Estimation of biological age and age-related outcomes with easily accessible parameters in Chinese

Abstract

Biological aging exhibits significant heterogeneity across individuals, and a precise evaluation at scale is urgently needed. Here, we developed a PCAge, based on common clinical, physiological, and laboratory indices routinely collected in primary healthcare, in the CHARLS. PCAge demonstrated strong correlations with chronological age (r = 0.86–0.88, P < 0.001) and robust performance in the prediction of mortality (C-index = 0.798) over a 10-year follow-up. Biologically older individuals (PCAge > chronological age) suffered from substantially higher risk of age-related diseases, including cardiovascular disease (HR = 1.30, P < 0.001), heart disease (HR = 1.35, P = 0.003), stroke (HR = 2.38, P < 0.001), hypertension (HR = 1.28, P = 0.007), and diabetes (HR = 1.51, P < 0.001). Furthermore, the generalizability of PCAge was validated in the South China Cohort (SCC, n = 68,920). Biologically older individuals were more likely to have hypertension, diabetes, cardiovascular disease, and respiratory diseases. Being female (proportion ratios [PR] = 1.94, P < 0.001), lower education attainment (PR = 1.18, P < 0.001), higher income (PR = 1.47, P < 0.001), and unfavorable lifestyles (PR = 1.41, P < 0.001) were associated with a higher probability of having accelerated aging. PCAge identified aging trajectories up to a decade before clinical disease onset, offering a cost-effective tool for population-level aging surveillance. Our findings underscore the potential of PCAge as a highly accessible tool for the evaluation of aging status, especially in resource-limited areas.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1007/s11357-025-01940-y.

Introduction

Population aging is a defining global trend of our time, imposing a heavy burden on all societies. In China, this demographic shift is even more pronounced, driven by declining birth rates and increasing life expectancy. By 2035, over 30% of China’s population is expected to be over 60 years, posing significant challenges to both healthcare systems and socioeconomic stability [1]. Of note, living longer does not necessarily mean healthy aging. Aging is associated with a progressive loss of physical, mental, and cognitive function, leading to increased vulnerability to morbidity and mortality [2, 3]. Given the large population size and unparalleled speed of aging, it is therefore crucial to evaluate the status of biological aging in the population and identify those with accelerated aging in time to introduce appropriate interventions.

Compared to chronological age (CA), biological age (BA), which reflects the cumulative physiological deterioration over the lifetime at the molecular and cellular level, is more accurate for the prediction of morbidity and aging-related outcomes [4]. Moreover, BA is amenable to various interventions, such as drugs and lifestyle changes [5], offering an opportunity to boost health. In the past few decades, a variety of BA estimators have been developed, primarily based on complex and costly biomarkers, such as telomere length, DNA methylation [6, 7], multi-omics data [8], or unconventional clinical parameters [9, 10]. Despite a high accuracy of these models to detect subtle differences in BA even in young individuals, the high cost and dependence on advanced laboratory infrastructure and specialized expertise substantially hinder the large-scale application, especially in less-developed areas, which in turn further exacerbates health inequalities. In light of an unprecedented scale and pace of aging alongside widening socioeconomic inequalities across China, a BA model integrating routine clinical biomarkers and physical characteristics, which can be easily conducted in primary healthcare centers, is an urgent need to enable an early identification of age-related risks and individuals with accelerated aging.

To address this gap, a clinical aging clock (PCAge) based on primary healthcare measurements was developed in the China Health and Retirement Longitudinal Study (CHARLS) [11] and then evaluated in the South China Cohort (SCC) [12]. Moreover, the overall status of aging across Southern China and individuals with a higher probability of accelerated aging were identified by PCAge.

Methods

Study design and population

This study was based on the CHARLS cohort [11] and the SCC cohort [12]. CHARLS involved a series of cross-sectional surveys, including demographic, socioeconomic, health-related questionnaires, medical and physiological measurements, and laboratory tests. In the present study, a nationally representative sample of 17,705 participants recruited from 28 provinces in 2011 (wave 1), via multistage probability sampling, was used as the discovery cohort. The follow-up surveys were conducted in 2013 (wave 2), 2015 (wave 3), 2018 (wave 4), and 2020 (wave 5) with questionnaire interviews and anthropometric measurements. To ensure robust model development and evaluation, we started with 17,705 subjects enrolled in 2011 (wave 1). First, we removed 5858 participants without blood samples for laboratory tests. Next, 650 participants with a high rate of missing data (> 30% among 41 clinical biomarkers) were excluded (Supplementary Tables 1 and 2). Finally, a total of 11,197 participants were included in this study and then randomly partitioned into a training set (n = 7838, 70% of the dataset) and a testing set (n = 3359, 30% of the dataset). A fixed random seed (seed = [235]) was applied during the splitting process to guarantee reproducibility and consistency across analyses.

