What Really Drives Biological Aging? Smoking, Diet, and SES

Biological aging does not move at the same speed for everyone. In the best head-to-head analysis available, smoking towered over other measured lifestyle exposures as a correlate of faster GrimAge acceleration. Diet quality and education moved in the opposite direction. Studies in Taiwan and Hawaii point to the same broader picture: metabolic health and social conditions track with epigenetic aging, but the clocks themselves still need better validation across populations.

That last clause matters. DNA-methylation clocks are useful risk markers, not crystal balls. They measure chemical patterns on DNA that correlate with aging-related outcomes; they do not prove that changing one habit will add or subtract a specified number of healthy years. Still, the studies offer a more useful hierarchy than the usual advice to sleep, move, eat well, and somehow also become a person with no chronic stress.

Smoking dominated the CARDIA lifestyle ranking

A 2022 CARDIA study examined 744 Black and White U.S. adults with a mean age of 45.9 years. The researchers modeled six correlated exposures at once: smoking, alcohol use, physical activity, sleep, diet quality, and education. That approach matters because these behaviors cluster. People do not usually draw smoking, diet, exercise, and education out of separate hats.

Using quantile-based g-computation, a one-quartile shift toward a more adverse overall exposure mix was associated with 5.15 additional years of GrimAge acceleration (95% CI, 4.16 to 5.86; p<0.001). A second method, Bayesian kernel machine regression, estimated 3.62 years (95% credible interval, 1.93 to 5.22). Both analyses pointed in the same direction.

The ranking was dramatically uneven. Smoking carried 83.5% of the harmful-direction weight in the quantile-based model and had a Bayesian posterior inclusion probability of 1.000. Alcohol ranked a distant second, with a weight of 0.129 and a posterior inclusion probability of 0.997. Physical activity and sleep had small harmful-direction weights in this observational analysis. Diet quality and education were independently associated with lower GrimAge acceleration.

For a smoker, this is the most actionable result in the article: cessation has a far stronger evidentiary claim than fine-tuning a supplement stack for improving a smoking-linked aging biomarker. A systematic review of 89 studies also found smoking consistently associated with epigenetic clock acceleration, while former smokers generally showed less acceleration than current smokers. The review supports partial reversibility, but it does not establish a reliable number of GrimAge years regained per year after quitting. A 2026 NHANES analysis reports time-since-cessation patterns for GrimAge and DunedinPACE, but it remains a preprint, so it cannot settle that question.

The smoking result has a built-in advantage

The 83.5% figure should not be read as smoking causing 83.5% of biological aging. CARDIA used GrimAge acceleration, and GrimAge includes DNAmPACKYRS, a methylation-based surrogate for lifetime smoking pack-years. Smoking therefore predicts a clock that partly encodes smoking biology by design.

That does not make the result meaningless. Smoking damages health through many pathways, and its association with worse aging-related outcomes has abundant evidence behind it. But it does mean the exact ranking may exaggerate smoking's independent share relative to a clock without a smoking-specific component, such as DunedinPACE. Replication with several clock algorithms would tell us whether smoking remains this dominant when the outcome does not contain a smoking surrogate.

CARDIA also found a U-shaped alcohol relationship in its Bayesian analysis. That pattern does not prove moderate drinking protects biological age. People who abstain may do so because of existing illness or past heavy drinking, which can make abstainers look less healthy than moderate drinkers. This is a familiar observational trap, not a prescription to pour a preventive glass of wine.

Metabolic markers connect behavior to aging clocks

A 2025 Taiwan Biobank study examined 2,474 Taiwanese East Asian adults and screened 81 candidate lifestyle and physiological predictors of GrimAge acceleration. Its final model identified HbA1c, HDL cholesterol, body mass index, waist-to-hip ratio, uric acid, and gamma-glutamyltransferase as independent predictors.

