Using a Biomarker of Biological Aging to Predict Incident Cardiometabolic Disease

Abstract

Background

The ability to predict cardiometabolic risk is essential to support efforts to prevent future onset of disease. Biomarkers that assess biological aging—the rate of decline in physiological functioning that occurs across the systems of the human body—have the potential to support this goal.

Methods

We investigated the extent to which Dunedin Pace of Aging Calculated From the Epigenome (DunedinPACE), an epigenetic measure of biological aging, predicted cardiometabolic disease onset in a cohort of 2062 US military veterans free of disease at baseline and followed for an average of 14.1 years via electronic health records. Other clinical biomarkers included hemoglobin A1C, blood pressure, pulse, and body mass index.

Results

DunedinPACE predicted incident cardiometabolic disease onset (hazard ratio [HR]=1.82 [95% CI, 1.64–2.04]; P<0.001) and remained associated with cardiometabolic disease onset when accounting for hemoglobin A1C, blood pressure, pulse, and body mass index (HR=1.38 [95% CI, 1.23–1.55]; P<0.001). DunedinPACE increased prediction of cardiometabolic disease (area under the receiver operating characteristic [AUROC] curve=0.76, ΔAUROC=0.04) compared with demographics alone and increased prediction when combined with clinical biomarkers, demographics, smoking, and alcohol use (AUROC=0.81, ΔAUROC=0.01, ΔΧ ² = 27.81, df=1, P<0.001).

Conclusions

These results suggest that DunedinPACE improves prediction of cardiometabolic disease compared with other clinical biomarkers and could help identify individuals who would benefit from preventative interventions that delay the onset of cardiometabolic disease.

Nonstandard Abbreviations and Acronyms

DunedinPACE

Dunedin Pace of Aging Calculated From the Epigenome

Clinical Perspective.

What Is New?

• A biomarker of biological aging, Dunedin Pace of Aging Calculated From the Epigenome, improved the prediction of incident cardiometabolic disease compared with using demographics, health behaviors, and clinical laboratory measures (eg, hemoglobin A1C, blood pressure, pulse, and body mass index).

• Improvements extended to the prediction of future onset of cardiovascular disease, diabetes, and chronic kidney disease when assessed independently.

What Are the Clinical Implications?

• Biomarkers of biological aging derived from DNA methylation, such as Dunedin Pace of Aging Calculated From the Epigenome, could aid in determining who might benefit from preventative interventions before the onset of cardiometabolic disease.

The prevalence of cardiometabolic disease—including diabetes, cardiovascular disease (CVD), and chronic kidney disease (CKD)—is expected to rise over the coming decades. This increase will result in substantial human and economic costs. ¹ , ² To address the burden of cardiometabolic diseases, it is critical to identify individuals who might benefit from preventive interventions, ideally using biomarkers that predict future disease risk. Novel biomarkers of aging ³ , ⁴ , ⁵ have been developed to assess the pace of biological aging, defined as the rate of decline in physiological functioning that occurs across the systems of the human body. Although people grow older at the same speed, the rate at which they are biologically aging is a measurable difference that varies among individuals. ⁶ Patterns of DNA methylation (DNAm) that reflect epigenetic signatures consistent with faster or slower aging ⁴ , ⁵ have been shown to predict the new onset of chronic disease ⁷ and could improve the prediction of future cardiometabolic disease. For example, Dunedin Pace of Aging Calculated From the Epigenome (DunedinPACE), an epigenetic measure of aging derived from 20‐year longitudinal change across 19 biomarkers, ⁶ is associated with incident myocardial infarction, kidney disease, and diabetes, with stronger associations than other recently developed epigenetic aging measures. ⁷ As new biomarkers of aging like DunedinPACE are validated, a natural question is whether they have prognostic value compared with clinical markers already used in clinical care—such as body mass index (BMI), blood pressure (BP), pulse, or hemoglobin A1C (HbA1c). To begin to answer this important question, we investigated the clinical utility of DunedinPACE ⁶ and commonly used clinical biomarkers in the prediction of incident cardiometabolic disease. To do so, we used data from 2062 veterans with 14.1 years of electronic health record (EHR) follow‐up, on average.