The SCC cohort was established in 13 cities spanning four provinces in Southern China [12]. The baseline survey recruited 116,520 participants between March 2018 and December 2020. Face-to-face interviews were conducted by the trained staff to collect data on sociodemographic information, lifestyles, health-related information, and blood samples. A total of 13,588 subjects with no response to the face-to-face questionnaire, 20,207 subjects without a blood sample, 1624 subjects with missing lifestyle information, and 12,181 participants with a high rate of missing data (> 30% among 39 clinical biomarkers) were excluded from the current analysis (Supplementary Tables 1 and2). Finally, a total of 68,920 individuals from the SCC were subjected to the calculation of PCAge and evaluation of age-related outcomes.

Outcome definition in CHARLS and SCC

For the CHARLS, the primary outcome was all-cause mortality until 2020 (wave 5). The secondary outcomes were incident cardiovascular disease (CVD) events until 2020, after excluding those with prevalent heart disease (n = 1380), stroke (n = 264), and hypertension (n = 2851) at baseline. CVD was ascertained based on the self-reported physician-diagnosed hypertension, heart disease, or stroke. Heart disease includes heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems. Dyslipidemia was defined as a self-reported diagnosis of dyslipidemia or lipid-lowering drugs usage or total cholesterol (TC) ≥ 6.22 mmol/L or triglyceride (TG) ≥ 2.26 mmol/L or high-density lipoprotein cholesterol (HDL-C) < 1.04 mmol/L or low-density lipoprotein cholesterol (LDL-C) ≥ 4.14 mmol/L [13]. Diabetes was defined as a self-reported diagnosis of diabetes or anti-diabetic medications usage or fasting blood glucose (FBG) ≥ 7.0 mmol/L or glycosylated hemoglobin A1c (HbA1c) ≥ 6.5% [14].

For the SCC, the primary outcome was prevalence of CVD, and the secondary outcome was the prevalence of various age-related chronic diseases, including hypertension, diabetes mellitus, gout, lung disease, tuberculosis, and asthma. CVD was ascertained based on the self-reported physician-diagnosed hypertension, coronary heart disease, or stroke. Hypertension was defined as a self-reported diagnosis of hypertension or anti-hypertensive medication usage or a measured systolic blood pressure (SBP) ≥ 140 mmHg or diastolic blood pressure (DBP) ≥ 90 mmHg [15] in the main analysis and self-reported diagnosis of hypertension or anti-hypertensive medication usage only in the sensitivity analysis. Diabetes was defined as a self-reported diagnosis of diabetes or anti-diabetic medications usage or fasting blood glucose (FBG) ≥ 7.0 mmol/L or glycosylated hemoglobin A1c (HbA1c) ≥ 6.5% [14]. Lung diseases included chronic bronchitis, emphysema, and chronic obstructive pulmonary disease (COPD), which were defined by self-reported physician diagnosis. For gout, tuberculosis, and asthma, each condition was defined based on participants’ self-reported physician diagnosis.

Definition of variables in CHARLS

The detailed information about CHARLS had been published in previous literature [11]. The CHARLS datasets can be accessed at the CHARLS home page at http://charls.pku.edu.cn/en. Sociodemographic characteristics, including age, sex, height, weight, waist, education (less than primary school, primary school, junior high school, high school, and college or above), and marital status (married, separated, and unmarried/divorced/widowed), and health‐related factors, including smoking status (never smoked and smoked [current or former smoker]), alcohol consumption (never, less than once a month, and more than once a month), daily sleep time, and depression, were obtained from validated questionnaire by trained staffs. Body mass index (BMI) was defined as the weight (kg) divided by height (m²). Depression was defined using the ten-item short form of the Center for Epidemiologic Studies Depression Scale (CESD-10), which has high validity and excellent psychometric properties among Chinese older individuals [16].

