How pregnancy history and unhealthy lifestyle relate to faster biological aging

The study used data from the UK Biobank (UKB), a large-scale, population-based prospective cohort that recruited over 500,000 participants aged 39 to 71 years from 22 assessment centers across England, Scotland, and Wales between 2006 and 2010. At baseline, UKB collected extensive sociodemographic, lifestyle, and health-related data through self-administered questionnaires, in-person interviews, physical measurements, and biological sample collection. Additionally, UKB linked participants' health records to national databases, including hospital admissions and mortality data, enabling long-term follow-up. Further details on the UKB study design and population have been described elsewhere (18).

From the initial UKB cohort of 502,230 participants, a total of 273,198 female participants were included in this study. We sequentially excluded 487 individuals with missing information on pregnancy history, 75,899 individuals with missing information on lifestyle factors that were used to generate an unhealthy lifestyle score, and 59,594 individuals without available clinical biomarkers for biological age estimation, finally 137,218 individuals were included in the study (Figure 1).

Pregnancy history was assessed using self-reported reproductive data collected through questionnaires administered at baseline. Participants provided information on age at first and last live birth, history and number of spontaneous or induced abortions, and stillbirths. Additional details on gestational complications were recorded, including gestational diabetes mellitus, gestational hypertension, eclampsia, hydatidiform mole, gestational edema and proteinuria, ectopic pregnancy, venous complications in pregnancy, hemorrhage in early pregnancy, malnutrition during pregnancy. Participants were categorized as either never pregnant or ever pregnant according to their reproductive information.

Given the interrelated nature of multiple lifestyle factors and their association with accelerated aging (15–17), an unhealthy lifestyle score was generated based on cigarette smoking, alcohol consumption, physical activity, television viewing/sedentary time, sleep duration, fruit and vegetable consumption, oily fish consumption, and red and processed meat consumption, according to a previous Australian study (19). At baseline, participants completed a self-administered touchscreen questionnaire on lifestyle behaviors. Each unhealthy behavior was considered as “at risk” and assigned 1 point: current smoking, daily or almost-daily alcohol consumption, insufficient physical activity (< 150 min/week of moderate-intensity or < 75 min/week of vigorous-intensity exercise), excessive television viewing/sedentary time (≥4 hours/day), suboptimal or excessive sleep duration (< 7 or >9 hours/day), low fruit and vegetable intake (< 400 g/day), insufficient oily fish intake (< 1 portion/week), excessive red meat intake (>3 portions/week), and high processed meat intake (>1 portion/week). The final unweighted unhealthy lifestyle score ranged from 0 (healthiest) to 9 (unhealthiest). For analysis, participants were categorized into three groups: most healthy (score 0–2), moderately healthy (score 3–5), and least healthy (score 6–9). Detailed definitions of the unhealthy lifestyle score and its categories are provided in Supplementary Table S1.

Biological age (BA) was estimated using two widely validated algorithms, the Klemera-Doubal Method (KDM) and PhenoAge, which incorporate clinical biomarkers and physical measurements to predict age-related health outcomes (20, 21). Briefly, KDM BA was derived through forced expiratory volume in 1 s, systolic blood pressure, and seven clinical biomarkers: total cholesterol, glycated hemoglobin, blood urea nitrogen, albumin, creatinine, C-reactive protein, and alkaline phosphatase. PhenoAge was calculated using nine clinical biomarkers: albumin, alkaline phosphatase, creatinine, C-reactive protein, glucose, mean cell volume, red blood cell distribution width, white blood cell count, and lymphocyte ratio. The distribution of these biomarkers was provided in Supplementary Table S2, and BA values were computed using the R package “BioAge” (https://github.com/dayoonkwon/BioAge↗).

KDM BA estimates an individual's physiological age relative to their chronological age using a biomarker-based regression approach applied to a reference population. The equation for calculating KDM BA incorporates multiple regression lines, each representing the relationship between a specific biomarker and age. The formula is:

where x is the value of biomarker i for the individual, and k_i, q_i, and s_i are the estimated parameters from the regression model for each biomarker, including the intercept, slope, and the root mean squared error, respectively. The s_BA is a scaling factor accounting for the variance in chronological age explained by the biomarker set, determined from the reference population.