METHODS

Participants included 2062 veterans in the PDMH (Post Deployment Mental Health) Study ⁸ who had DunedinPACE aging scores, were free of cardiometabolic disease, and had follow‐up EHR data (see Figure S1). The Durham, Richmond, W. G. Bill Hefner Veterans Affairs and Central Virginia Veterans Affairs Health Care Systems' Institutional Review Boards approved the study protocol and all participants provided informed consent. Data and materials from the PDMH Study are available to researchers who request access through the Veterans Integrated Service Network 6 Mental Illness Research, Education, and Clinical Center and follow required data access protocols.

Measures

DNA Methylation Data Quality Control and Normalization Pipeline

Blood was collected during the baseline assessments and analyzed using the Infinium HumanMethylation450 or MethylationEPIC Beadchip (Illumina Inc., San Diego, CA) to derive DNAm data. Internal replicates were included and checked for consistency using single nucleotide polymorphisms on each array. Quality control was performed using the minfi ⁹ and ChAMP ¹⁰ R packages. Samples were excluded if average fluorescence signal intensity was <2000 arbitrary units or <50% of the mean intensity of all samples, >10% of probes were not detectable (P value >0.001), if a sex mismatch was detected, or if the sample was deemed an outlier on principal component analysis plots. Probe quality control and data normalization were performed within each batch using the R package wateRmelon. ¹¹ Probes not detected (P value >0.001) in >10% of samples and those hybridizing to multiple locations in the genome were removed. Raw beta values were normalized using the dasen approach ¹¹ and batch and chip adjustments were completed using ComBat in the R package sva ¹² to produce the resulting normalized beta values.

DunedinPACE Aging Scores

DunedinPACE aging scores ⁶ were calculated by applying the standard code for the DunedinPACE algorithm ¹³ to the DNAm data.

DNAm Covariates

There were 3 sets of technical covariates derived from the DNAm data. First, DNAm data were used to produce estimated white blood cell counts using the Houseman approach. ¹⁴ Second, DNAm data were also used to generate polyepigenetic smoking methylation scores. ¹⁵ , ¹⁶ Third, a binary variable was also included to index DNAm chip type (450k or EPIC v1.0).

Clinical Biomarkers

Baseline HbA1c, BP, pulse, and BMI were ascertained using clinical values present in the EHR in the 2 years before and following enrollment in the PDMH (Data S1).

Cardiometabolic Health

Incident diabetes, CVD, and CKD diagnoses were ascertained using International Classification of Diseases, Ninth Revision (ICD‐9) and Tenth Revision (ICD‐10) codes ¹⁷ in the Veterans Affairs EHR to a censor date of December 31, 2024. Data S1 provides the full details of the EHR data access and cleaning, as well as the chronic disease diagnosis ascertainment.

Demographics

Participants reported their age, sex, race and ethnicity, and education. Sex (men or women), as well as race and ethnicity (non‐Hispanic Black veterans or non‐Hispanic White veterans) were confirmed by genetic data. ⁷

Health Behaviors

Participants reported their smoking (0=never smoker, 1=past smoker, 2=current smoker) and alcohol use was assessed using the Alcohol Use Disorders Identification Test. ¹⁸ These health behaviors were selected due to their availability in the PDMH cohort at baseline.

Statistical Analysis

We conducted a series of analyses to characterize the clinical utility of DunedinPACE in the prediction of future cardiometabolic disease. We first present demographic characteristics of the sample, as well as bivariate correlations between DunedinPACE and common clinical biomarkers (HbA1c, BP, pulse, and BMI) to contextualize their baseline associations. We next examined whether DunedinPACE predicted cardiometabolic disease onset using Cox‐proportional hazard models. We generated predicted probabilities from these models and used these probabilities to conduct area under the receiver operating characteristic (AUROC) curve analyses for the prediction of cardiometabolic disease onset for DunedinPACE, producing statistics that are comparable to a C‐statistic and represent external validation performance estimates. When testing the incremental predictive utility of adding variables, we compared the change in chi‐square model fit (ΔΧ ²) in nested models. We also present results for diabetes, CVD, and CKD onset separately and present stratified analyses to ensure observed effect sizes are similar for men and women, as well as for non‐Hispanic Black and non‐Hispanic White veterans. CIs (95%) were constructed using SEs, with a nonparametric approach used for the AUROC CIs. All models included demographic (age, sex, race and ethnicity, education level) and health behavior covariates (alcohol use, smoking) unless otherwise noted and used 2‐sided significance tests with an alpha of 0.05. When testing results using bootstrapping, all model results were consistent with the results reported in the study. Correlations between time and partial residuals from the Cox‐proportional hazard models indicated the proportional hazard assumptions were met. Models with DunedinPACE also included DNAm technical covariates (white blood cell proportions, DNAm chip type, smoking methylation scores). Three equal groups were created for the figure tertiles based on DunedinPACE values. These groups were used for visualization purposes only; all statistical models used continuously measures DunedinPACE.