Definition of lifestyles in the SCC

Lifestyle information was collected through face-to-face interviews conducted by trained interviewers using structured questionnaires. Smoking and drinking status were defined as category variables, while diet and sleep duration were defined as continuous variables. Questions about tobacco smoking included the age at initiation or cessation and the amount of tobacco smoked per day by current and former smokers. Based on the response, smoking status was categorized into two groups: “never smoker” (the reference group) and “smoker” (combining current smoker and former smoker), and never smoking was defined as a low-risk lifestyle factor, as previously described [17]. Data on alcohol consumption included the age of initiation or cessation, drinking frequency, type of alcoholic beverage typically consumed, and volume of alcohol consumed per occasion. Drinking status was classified into three categories: never, drinking but less than once a month, and more than once a month. We defined participants who never or drink less than once a month as a low-risk lifestyle factor. Diet was evaluated from a food frequency questionnaire, as previously described [18]. We awarded 1 point for each food item that aligned with the 2022 Chinese Dietary Guidelines [19]. Detailed scoring was as follows: (1) fruit or (2) vegetables consumed more than three times per week; (3) whole grains, (4) eggs, (5) nuts, (6) dairy, (7) poultry, and (8) fish consumed at least once per week; (9) red meat consumed less than once per week; and (10) high-salt foods, (11) juices, and (12) sugary beverages consumed less than once per month. The diet quality score ranged from 0 to 12. For the current analysis, we defined a healthy diet with a diet diversity score in the top 40% of the study population [17]. Optimal nighttime sleep duration was defined as 7–9 h of sleep per night [20]. BMI was categorized as underweight (< 18.5), optimal weight (18.5–23.9), overweight (24–27.9), and obesity (≥ 28) [21]. Subsequently, the comprehensive lifestyle score was the number of low-risk factors (never smoking, never drinking or less than once a month, healthy diet, optimal nighttime sleep duration, and optimal weight).

Model construction

For samples with missing values, if features are missing randomly for only a small fraction of subjects, missing values in continuous datasets can be imputed using mean imputation or median imputation. However, when imputing data, care needs to be taken to ensure that variables are missing completely at random, with the probability of being missing not related to parameter values.

The “Gompertz” framework was employed to construct BA in this study [22]. Given the established sex difference in mortality trajectories, sex-specific Cox proportional hazards regression models were constructed in the training set. For these models, we implemented a two-step approach aligned with sex stratification: Separate null models were run for males and females, respectively. These null models were used to estimate the baseline mortality hazard (h₀) across the follow-up period and simultaneously derive sex-stratified mortality rate doubling times (MRDT_sex), a key parameter of the Gompertz model that quantifies the rate at which mortality risk increases with age for each sex (Supplementary Fig. 1). Subsequently, clinical parameters from the training set were standardized into z-scores to ensure comparability across features. The normalized data were then transformed into PC coordinates using the singular value decomposition function in R version 4.2.0 (https://www.R-project.org/). The first 19 PCs derived from the training dataset were selected as covariates (Supplementary Fig. 2), capturing 70% of the total variance, as previously described [23]. The factor loadings of the 41 clinical parameters across the 19 PCs were graphically represented through heatmap visualization (Fig. 1). These PCs, along with each subject’s CA, were incorporated into a Cox proportional hazards regression model, which was trained using 10-year mortality follow-up data (Supplementary Fig. 3). This optimized model enabled mortality hazard prediction (h_pc) based on individual positioning within the PC-transformed feature space. Finally, we computed equivalent age differentials (“Δage “) by determining the Gompertz age adjustment required to equate the relative hazard ratio (h_pc/h₀), formalized as:

$$▵age=\frac{\mathrm{In}{\displaystyle \frac{{\mathrm{h}}_{\mathrm{pc}}}{{\mathrm{h}}_0}}}{\mathrm{In}2}MRDTsex$$

Fig. 1.