These are associations from a cross-sectional study, so they cannot show whether metabolic dysfunction speeds epigenetic aging, whether faster biological aging worsens metabolic health, or whether a third factor drives both. Yet the pattern is biologically coherent with CARDIA's results. Smoking can alter liver and lipid-related measures; diet quality relates to glycemic control and adiposity; inactivity contributes to body-size measures. The Taiwan findings do not prove that pathway, but they show that the metabolic correlates of aging clocks also appear in a large East Asian cohort.

Longitudinal evidence gives diet a firmer footing than a single cross-sectional snapshot. A study of 1,041 participants in the Melbourne Collaborative Cohort Study assessed diet, body size, and several epigenetic aging measures at baseline and again more than 10 years later. Higher diet quality and lower body size were associated with slower epigenetic aging. Associations still fall short of proof that a particular diet causes a particular clock change, but they fit the broader metabolic pattern.

Randomized work has begun to test whether changing exposures shifts clocks. The DO-HEALTH trial reported effects of vitamin D, omega-3, and exercise on DNA-methylation clocks in older European adults. Its exercise finding differs from CARDIA's small physical-activity weight, but the studies asked different questions: an intervention in older adults is not equivalent to a self-reported activity measure in a middle-aged observational cohort. The discrepancy is a reason to avoid treating CARDIA's activity ranking as the final word.

Your neighborhood may enter the equation

Individual behavior does not fully explain who ages faster on these measures. A 2024 NIMHD Social Epigenomics Program study examined 376 nonsmoking Japanese American, Native Hawaiian, and White Hawaii residents. It used DunedinPACE, which estimates the pace of aging per chronological year rather than age acceleration expressed in years.

Neighborhood socioeconomic status was associated with DunedinPACE after adjustment for individual education, BMI, diet quality, and physical activity. That result supports the possibility that neighborhood conditions contribute beyond personal behavior, through exposures such as chronic stress, environmental conditions, or access to care. A systematic review of trauma and racial discrimination similarly found evidence linking psychosocial stressors with epigenetic age acceleration.

The Hawaii study also found that physical activity was associated with slower DunedinPACE in Native Hawaiian participants but not in Japanese American or White participants. This is an interesting signal, not a clinical conclusion. With 376 participants divided among three ethnic groups, an interaction analysis has limited precision and a substantial false-positive risk. It needs replication in a much larger prospective cohort.

The nonsmoker-only design creates another interpretive limit. Excluding the exposure that dominated the CARDIA GrimAge analysis makes other variables appear relatively more important. The NIMHD findings therefore illuminate non-smoking pathways; they do not provide a population-wide ranking of all aging drivers.

The clock may be less calibrated than the confidence suggests

Second-generation clocks are more relevant to health than early clocks trained mainly to predict chronological age. GrimAge was trained against mortality-related biomarkers, while DunedinPACE was developed using roughly 20 years of follow-up in the Dunedin birth cohort and 19 measures of organ-system integrity.

But DunedinPACE was developed in approximately 1,000 people born in Dunedin, New Zealand, a predominantly European-ancestry cohort. GrimAge also lacks full validation in many non-European populations. Researchers have not yet quantified whether these clocks show intercept shifts, weaker prediction, or other calibration problems in East Asian and Pacific Islander groups. That uncertainty applies to comparisons across Taiwan, Hawaii, and U.S. cohorts.

Epigenetic clocks can reveal a hierarchy of correlated risks, but they cannot yet provide an ancestry-neutral receipt for every lifestyle choice.

The sensible ranking remains plain. If you smoke, quitting is the lifestyle change with the strongest evidence for improving smoking-linked biological-aging measures, even though the size and timing of recovery remain uncertain. Diet quality and metabolic health deserve attention because several cohorts connect them with slower epigenetic aging. Exercise likely matters for health and may affect clocks under some conditions, but CARDIA's observational ranking does not quantify its causal effect. Finally, social conditions belong in the analysis. A person is not a lone variable in a regression model, however much the regression model would appreciate the simplification.