RESULTS

The 2062 veterans (78.6% men; 1000 non‐Hispanic Black veterans and 1062 non‐Hispanic White veterans) averaged 37.2 years old at baseline. Over an average of 14.1 years of follow‐up (SD=2.7), 24.7% of veterans (n=510) developed a cardiometabolic disease—17.9% developed diabetes (n=369), 6.8% developed CVD (n=140), and 4.7% developed CKD (n=96), with 4.1% developing multiple conditions (n=86). At baseline, DunedinPACE was significantly correlated with HbA1c (r=0.26), BP (r=0.18), pulse (r=0.26), and BMI (r=0.27, all ps<0.001).

Time‐To‐Event Analyses

In separate models, DunedinPACE and the clinical biomarkers (HbA1c, BP, pulse, and BMI) were independently associated with incident cardiometabolic disease (Table 1). These associations replicated across the individual disease categories, with the exception that pulse was not associated with CKD.

Table 1.

Model Results for Cox‐Proportional Hazard Models Predicting Cardiometabolic Health

	Cardiometabolic disease onset (n onset=510)		Diabetes onset (n onset=369)		Cardiovascular disease onset (n onset=140)		Chronic kidney disease onset (n onset=96)
No.=2062	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Independent associations (with covariates)
DunedinPACE	1.82*	1.64–2.04	1.97*	1.74–2.24	1.64*	1.33–2.02	1.59*	1.22–2.05
HbA1c	1.56*	1.49–1.64	1.71*	1.63–1.80	1.24*	1.10–1.39	1.19†	1.05–1.35
Systolic BP	1.43*	1.31–1.57	1.37*	1.24–1.53	1.68*	1.42–1.98	1.36†	1.12–1.65
Pulse	1.44*	1.32–1.59	1.66*	1.48–1.86	1.25†	1.04–1.49	1.10	0.90–1.35
BMI	1.65*	1.51–1.79	1.91*	1.74–2.11	1.41*	1.19–1.67	1.23‡	1.01–1.52
Associations in combined model
DunedinPACE	1.38*	1.23–1.55	1.33*	1.16–1.53	1.45*	1.16–1.81	1.42‡	1.08–1.87
HbA1c	1.44*	1.37–1.52	1.59*	1.51–1.68	1.10	0.96–1.26	1.17‡	1.01–1.35
Systolic BP	1.20*	1.09–1.32	1.09	0.97–1.22	1.55*	1.30–1.85	1.46*	1.17–1.82
Pulse	1.15†	1.05–1.27	1.27*	1.13–1.42	1.02	0.85–1.23	0.97	0.79–1.21
BMI	1.26*	1.14–1.40	1.48*	1.32–1.66	1.07	0.87–1.30	0.96	0.76–1.21

All models included demographic (age, sex, race and ethnicity, and education) and health behavior covariates (smoking, alcohol use). Models with DunedinPACE also included DNAm technical covariates (white blood cell proportions, DNAm chip type, smoking methylation scores). Combined model includes all measures in the same model with demographic, health behavior, and DNAm covariates. All measures have been z‐scored to allow for a comparison of effect sizes. BMI indicates body mass index; BP, blood pressure; DNAm, DNA methylation; DunedinPACE, Dunedin Pace of Aging Calculated From the Epigenome; HbA1c, hemoglobin A1C; and HR odds ratio.

*P<0.001, †P<0.01, ‡P<0.05.

In a combined model adjusting for all other biomarkers, DunedinPACE continued to predict incident cardiometabolic disease, diabetes, CVD, and CKD (Figure 1).

Figure 1. Survival curve showing cardiometabolic disease onset associated with DunedinPACE, accounting for demographic covariates, health behaviors, and clinical biomarkers.

DunedinPACE aging groups were created by splitting the sample in 3 even groups based on DunedinPACE aging score for visualization purposes only. All study models used DunedinPACE measured continuously. Each period (baseline, 5 y, etc.) includes the participants at risk across the total sample and in each of the 3 groups. DunedinPACE indicates Dunedin Pace of Aging Calculated From the Epigenome.