Heatmap illustrating the loadings of the 41 clinical parameters onto the 19 PCs with significant weights in PCAge. We selected the 19 PCs (V1 to V19) with significant weights in the PCAge Cox models and extracted the corresponding right singular vectors (rotations) for each PC. To visualize the loadings of the clinical parameters on each of these PCs, we applied hierarchical clustering analysis to the 41 parameters before visualizing the re-ordered parameter weights (rotations) as a heatmap

The final BA was then calculated by adding this age correction to the subject’s actual CA:

$$\mathrm{BA}=\mathrm{CA}+▵\mathrm{age}$$

To account for potential sex-specific differences in aging trajectories, separate BA clocks were developed for males and females, exclusively using the training set.

Model evaluation

The BA model was then evaluated in the testing set by discrimination (Harrell’s concordance index [C-index] and receiver operator characteristic [ROC] curve). We calculated the time-dependent area under the ROC curve (AUC) for 4-, 7-, and 10-year intervals using the R package timeROC, of which 95% CIs were computed by 1000 bootstrap repetitions. We applied Cox models with PCAge as predictors of mortality to estimate the mortality risk associated with accelerated aging. The effect size was quantified as HR per standard deviation (SD) increase in PCAge.

Furthermore, we evaluated the predictive value of PCAge on the incidence of CVD, heart disease, stroke, hypertension, diabetes, and dyslipidemia. Unadjusted Cox proportional hazards regression models were implemented, with PCAge as the sole predictor. Participants who experienced mortality events before disease diagnosis were censored at their date of death in these survival analyses.

Application of PCAge in the SCC

The BA of each individual in the SCC was calculated using the model derived from the CHARLS training set (Supplementary Fig. 1). Participants were stratified into two groups: “biologically younger” (BA < CA) and “biologically older” (BA ≥ CA), based on the Δage difference between predicted BA and CA, as previously described [24] in main analysis. Given a fraction of individuals with only minor deviation between BA and CA, participants were stratified into three groups by Δage as the lowest 25% in the population (biologically younger), middle 50% (biologically concordant), and the highest 25% (biologically older) [22] in sensitivity analysis. Multivariable logistic regression analyses were conducted to examine associations between accelerated aging and seven age-related chronic conditions, including CVD, hypertension, diabetes mellitus, gout, lung disease, tuberculosis, and asthma. The primary exposure was the BA subgroup (reference: biologically younger), with CA included as a continuous covariate to account for residual age-related confounding. Associations were quantified using OR with corresponding 95% CIs. BA subgroup differences between biologically younger and biologically older were assessed using multivariable logistic regressions with adjustment for age, sex, education, income level, and lifestyle. Proportion ratios and 95% CIs were calculated.

Statistical analysis

Physiological and laboratory measurements with more than a 10% missing rate were excluded from the current analysis. For descriptive statistics, continuous variables were expressed as mean (SD) or median (interquartile range [IQR]), while categorical variables were expressed as number (percentage). Correlation analyses were performed using linear regression, and the strength of correlation was determined using the Pearson correlation coefficient. We employed Cox proportional hazards regression models to quantify HR per SD increase in biological age for mortality and disease incidence. To account for competing mortality risks in disease prediction, sensitivity analyses were conducted using Fine-Gray sub-distribution hazard models. Model discriminative accuracy was assessed via C-index and time-dependent ROC curves. For cross-sectional analyses in the SCC, multivariable logistic regression was implemented to evaluate associations between biological age subgroups (biologically older vs. younger) and disease prevalence. Adjusted PR with 95% CI were calculated for each outcome, adjusting for age and sex as covariates. All analyses were performed using R version 4.4.0 (R Foundation), with two-tailed P < 0.05 considered statistically significant.

Results

Biological aging acceleration and mortality risk

The training set was initially composed of 3674 males and 4164 females aged 30–96 years, with a feature set comprising data from medical examination, physiological, and laboratory measurements. Participants who died during the follow-up were older, more likely to be smokers or drinkers, and had metabolic diseases and CVD at the baseline (P < 0.001; Supplementary Table 3). PCAge was highly correlated with CA in both males (Pearson’s r = 0.87) and females (Pearson’s r = 0.87; Supplementary Fig.  4A) in the training set. According to BA (PCAge) and CA, participants were stratified into “biologically younger” (BA < CA) and “biologically older” (BA ≥ CA) subgroups. This classification revealed substantial heterogeneity in aging trajectories, with 45.6% of individuals identified as biologically older (Supplementary Table 4). Compared to subjects who were biologically younger, subjects in the biologically older group experienced a significantly higher rate of mortality over the 10-year follow-up (HR = 2.38, 95% CI: 2.08–3.14, P < 0.001; Supplementary Fig. 4B).

