshold for marginal or statistical significance, we also observed trends of lower DNAmTL and DunedinPoAm but greater Hannum, Horvath, and PhenoAge acceleration in veterans when compared to nonveterans in all models (Table 1). Supplementary Table S4 presents the results of the age-stratified sensitivity analysis. In participants 55-74 years of age, veterans had a significantly higher SkinBloodAge when compared to nonveterans (β = 1.37 years, 95% CI: 0.04, 2.69, P = .05). The SkinBloodAge relationship in the remainder of study participants was null with an attenuated model estimate (β = 0.61 years, 95% CI: −1.13, 2.35, P = .42).

Table 1.

Relationships of Veteran Status with Epigenetic Aging

Model/measure	Estimate (95% CI)	P-value	n
Minimally-adjusted
HannumAge (years)	0.72 (−0.52, 1.96)	.24	2344
HorvathAge (years)	0.62 (−0.44, 1.69)	.24	2344
SkinBloodAge (years)	0.86 (−0.10, 1.81)	.08	2344
PhenoAge (years)	0.37 (−0.88, 1.61)	.55	2344
GrimAge2 (years)	0.71 (−0.07, 1.49)	.07	2344
DNAmTL (kb)	−0.01 (−0.05, 0.03)	.53	2344
DunedinPoAm	−0.003 (−0.01, 0.01)	.65	2344
Imputed fully-adjusted
HannumAge (years)	0.86 (−0.41, 2.13)	.15	2344
HorvathAge (years)	0.78 (−0.37, 1.93)	.15	2344
SkinBloodAge (years)	1.00 (−0.01, 2.00)	.05	2344
PhenoAge (years)	0.58 (−0.77, 1.92)	.33	2344
GrimAge2 (years)	0.69 (−0.14, 1.52)	.09	2344
DNAmTL (kb)	−0.02 (−0.07, 0.02)	.23	2344
DunedinPoAm	−0.005 (−0.02, 0.01)	.32	2344
Fully-adjusted complete case sensitivity
HannumAge (years)	0.79 (−0.4, 1.98)	.15	1829
HorvathAge (years)	0.79 (−0.29, 1.86)	.12	1829
SkinBloodAge (years)	0.75 (−0.22, 1.71)	.11	1829
PhenoAge (years)	0.62 (−0.64, 1.87)	.27	1829
GrimAge2 (years)	0.64 (−0.33, 1.60)	.16	1829
DNAmTL (kb)	−0.02 (−0.07, 0.03)	.30	1829
DunedinPoAm	−0.01 (−0.02, 0.004)	.19	1829

Model estimates are for veterans (nonveterans are the reference group).

Minimally-adjusted model adjustments: chronological age, chronological age², sex, and race/ethnicity.

Fully-adjusted model adjustments: Minimally-adjusted + alcohol, BMI, education, general health, occupation, physical activity, PIR, smoking, and estimated leukocyte proportions. BMI, body mass index; PIR, poverty-to-income ratio.

P < .05: statistically significant.

P < .10: marginally significant.

GrimAge Component Relationships

Given the marginal GrimAge2 findings in models using the entire study sample, we examined associations of GrimAge2 predicted biomarker components with veteran status. Table 2 describes these results. Being a veteran was associated with higher predicted A1c (β = 0.006, 95% CI: 0.0005, 0.01, P = .03) and TIMP1 levels (β = 71.14, 95% CI: 8.28, 134.01, P = .03) in minimally-adjusted models. The A1c (β = 0.005, 95% CI: −0.0005, 0.01, P = .07) and TIMP1 relationships (β = 96.40, 95% CI: −15.05, 207.85, P = .08) remained marginally significant in a fully-adjusted imputed models. Veterans also had marginally greater levels of cystatin C (β = 3827.72, 95% CI: −76.18, 7731.61, P = .05) and B2M (β = 15,351.93, 95% CI: −2929.33, 33,633.19, P = .08) in imputed fully-adjusted models. Only the TIMP1 relationship (β = 98.66, 95% CI: −25.24, 222.55, P = .099) was marginally significant in the fully-adjusted complete case sensitivity analysis.

Table 2.