The PhenoAge algorithm, developed using elastic-net Gompertz regression, predict biological age based on multivariate analysis of mortality risk using a set of biomarkers. Originally derived from the NHANES III dataset based on 42 biomarkers from, its formula is:

where xb represents the linear combination of biomarkers from the regression model, and γ is an ancillary parameter estimated from the data.

To evaluate individual variations in BA, we regressed BA against chronological age at biomarker measurement and used the residuals as the indicator of biological aging, referred to as BA acceleration. To facilitate comparison between biological aging measures, BA acceleration was standardized to a mean of 0 and a standard deviation (SD) of 1.

Chronological age was calculated using participants' birthdates and the date of their baseline assessment. Race and ethnicity was self-reported as white, mixed, Asian or Asian British, black or black British, Chinese, or other ethnic group; for analysis, due to small sample sizes in non-White ethnic subgroups, participants were classified into White and non-White groups. Educational level was determined based on the highest self-reported qualification and categorized as high (college or university degree), intermediate (A/AS levels, O-levels, General Certificate of Secondary Education, or equivalent, corresponding to grades 6–12 in the US school system), and low (no formal qualifications) (22). Socioeconomic status was assessed using the Townsend Deprivation Index (TDI), derived from national census data on car ownership, household overcrowding, home ownership, and unemployment aggregated at the postcode level (23). Higher TDI scores indicate greater socioeconomic deprivation. The Charlson Comorbidity Index (CCI) was calculated using ICD-9 and ICD-10 diagnosis codes extracted from hospital administrative data to quantify comorbidity burden. The CCI scores were derived following the original method described by Charlson et al. (24), with weighted scores reflecting comorbidity severity. Body mass index (BMI) was calculated as weight in kilogram divided by squared of height in meter, measured by trained staffs at baseline. BMI was classified according to WHO criteria as underweight (< 18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), and obese (≥30.0 kg/m²).

Descriptive statistics were used to summarize baseline characteristics, with continuous variables presented as means ± standard deviations (SD) and categorical variables as frequencies (percentages).

General linear regression models were used to examine the associations between pregnancy history and both BA and BA acceleration, and regression coefficients with their corresponding 95% confidence intervals (95% CI) were reported from models. For joint effect analyses, participants were classified into six groups based on pregnancy history and unhealthy lifestyle score categories, with participants who adhered to the healthiest lifestyle and had never been pregnant serving as the reference group. Interaction between pregnancy history and unhealthy lifestyle score was assessed exploratory analyses using likelihood ratio tests, by comparing models with and without cross-product terms. The independent associations of individual lifestyle factors (categorized as at risk vs. not at risk) with BA and BA acceleration was also assessed. A sequence of regression models was performed: (1) adjusted for socioeconomic variables, including chronological age, race and ethnicity, educational level, BMI, and TDI; (2) further adjusted for CCI; (3) further adjusted for unhealthy lifestyle score. Missing data for covariates were not imputed but coded as “unknown,” which was classified as a separate category in regression models.

Several sensitivity analyses were performed to assess the robustness of the findings. First, stratified analyses were conducted by chronological age (39–59 and 60–71 years). Second, the analyses were repeated without adjusting for chronological age to evaluate its potential effect. Third, the primary analysis was repeated using a complete-case sample of 136 058 participants. Fourth, three reproductive factors reflecting pregnancy history—including the number of live births, stillbirths, and miscarriages, were individually included in the models to examine their specific associations with biological age acceleration. Fifth, baseline characteristics were compared between participants excluded due to missing clinical biomarkers data and the total population. Sixth, Finally, additional adjustments were made for gestational diabetes mellitus, hypertensive disorders of pregnancy, and hormone replacement therapy use.

All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA) and R 4.3.1 (R Project for Statistical Computing). A two-tailed P < 0.05 was considered statistically significant.