To address sex-specific aging patterns, gender-stratified analyses revealed elevated mortality risks in biologically older subgroups for both females and males. Females with accelerated aging showed a 2.56-fold higher mortality risk compared to biologically younger counterparts (HR = 2.56, 95% CI: 2.08–3.14, P < 0.001; Supplementary Fig.  4 C). Similarly, biologically older males exhibited a 2.25-fold increase in mortality risk compared to those who were biologically younger (HR = 2.25, 95% CI: 1.91–2.66, P < 0.001; Supplementary Fig. 4D). Notably, the overall mortality was higher in males, but the magnitude of mortality risk associated with accelerated aging was greater in females, suggesting potential sex-specific vulnerabilities in aging-related pathways.

The potential of PCAge in the prediction of mortality

We then tested the performance of PCAge in the testing set, comprised of 1552 males and 1807 females who were aged between 33 and 96 years. The characteristics of the study participants are shown in Supplementary Table 3. Like the training set, participants who died during follow-up were older, more likely to be current smokers, and patients with metabolic diseases or CVD (Supplementary Table 3). As expected, PCAge demonstrated strong correlations with CA for both females (Pearson’s r = 0.86; Fig. 2A) and males (Pearson’s r = 0.88; Fig. 2A). A total of 47.8% of the participants were classified as biologically older (Supplementary Table 4). Over the 10-year follow-up, biologically older individuals demonstrated a substantially higher mortality risk compared to those biologically younger counterparts (HR = 4.46, 95% CI: 3.51–5.67, P < 0.001; Fig. 2B). Given the existence of biologically concordant participants (with minor BA-CA deviation), individuals were further grouped into three groups by Δage quartiles: the lowest 25% (biologically younger), middle 50% (biologically concordant), and the highest 25% (biologically older). The results remained materially unchanged (HR: concordant vs. older = 2.98; younger vs. older = 10.74, both P < 0.001; Supplementary Fig. 5), further supporting the robustness of PCAge in the prediction of mortality. Sex-stratified analyses showed remarkable differences in mortality risk. Females with accelerated aging exhibited a 5.74-fold elevated mortality risk compared to their biologically younger counterparts (HR = 5.74, 95% CI: 3.86–8.55, P < 0.001; Fig. 2C), while biologically older males displayed a 3.55-fold increase in mortality risk relative to biologically younger males (HR = 3.55, 95% CI: 2.63–4.80, P < 0.001; Fig. 2D), suggesting that the impact of accelerated aging on mortality is more pronounced in females. Subsequently, the performance of PCAge was further evaluated across different age groups. Among participants aged between 55 and 64 years, biologically older individuals demonstrated a 1.99-fold elevated mortality risk compared to their biologically younger counterparts (HR = 1.99, 95% CI: 1.32–3.01, P = 0.001; Fig. 2E). Likewise, a 2.12-fold increased mortality risk in biologically older individuals compared to their biologically younger peers was also found in the 65–74 years age group (HR = 2.12, 95% CI: 1.26–3.57, P < 0.001; Fig. 2F). The predictive accuracy of PCAge was rigorously evaluated in testing sets. Moreover, the PCAge demonstrated an excellent performance in the prediction of mortality, achieving a C-index of 0.798 (95% CI: 0.797–0.799). The time-dependent accuracy of PCAge was further improved, with 4-year, 7-year, and 10-year AUC achieving 0.837 (95% CI: 0.780–0.894), 0.807 (95% CI: 0.770–0.845), and 0.809 (95% CI: 0.756–0.861), respectively (Supplementary Fig. 6). Collectively, these findings indicated that PCAge robustly predicted survival and mortality risk in populations aged between 55 and 75 years, with risk gradients intensified in advanced aging stage.

Fig. 2.