In stratified analyses, DunedinPACE predicted cardiometabolic disease onset similarly for men and women, as well as for non‐Hispanic Black veterans and non‐Hispanic White veterans (Table S1).

AUROC Curve Results

Using the predicted probabilities from the time‐to‐event analyses reported previously, we modeled AUROC curves for DunedinPACE and the clinical biomarkers (Table 2). DunedinPACE predicted cardiometabolic disease when combined with demographic characteristics and DNAm covariates (AUROC=0.76), improving prediction over demographics, health behavior, and DNAm covariates (ΔΧ ²=112.19, df=1, P<0.001). This level of prediction matched or exceeded the other clinical biomarkers, with the exception of HbA1c (AUROC=0.79). DunedinPACE improved the prediction of incident cardiometabolic disease when added to the other clinical biomarkers, demographics, and DNAm covariates (AUROC=0.81, ΔAUROC=0.01, ΔΧ ²=27.81, df=1, P<0.001).

Table 2.

Model Results for Area Under the Operator Characteristic Curve Analyses

	Cardiometabolic disease onset (n onset=510)		Diabetes onset (n onset=369)		Cardiovascular disease onset (n onset=140)		Chronic kidney disease onset (n onset=96)
No.=1899	AUROC	95% CI	AUROC	95% CI	AUROC	95% CI	AUROC	95% CI
Baseline prediction
Covariates only	0.72	0.70–0.75	0.71	0.69–0.74	0.69	0.64–0.74	0.74	0.69–0.79
Biomarker plus covariates
DunedinPACE	0.76	0.74–0.79	0.77	0.74–0.79	0.73	0.69–0.77	0.77	0.72–0.81
HbA1c	0.79	0.77–0.81	0.82	0.80–0.85	0.70	0.65–0.75	0.74	0.70–0.79
Systolic BP	0.74	0.72–0.77	0.74	0.71–0.77	0.72	0.68–0.76	0.75	0.70–0.80
Pulse	0.73	0.71–0.76	0.74	0.71–0.77	0.70	0.65–0.74	0.74	0.69–0.79
BMI	0.76	0.74–0.78	0.78	0.73–0.78	0.71	0.67–0.75	0.74	0.69–0.79
Combined measures
Clinical biomarkers	0.80	0.78–0.83	0.85	0.83–0.87	0.73	0.69–0.77	0.75	0.70–0.80
Adding DunedinPACE	0.81	0.79–0.84	0.85	0.83–0.87	0.76	0.71–0.80	0.77	0.72–0.82

All models included demographic (age, sex, race and ethnicity, and education) and health behavior covariates (smoking, alcohol use). Models with DunedinPACE also included DNAm technical covariates (white blood cell proportions, DNAm chip type, smoking methylation scores). Combined measures indicate the combination of demographics and all clinical biomarkers (“Clinical biomarkers”) and the combination of demographics and all clinical biomarkers with DunedinPACE. The AUROC is derived from the predicted probabilities from the Cox‐proportional hazard models and can be interpreted as a C‐statistic. AUROC indicates area under the receiver operating characteristic curve; BMI, body mass index; BP, blood pressure; DunedinPACE, Dunedin Pace of Aging Calculated From the Epigenome; and HbA1c, hemoglobin A1C.

DISCUSSION

Accelerated biological aging, assessed by DunedinPACE, predicted incident cardiometabolic disease in a cohort of 2062 veterans. Each SD increase in DunedinPACE corresponded to an 82% increased risk for cardiometabolic disease (95% CI, 64%–104%), equivalent to the risk associated with a 0.7% higher HbA1c or a 5.6 unit (kg/m²) increase in BMI at baseline. For individual disease categories, each SD increase in DunedinPACE was associated with a 97% increased risk for diabetes (95% CI, 74%–124%), 64% increased risk for CVD (95% CI, 33%–102%), and 59% increased risk for CKD (95% CI, 22%–105%). These associations adjusted for demographics, smoking, alcohol use, and DNAm covariates. In AUROC curve analyses, DunedinPACE improved prediction when added to other clinical biomarkers (HbA1c, BP, pulse, BMI), and results largely replicated for diabetes, CVD, and CKD, with the largest improvements in prediction for CVD and CKD compared with diabetes.