Relationships of Veteran Status with GrimAge2 Components

Model/measure	Estimate (95% CI)	P-value	n
Minimally-adjusted
A1c	0.006 (0.0005, 0.01)	.03	2344
ADM	1.14 (−1.14, 3.41)	.31	2344
B2M	12,593.58 (−2628.63, 27,815.78)	.10	2344
CRP	0.06 (−0.01, 0.13)	.08	2344
Cystatin C	3009.40 (−534.21, 6553.01)	.09	2344
GDF15	2.97 (−12.71, 18.65)	.70	2344
Leptin	166.99 (−153.08, 487.05)	.29	2344
Pack-years	0.67 (−1.24, 2.58)	.48	2344
PAI1	46.03 (−378.12, 470.17)	.82	2344
TIMP1	71.14 (8.28, 134.01)	.03	2344
Imputed fully-adjusted
A1c	0.005 (−0.0005, 0.01)	.07	2344
ADM	1.08 (−1.29, 3.45)	.30	2344
B2M	15,351.93 (−2929.33, 33,633.19)	.08	2344
CRP	0.06 (−0.02, 0.14)	.10	2344
Cystatin C	3827.72 (−76.18, 7731.61)	.05	2344
GDF15	4.92 (−17.33, 27.17)	.60	2344
Leptin	136.09 (−266.84, 539.03)	.43	2344
Pack-years	0.15 (−1.72, 2.02)	.85	2344
PAI1	22.73 (−472.09, 517.54)	.91	2344
TIMP1	96.40 (−15.05, 207.85)	.08	2344
Fully-adjusted complete case sensitivity
A1c	0.003 (−0.003, 0.01)	.29	1829
ADM	2.11 (−0.66, 4.88)	.11	1829
B2M	13,593.07 (−6649.6, 33,835.73)	.15	1829
CRP	0.06 (−0.03, 0.14)	.14	1829
Cystatin C	3057.47 (−929.01, 7043.94)	.11	1829
GDF15	−1.10 (−22.47, 20.27)	.90	1829
Leptin	167.03 (−318.93, 653)	.43	1829
Pack-years	0.14 (−1.97, 2.25)	.88	1829
PAI1	−47.89 (−624.6, 528.82)	.85	1829
TIMP1	98.66 (−25.24, 222.55)	.099	1829

Model estimates are for veterans (nonveterans are the reference group).

Minimally-adjusted model adjustments: chronological age, chronological age², sex, and race/ethnicity.

Fully-adjusted model adjustments: Minimally-adjusted + alcohol, BMI, education, general health, occupation, physical activity, PIR, smoking, and estimated leukocyte proportions. BMI, body mass index; PIR, poverty-to-income ratio.

P < .05: statistically significant.

P < .10: marginally significant.

DISCUSSION

We examined relationships of self-reported veteran status with epigenetic aging biomarkers in a representative cross-sectional sample of U.S. adults aged 50-84 years and participating in NHANES. To the best of our knowledge, this is the largest study directly comparing epigenetic aging in veterans and nonveterans. Across models using the entire study sample and adjusted for demographic, health, and lifestyle factors, we observed marginal trends of veterans having greater epigenetic ages in the SkinBloodAge and GrimAge2 biomarkers as well as higher DNAm-predicted blood levels of GrimAge2-component A1c and protein TIMP1 when compared to nonveterans.

Given previous research reporting that veterans had greater all-cause mortality when compared to non-veterans,⁴ we hypothesized that being a veteran would be associated with greater morbidity and mortality risk reflected by greater epigenetic age acceleration. Consistent with our hypothesis, in our minimally-adjusted models, we observed marginal trends of greater epigenetic aging in SkinBloodAge and GrimAge2. SkinBloodAge is one of the most accurate predictors of chronological age and was developed for better performance in skin and blood tissues among other cell types when compared to earlier chronological age estimators like the HorvathAge and HannumAge biomarkers. Still, SkinBloodAge does have reported relationships with age-related phenotypes like hyperlipidemia, inflammation, and insulinemia.²⁰ Our GrimAge2 findings are particularly fitting given that among all the epigenetic age measures examined, GrimAge2 is not only highly associated with age-related health conditions but it is also the measure that most closely predicts mortality across racial and ethnic groups.⁹ Furthermore, with respect to magnitude and temporality, our effect estimates with the 1999-2000 and 2001-2002 NHANES cycle veterans having marginally greater average SkinBloodAge of 1.00 years and GrimAge2 of 0.69 years, are similar to National Center for Veterans Analysis and Statistics findings that expected life-years for veterans from 2000-2014 were 0.8 to 1.2 life-years shorter when compared to the total U.S. population.⁴

To better understand what was driving our observed marginal GrimAge2 relationship, we examined the relationships of being a veteran with the 10 GrimAge2 DNAm-predicted components. In our minimally-adjusted analyses, we observed statistically significant associations with A1c (a measure related to diabetes) and TIMP1 (a protein that primarily functions to inhibit other proteins that are involved in degrading the extracellular matrix). Across our fully-adjusted models, our most robust findings were marginally significant higher levels of TIMP1 in veterans compared to nonveterans. TIMP1 has a role in preventing programmed cell death and conversely promoting cellular proliferation in a variety of cell types and processes like cancer.^24–26 TIMP1 has been previously implicated as a marker of 24 month all-cause mortality and myocardial infarctions in male human veterans²⁷ and a marker of military burn pit exposure in male rat models.²⁸