Sex-specific predictive performance of PCAge in mortality. A Scatter plots with linear regression lines demonstrating the correlation between chronological age and PCAge in males and females, respectively. Blue and red indicated male and female, respectively. B Kaplan–Meier survival curves comparing 10-year mortality risk between biologically older and younger participants in the testing set. Sex-stratified survival analysis showing increased mortality risk in biologically older C females and D males, compared to their biologically younger counterparts. Sex-stratified survival analysis showing increased mortality risk in biological individuals aged between E 55–64 years and F 65–74 years, compared to their biologically younger counterparts. Yellow and blue indicated biologically older and younger individuals, respectively

The potential of PCAge in the prediction of incident chronic diseases

We further evaluated the ability of PCAge in the prediction of the incidence of age-related chronic diseases, including CVD, heart disease, stroke, hypertension, diabetes, and dyslipidemia. Significant associations between accelerated biological aging (BA ≥ CA) and risk of all these age-related diseases were found, except for dyslipidemia (Fig. 3). Specifically, individuals with accelerated aging demonstrated the highest hazard ratio for stroke (HR = 2.38, 95% CI: 1.83–3.10, P < 0.001; Fig. 3C), followed by diabetes (HR = 1.51, 95% CI: 1.19–1.90, P < 0.001; Fig. 3E), heart disease (HR = 1.35, 95% CI: 1.10–1.65, P = 0.003; Fig. 3B), CVD (HR = 1.30, 95% CI: 1.12–1.52, P < 0.001; Fig. 3A), and hypertension (HR = 1.28, 95% CI: 1.07–1.52, P = 0.007; Fig. 3D). In contrast, risk of dyslipidemia showed no obvious difference in individuals with accelerated or delayed biologically aging (HR = 1.16, 95% CI: 0.98–1.37, P = 0.093; Fig. 3E). To address potential bias from competing risk of mortality, we conducted Fine-Gray sub-distribution hazard analyses. After accounting for death-related censoring, the associations of accelerated biologically aging with all these aging-related chronic diseases remained largely the same (Supplementary Fig. 7).

Fig. 3.

Predictive performance of PCAge in the incidence of aging-related diseases. Kaplan–Meier curves illustrating 10-year risk of incident A cardiovascular disease, B heart disease, C stroke, D hypertension, E diabetes, and F dyslipidemia in the testing set. Yellow and blue indicated biologically older and younger individuals, respectively

To further assess the gradient of risk across different levels of biological aging, we stratified the testing set into quartiles based on Δage (the difference between BA and CA). Notably, the contrast between the lowest 25% (delayed biologically aging) and the highest 25% quartiles (accelerated biologically aging) of Δage demonstrated a more pronounced difference in the risk of aging-related diseases (Supplementary Fig. 8). Individuals in the Q4 exhibited significantly higher risks of CVD, heart disease, stroke, hypertension, and diabetes compared to those in the Q1 (all P < 0.01), while dyslipidemia risk showed no significant difference (P = 0.667). For each one-unit increase in Δage, a progressively increased risk of CVD (HR = 1.02, 95% CI: 1.01–1.03, P < 0.001), heart disease (HR = 1.02, 95% CI: 1.01–1.04, P = 0.003), stroke (HR = 1.05, 95% CI: 1.03–1.07, P < 0.001), hypertension (HR = 1.03, 95% CI: 1.01–1.04, P < 0.001), and diabetes (HR = 1.03, 95% CI: 1.01–1.05, P < 0.001) was found (Supplementary Fig. 8), suggesting that PCAge captures a continuous gradient of risk of aging-related diseases. Consistently, PCAge also showed excellent discrimination in the prevalence of age-related diseases. Biologically older individuals had substantially higher probability of CVD (odds ratio [OR] = 3.91, 95% CI: 3.36–4.57, P < 0.001), heart disease (OR = 3.59, 95% CI: 2.86–4.55, P < 0.001), stroke(OR = 3.17, 95% CI: 1.97–5.29, P < 0.001), hypertension (OR = 3.74, 95% CI: 3.16–4.43, P < 0.001), diabetes (OR = 7.68, 95% CI: 5.14–11.9, P < 0.001) and dyslipidemia (OR = 1.85, 95% CI: 1.45–2.36, P < 0.001), compared to their biologically younger counterparts (Supplementary Fig. 9).

Identification of individuals with accelerated aging in Southern China

The baseline characteristics of the SCC participants, stratified by biological age, are presented in Supplementary Table 5. Of the 68,920 individuals, 35,746 (51.9%) were classified as biologically older. Significant differences were observed between the two subgroups across multiple demographic and clinical variables. For example, biologically older individuals were more likely to be female and had higher BMI but were less likely to be smokers or frequent drinkers (Supplementary Table 5).