Our results suggest that DunedinPACE could be useful in assessing risk for cardiometabolic disease when combined with other common clinical biomarkers, such as HbA1c. The prediction of future cardiometabolic disease (AUROC=0.81) reached levels consistent with, or greater than, widely used risk measures, such as the Framingham Risk Score (AUROC=0.76–0.79). ¹⁹ Similarly, the identification of diabetes risk using DunedinPACE and clinical biomarkers (AUROC=0.85) compared well with a recently developed DNAm risk score trained to predict 10‐year diabetes onset when applied in a validation cohort (AUROC=0.82). ²⁰ As DNAm technology becomes more cost effective and reliable, DunedinPACE could provide an intuitive measure of cardiometabolic disease risk that complements other clinical biomarkers available to clinicians and patients, such as HbA1c and BMI.

These findings suggest the pace of biological aging, as measured by DunedinPACE, could be a useful physiological mechanism to understand whether exposures might be associated with future diabetes or CVD onset. For example, prior studies have found posttraumatic stress disorder is associated with faster rates of biological aging, ²¹ which might explain increased risk for CVD among these individuals. DunedinPACE could be as a useful surrogate health outcome for intervention trials seeking to prevent the onset of cardiometabolic diseases, reducing the large samples and long time frames necessary to observe clinical outcomes.

Our study had limitations relevant to interpreting the results. First, the clinical biomarkers were assessed at baseline. Testing the longitudinal dynamics of DunedinPACE, clinical biomarkers, and disease onset would be an important future direction, as cardiometabolic traits might influence rates of aging. Second, our diagnoses can only capture Veterans Affairs encounters or referrals to community care and cannot ascertain diagnoses outside those sources. Third, it is not clear to what extent results would generalize to other groups of veterans from earlier service eras (eg, Vietnam War era) or nonveteran populations, which would require additional validation. Fourth, the sample averaged 37.2 years old at baseline and 51.3 years old at follow‐up. Cardiometabolic diseases can onset at older ages and results might vary based on the age at which individuals are assessed. Finally, our analyses did not account for medication use that might affect the onset of cardiometabolic diseases. Future research should determine to what extent medication use might impact the observed results.

CONCLUSIONS

In 2062 veterans free of cardiometabolic disease at baseline, accelerated biological aging predicted the future onset of cardiometabolic disease while accounting for other clinical biomarkers. These findings suggest DunedinPACE could help identify individuals who would benefit from prevention efforts that may reduce, delay, or prevent the onset of cardiometabolic diseases.

Sources of Funding

This research was supported by Award #IK2CX002694 to Dr Bourassa and a Senior Research Career Scientist Award to Dr Beckham (#lK6BX003777) from the Clinical Science and Research and Development Service of Veterans Affairs Office of Research and Development, as well as a Research Career Scientist Award to Dr Kimbrel (#IK6BX006523) from the Biomedical Laboratory Research & Development Service. Dr Serier was supported by an award from the National Institute of Diabetes and Digestive and Kidney Diseases (K08DK138296). Dr Crowley acknowledges funding from the National Institutes of Health (1R01NR019594‐01), Veterans Affairs Quality Enhancement Research Initiative (QUE 20‐012), and the Veterans Affairs Office of Rural Health (NOMAD 3480), as well as in‐kind support from the Durham Center of Innovation to Accelerate Discovery and Practice Transformation (VA CIN 13‐410) within the Durham Veterans Affairs Health Care System.

Disclosures

None.

Supporting information

Data S1. Supplemental Methods.

Table S1

Figure S1

STROBE Checklist

Veterans Affairs Mid‐Atlantic MIRECC Workgroup

The Veterans Affairs Mid‐Atlantic MIRECC Workgroup includes (affiliation for all authors: Veterans Integrated Service Network 6 Mid‐Atlantic Mental Illness Research, Education and Clinical Center, Department of Veterans Affairs):

Patrick S. Calhoun, PhD, Eric Dedert, PhD, Eric B. Elbogen, PhD, Robin A. Hurley, MD, Jason D. Kilts, PhD, Angela Kirby, MS, Scott D. McDonald, PhD, Sarah L. Martindale, Ph.D, Christine E. Marx, MD, MS, Scott D. Moore, MD, PhD, Rajendra A. Morey, MD, MS, Jared A. Rowland, PhD, Robert D. Shura, PsyD, Cindy Swinkels, PhD, H. Ryan Wagner, PhD.