Of 137,218 female participants, the mean chronological age at baseline was 55.73 ± 8.01 years, and 115,675 reported a history of at least one pregnancy (Table 1). Compared to women who had never been pregnant, those with a history of pregnancy had a lower TDI (−1.61 ± 2.89 vs. −0.98 ± 3.07) and unhealthy lifestyle score (2.03 ± 1.33 vs. 2.04 ± 1.33), had a higher CCI (0.86 ± 1.54 vs. 0.79 ± 1.50), and prevalence of diabetes (6.1% vs. 5.4%), hypertension (9.6% vs. 7.4%), and overweight and obese (58.3% vs. 53.8%). In addition, women with a history of pregnancy were more likely to be current smokers (8.3% vs. 7.8%), but less likely to engage in excessive alcohol consumption at baseline (17.0% vs. 19.0%) (Table 1).

In crude analysis, the mean KDM BA, PhenoAge, KDM BA acceleration, and PhenoAge acceleration were 53.64 ± 11.86 years, 48.63 ± 8.75 years, 0.09 ± 0.89 years, and −0.19 ± 0.99 years, respectively. Compared to women who had never been pregnant, those with a history of pregnancy had a higher mean KDM BA (54.01 ± 11.82 vs. 51.66 ± 11.91 years) and PhenoAge (48.96 ± 8.71 vs. 46.85 ± 8.76 years). However, lower values were observed for KDM BA acceleration (0.09 ± 0.89 vs. 0.09 ± 0.9 years) and PhenoAge acceleration (−0.20±0.98 vs. −0.11 ± 1.00 years) (Figure 1).

After adjusting for potential confounders as in model 1, women with a history of pregnancy had a decrease in BA [β = −0.317 (95% CI, −0.438 to −0.196) for KDM BA; β = −0.249 (95% CI, −0.309 to −0.189) for PhenoAge], and BA acceleration [β = −0.019 (95% CI, −0.031 to −0.006) for KDM BA acceleration; β = −0.085 (95% CI, −0.099 to −0.071) for PhenoAge acceleration] (Table 2). These associations remained largely unchanged after further adjustment for CCI and unhealthy lifestyle score in models 2 and 3 [β = −0.312 (95% CI, −0.432 to −0.191) for KDM BA; β = −0.242 (95% CI, −0.302 to −0.183) for PhenoAge], and BA acceleration [β = −0.021 (95% CI, −0.034 to −0.009) for KDM BA acceleration; β = −0.088 (95% CI, −0.102 to −0.075) for PhenoAge acceleration].

In the fully adjusted model examining the joint effect, compared to women who had never been pregnant and with the most healthy lifestyle, those who had never been pregnant and with the least healthy lifestyle had the highest BA [β = 3.396 (95% CI, 2.341 to 4.452) for KDM BA; β = 2.004 (95% CI, 1.772 to 2.237) for PhenoAge] and BA acceleration [β = 0.340 (95% CI, 0.232 to 0.449) for KDM BA acceleration; β = 0.467 (95% CI, 0.412 to 0.522) for PhenoAge acceleration] (Table 3). However, no significant interaction was found between pregnancy history and unhealthy lifestyle score on BA acceleration (P_interaction > 0.05). For example, in the joint analysis of pregnancy history and smoking status, women who had never been pregnant but were classified as “at risk” smokers showed a higher BA acceleration compared with the reference group of women who had never been pregnant and were not at risk (β = 0.39, 95% CI: 0.34 to 0.43). Similarly, among women with a history of pregnancy, those with at-risk smoking status also showed increased BA acceleration (β = 0.32, 95% CI: 0.29 to 0.34), whereas women with a history of pregnancy but not at-risk smoking status had a lower BA acceleration (β = −0.10, 95% CI: −0.11 to −0.08) (Figure 2). Additionally, analyses adjusting for individual lifestyle factors showed that women classified as “at risk” due to smoking, prolonged TV viewing/sedentary time, inadequate/excessive sleep, insufficient fruit and vegetable intake, and had excessive processed meat intake an increase in BA and BA acceleration compared to those classified as “not at risk” (Supplementary Table S3).