Similarly, PCAge demonstrated strong correlations with CA for both males (Pearson’s r = 0.87, R² = 0.757, P < 0.001) and females (Pearson’s r = 0.87, R² = 0.765, P < 0.001) in the SCC (Fig. 4A). Compared to the biologically younger group, participants with accelerated biologically aging exhibited significantly higher odds of hypertension (OR = 1.34, 95% CI: 1.29–1.40, P < 0.001; Fig. 4B), CVD (OR = 1.59, 95% CI: 1.53–1.66, P < 0.001; Fig. 4C), diabetes (OR = 1.20, 95% CI: 1.13–1.27, P < 0.001; Fig. 4D), gout (OR = 2.16, 95% CI: 1.98–2.37, P < 0.001; Fig. 4E), lung disease (OR = 10.80, 95% CI: 8.75–13.50, P < 0.001; Fig. 4F), tuberculosis (OR = 6.50, 95% CI: 5.15–8.31, P < 0.001; Fig. 4G), and asthma (OR = 20.15, 95% CI: 12.65–34.66, P < 0.001; Fig. 4H). Considering the physiological difference in diagnosed and untreated individuals with abnormal blood pressure, sensitivity analysis was further performed exclusively in individuals with physician diagnosed hypertension, and the results remained materially unchanged (OR = 1.25, P < 0.001; Supplementary Fig. 10). Collectively, these findings highlighted the utility of PCAge as a robust predictor of both BA and the risk of multiple age-related diseases in an independent population.

Fig. 4.

Application of PCAge in the SCC. A Scatter plot and linear regression between chronological age and PCAge in males and females, respectively. Blue and red indicated male and female, respectively. Logistic regression analyses showing the associations between biological aging and prevalent B hypertension, C cardiovascular disease, D diabetes, E gout, F lung disease, G tuberculosis, and H asthma. Red and blue indicated biologically older and younger individuals, respectively

Individuals with a higher probability of accelerated aging

To further elucidate the social and behavioral determinants of biological aging, we conducted stratified logistic regression analyses across the entire cohort and in individuals with and without chronic diseases (Fig. 5). All models were adjusted for age and sex. In the full cohort, after correcting for CA and sex, biologically older individuals were more likely to be female (PR = 1.94, 95% CI: 1.88–2.01, P < 0.001), but less likely to be chronologically older individuals. Higher educational attainment was associated with a lower likelihood of being accelerated aging (college or high: PR = 0.74, 95% CI: 0.69–0.80, P < 0.001), while a high income, reaching more than 100,000 CNY per year, demonstrated a positive correlation with biologically aging (PR = 1.47, 95% CI: 1.40–1.53, P < 0.001). Not surprisingly, individuals with unfavorable lifestyles were more likely to be biologically older. In individuals with no chronic diseases, the findings remained largely the same. Females exhibited a substantially higher risk of biological aging (PR = 3.49, 95% CI: 3.32–3.67, P < 0.001). While higher education (college or higher: PR = 0.76, 95% CI: 0.68–0.85, P < 0.001) remained a protective factor for biological aging, the favorable effect of higher income substantially attenuated. On the contrary, in individuals with prevalent chronic diseases, a distinct pattern of biological aging was found. Higher education paradoxically increased the risk of accelerated aging, and the risk gradients of higher income for accelerated aging were remarkably steeper. Moreover, a larger adverse effect of suboptimal lifestyles on accelerated aging was found in individuals with chronic diseases (Fig. 5). Collectively, the above findings underscored a complex interplay between the social and behavioral factors and biological aging, highlighting the importance of large-scale BA screening and targeted interventions to relieve aging-related disorders.

Fig. 5.

Sociodemographic and lifestyle characteristics of biologically older individuals in the SCC. Proportion ratios were derived from logistic regression models adjusted for age and sex. Healthy lifestyle factors included normal BMI, non-smoking status, minimal alcohol consumption, dietary diversity, and adequate sleep duration

Discussion

In this study, we developed and validated the performance of PCAge, a multidimensional biological age estimator derived from routine primary care parameters, in the quantification of aging trajectories. PCAge demonstrated capturing both mortality risk and chronic disease risk with high precision. Based on parameters routinely collected in primary care, such as blood pressure, lipid profiles, and anthropometric measurements, PCAge held a unique advantage of large-scale implementation, even in remote and resource-limited regions. Our findings proposed a feasible tool for the monitoring of aging status and the evaluation of and the prioritization of anti-aging interventions.