Although we consider the findings from our study to be novel, many of the results are of marginal statistical significance at best. One possibility for results trending toward the null could be reflective of the mixed findings regarding veterans’ mortality in the broader literature. More specifically, although some studies have associated veteran status with increased mortality,^3–5 other studies report a “healthy soldier effect” (i.e., decreased mortality in veterans) thought to be driven by recruitment screening procedures and access to medical services afforded to veterans.^2,29 One study using Veterans Health Administration data from 2014 to 2020 reported lower overall standard mortality ratios when veterans were compared to the general U.S. population.²³ However, when the authors stratified their data by age groups, they observed that the lower standard mortality ratios were being driven by younger and very old veterans. In fact, veterans 55-74 years of age had greater mortality than the general U.S. population. The authors speculate that some career military members transition into retirement in this age range and that changes in diets and other lifestyle practices may be responsible for this increased mortality.²³ Although we did not observe less mortality in veterans compared to nonveterans in our study sample, we did observe that the SkinBloodAge estimate for participants 55-74 years of age was much larger than the SkinBloodAge estimate for the remaining study participants—similarly suggesting increased morbidity risk in this age cohort. Moreover, this relationship was statistically significant. We did not observe the same relationship or statistical significance for GrimAge2. Future research will be important for understanding this phenomenon. Furthermore, we have limited details on the veterans’ military service. For instance, we do not know which veterans deployed or which veterans may have been National Guard and Reserve members that were never activated. Differences in types of military service and exposures are known to impact mortality ratios,^30–32 and it is possible that if we could compare our data by high risk/exposure and lower risk/exposure groups that we may observe significant differences in veteran epigenetic aging compared to the general population.

Strengths of our study sample include the use of DNA methylation-based biomarkers of aging in a large study sample to compare U.S. veterans and nonveterans. Nonetheless, our study has some limitations. First, our study uses data that are approximately 20 years old at the time of the analysis. Military service and the social landscape in the United States have changed in many ways since that time and this may limit the generalizability of our findings. Unfortunately, this is the most recent available methylation data in NHANES at the time of this analysis. Nonetheless, our study questions are novel and the findings from this nationally-representative study sample can help inform future research in this area. Second, this was a cross-sectional analysis and cannot assess longitudinal relationships that are key for understanding the aging process. Third, although NHANES includes a wide range of demographic and lifestyle variables, some participants were missing some variables subsequently reducing the sample size of the fully-adjusted complete case analyses. Nonetheless, we performed analyses including imputation models and observed similar effect estimates. Fourth, we do not have information on military rank, branch, length of service, or specific timing of service. These and similar variables could offer more accurate assessments of veterans’ exposures and risk profiles, with significant implications for their health and a potential impact on epigenetic aging.

In conclusion, this study reports marginal trends of veteran status with epigenetic aging-related measures (SkinBloodAge, GrimAge2, and DNAm-predicted GrimAge2 protein TIMP1) in U.S. adults aged 50-84 years. Future studies using other more recent, large datasets with more complete demographic data and more comprehensive service data will be helpful for further characterizing veteran status and epigenetic aging relationships. Given that the United States has over 18 million veterans and had 2024 U.S. Department of Veterans Affairs budgetary resources exceeding $400 billion,^16,17 understanding these relationships could carry important implications for population health and health policy.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Military Medicine online.

FUNDING

J.C.N.E. and A.C. are supported by National Institutes of Health grant (R01ES031259). This research was also supported by the National Institute on Minority Health and Health Disparities (R01MD011721, MPI: to B.L.N., and D.H.R.).

CONFLICT OF INTEREST STATEMENT

None declared.

DATA AVAILABILITY

The datasets analyzed in the current study are available from the NHANES website.

INSTITUTIONAL REVIEW BOARD (HUMAN SUBJECTS):

All NHANES participants gave written informed consent, and the study protocols received approval from the NCHS Research Ethics Review Board (protocol #98-12).

INSTITUTIONAL ANIMAL CARE AND USE COMMITTEE. (IACUC)

Not applicable.

INSTITUTIONAL CLEARANCE

Not Applicable. All authors reviewed and approved the final manuscript.

INDIVIDUAL AUTHOR CONTRIBUTION STATEMENT

J.C.N.E. conceived of the analyses, performed data analysis, visualization, original writing. D.K., H.S., N.G., S.D., and B.L.N. contributed to the analysis. N.G., B.L.N., D.K., and H.S. calculated the epigenetic age measures. A.C. and D.H.R. supervised the work and contributed to writing/editing of the manuscript.