The lifestyle score did not mediate the association of pregnancy history with BA and BA acceleration (Figure 3). For example, for KDM BA, a moderate total effect [β = −0.294 (95% CI, −0.415 to −0.173)] and direct effect [β = −0.298 (95% CI, −0.419 to −0.177)] were observed, while the indirect effect was non-significant [β = −0.002 (95% CI, −0.003 to 0.008)], with a mediation proportion of −1.26% (95% CI, −4.04 to 1.52). Similarly, for KDM BA acceleration, a smaller total effect [β = −0.030 (95% CI, −0.043 to −0.018)] and direct effect [β = −0.031 (95% CI, −0.043 to −0.018)] were found, with a non-significant indirect effect [β = −0.0003 (95% CI, −0.0004 to 0.001)], yielding a mediation proportion of −1.81% (95% CI, −5.88 to 2.26).

A series of additional analyses were conducted to evaluate the robustness of the findings. First, stratified analyses showed significant interactions between chronological age and pregnancy history on BA and BA acceleration (Supplementary Table S4). For example, the association between pregnancy history and KDM BA were more pronounced in women aged 39–59 years [β = −0.381 (95% CI, −0.524 to −0.238)] than in those aged 60–71 years [β = −0.134 (95% CI, −0.356 to −0.087); P for interaction < 0.001]. Second, when analyses were repeated without adjusting for chronological age, the results changed substantially, as women who with a history of pregnancy had an increase in KDM BA (β = 1.902, 95% CI, 1.738 to 2.066) and PhenoAge (β = 1.870, 95% CI, 1.748 to 1.992) (Supplementary Table S5). Third, analyses conducted using a complete-case sample yielded similar results (Supplementary Table S6). Fourth, women with a higher number of live births or miscarriages had a decrease in BA or BA acceleration, whereas those with more stillbirths had an increase in BA and BA acceleration (Supplementary Table S7). Fifth, participants excluded due to missing data for clinical biomarkers were more likely to be chronologically older, have a higher educational level, a higher TDI, and to be obese, and less likely to be White, compared with the total population (Supplementary Table S8). Finally, additional adjustment for gestational diabetes mellitus, hypertensive disorders of pregnancy, and hormone replacement therapy use resulted in some changes in the magnitude of the associations, while the direction and statistical significance remained consistent (Supplementary Table S9).

Prior prospective and retrospective studies have yielded conflicting results on the relationship between pregnancy and clinical biomarker-based BA (7, 10–12, 14). A cohort study of Swedish pregnant women reported a progressive increase in PhenoAge acceleration from pre-pregnancy to a few days postpartum (11, 12). However, a cross-sectional study of young Filipino women found no significant relationship between gravidity and PhenoAge acceleration (7). Additionally, a study of American women observed no relationship between parity and KDM-BA acceleration in premenopausal women but found an association in postmenopausal women (14). Meanwhile, several studies have suggested that adherence to more health-promoting factors was associated with slower KDM BA and PhenoAge acceleration (25, 26). Our findings suggested that pregnancy was negatively associated with BA and BA acceleration, even after adjusting for sociodemographic, clinical, and lifestyle factors. In addition, a higher unhealthy lifestyle score or engagement in most individual unhealthy lifestyle behaviors was associated with increased BA and BA acceleration. However, no significant interaction or mediation effect was observed between pregnancy history and unhealthy lifestyle score on BA and BA acceleration. These null findings should be interpreted descriptively rather than as evidence against potential biological interplay. Several factors may explain these inconsistent findings. Differences in study design, particularly longitudinal vs. cross-sectional approaches, may capture distinct phases of biological aging. In addition, variation in age, reproductive stage, and parity across study populations may influence observed associations. Finally, the use of different biological age measures and biomarker sets may further contribute to heterogeneity across studies.

The observed associations between pregnancy history and lower BA may reflect hormonal, immunological, and metabolic adaptations induced by pregnancy; however, given the cross-sectional design, these mechanistic interpretations were descriptive and should be considered hypothesis-generating. Pregnancy triggers significant hormonal shifts, notably increased estrogen and progesterone levels, which exert protective effects on cardiovascular, metabolic, and immune systems (27). Estrogen enhances endothelial function, reduces oxidative stress, and promotes vasodilation (28), while also providing neuroprotection against cognitive decline (29). Progesterone contributes by modulating immune responses and reducing systemic inflammation, a key driver of biological aging (30). Additionally, long-term immune adaptations, including enhanced regulatory T-cell activity and cytokine modulation, may lower chronic inflammation and decelerate biological aging (31). Pregnancy might induce pronounced metabolic adaptations. Although transient insulin resistance occurs during late gestation, evidence indicates that improvements in insulin sensitivity, lipid metabolism, and energy homeostasis may persist postpartum, potentially modulating subsequent biological aging (32). Pregnancy-related epigenetic modifications in immune cells could further enhance immune resilience, influencing the aging process (33). Another plausible mechanism is fetal microchimerism, where fetal cells persist in maternal circulation and various tissues for decades postpartum (34). Studies suggested these fetal cells contribute to maternal tissue repair and regeneration, potentially mitigating age-related deterioration (35). Moreover, pregnancy-induced metabolic adaptations, such as improved insulin sensitivity, lipid metabolism, and energy homeostasis (36, 37), may further slow biological aging long-term metabolic disease risk. Further research is warranted to elucidate the biological pathways linking pregnancy to BA modulation.

The large sample size of a well-characterized population increased statistical power to identify interactions through multiplicative interaction terms in regression models and stratified analyses. Additionally, the objective assessment of BA using clinical biomarkers, combined with comprehensive adjustments for potential confounding factors, enhanced the validity of our findings. Beyond its large sample size and use of multiple biological aging metrics, a key strength of this study is its life-course–oriented analytical framework. By jointly examining pregnancy history and lifestyle behaviors, we move beyond isolated risk factor analyses to explore how distinct life stages contribute to biological aging. This study has several limitations. First, the pregnancy history and lifestyle behaviors were self-reported, which may lead to some misclassification. Although these self-reported data used in our study have not been validated directly, previous studies have suggested that such data is highly reliable (38, 39). Second, the cross-sectional design limited causal inference because biological age was assessed only at baseline. Given the relatively older age of the cohort, pregnancy history likely reflects long-term reproductive exposure rather than recent physiological changes, and the biomarkers used primarily capture chronic biological processes. Mediation analyses were exploratory, and the temporal ordering between pregnancy, lifestyle factors, and biological aging was uncertain; these results should therefore be interpreted cautiously. Future longitudinal studies with detailed reproductive timing and repeated biomarker assessments are needed to clarify temporal relationships and life-course effects. Third, the study population was volunteer-based and not fully representative of the UK population, which may limit the generalization of the results to other racial or ethnic groups. Fourth, the lack of information on time since pregnancy restricted our ability to assess time-dependent associations with biological aging. Fifth, despite the large sample size, power to detect multiplicative interactions between pregnancy history and unhealthy lifestyle may be limited. Thus, the absence of significant interactions could reflect either a true lack of effect or undetected smaller effects, warranting confirmation in larger studies or with alternative analytic approaches. Sixth, although we used two well-established biological aging measures based on clinical biochemistry biomarkers, other aging measures, such as those derived from proteomic or metabolomic signatures, were not assessed. Future research incorporating these measures may provide further understanding of the complex relationship between pregnancy and aging. In addition, unmeasured psychosocial and environmental factors may contribute to residual confounding and warrant cautious interpretation. Although BA was not a direct clinical outcome, it represented a validated integrative biomarker that captures multisystem physiological aging and was associated with cardiometabolic disease, functional decline, and mortality (40). BA acceleration might therefore serve as an early indicator of long-term disease risk and healthspan. Thereforce, these findings supported a life-course approach that integrated reproductive health and lifestyle modification to promote healthy aging and might contribute to efforts to reduce health inequalities among women.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ukbiobank.ac.uk/↗.