Similar to the well-established epigenetic clocks like Horvath’s and PhenoAge [6, 24], PCAge, based on readily available parameters in primary care, also demonstrated a strong correlation with CA (Pearson’s r = 0.86–0.88, P < 0.001). Contrary to the costly DNA methylation technique, PCAge, incorporating clinical and physiological parameters, offered a clinically actionable framework for aging assessment at the population level. Of note, despite an appeal of the World Health Organization for “implementable aging metrics” in primary care [25], existing BA tools are costly and fail to be implemented at a population scale, especially in low- and middle-income countries, where over than 75% of health systems lack methylation sequencing capabilities [26]. In this connection, the PCAge proposed by our study functions as a feasible alternative, which transforms primary care data into frontline aging surveillance networks.

The sex-specific modeling of PCAge addresses known biological and lifestyle divergences between males and females [27], which may explain the stronger mortality risk association observed in females (HR = 5.74 vs. 3.55 in males). This sex disparity mirrors findings from large-scale cohort studies where female-specific aging trajectories often show higher sensitivity to metabolic dysregulation [28]. Such an observation was also consistent with the emerging evidence of sex-specific aging mechanisms. For instance, estrogen deficiency in postmenopausal women may induce mitochondrial dysfunction and inflammation [29], which in turn accelerates aging in females. Furthermore, age-stratified analyses revealed age-dependent patterns in the predictive capacity of PCAge, with a superior performance in the 55–75 age stratum. Participants aged 65–74 years exhibited significantly greater mortality risk disparities between BA subgroups compared to their counterparts aged 55–64 years (ΔHR =  + 0.13), reflecting an exponential accumulation of cellular damage postulated in the hallmarks of aging [30]. This age-dependent risk gradient underscored the progressive amplification of biological aging impacts in older populations. The model exhibited robust temporal validity, maintaining predictive accuracy over 10-year follow-up periods (time-dependent AUC: 0.807–0.837) across broad age ranges (33–96 years). Stable decade-long validity enables continuous aging monitoring from early adulthood, specifically mid-life risk stratification and elderly frailty monitoring.

Furthermore, the PCAge demonstrates a robust clinical utility in forecasting the incidence of age-related chronic diseases, including CVD, heart disease, stroke, and diabetes. Notably, these associations remained significant in Fine-Gray competing risk models accounting for mortality, further confirming the specificity of PCAge for disease prediction independent of survival bias. The pan-disease predictive capability of PCAge further supported the notion that it captured the fundamental aging mechanisms driving multimorbidity rather than organ-specific pathologies [10, 30]. Biological aging and the progression of chronic diseases always demonstrate a mutually reinforcing bidirectional relationship—a vicious cycle in which cellular senescence accelerates the onset of disease, subsequently exacerbating biological aging [31, 32]. The management of age-related disease progression is intrinsically linked to the mitigation of biological aging. By identifying accelerated aging trajectories up to a decade before clinical manifestation, PCAge could revolutionize preventive care frameworks through early, targeted interventions.

Given that all the components of PCAge are readily available from the primary health care system across the country, it serves as a unique opportunity to establish an aging surveillance network, enabling an early identification of high-risk individuals. Of note, accelerated aging was closely associated with unhealthy lifestyles and unfavorable socioeconomic status, which were modifiable. Lower educational attainment (PR = 0.74, 95% CI: 0.69–0.80) and suboptimal lifestyle scores were strongly associated with accelerated biological aging, consistent with the socioeconomic gradient in aging observed in other cohorts [33, 34]. Lifestyle factors emerged as modifiable drivers of biological aging. Suboptimal lifestyle scores increased aging risk across all subgroups, with effects magnified in chronic disease populations (PR = 1.48, 95% CI: 1.31–2.67). These results align with the “Gompertzian” model of aging, where cumulative environmental exposures accelerate mortality risk [35]. The calculation of PCAge requires only routine clinical measurements, making it a pragmatic and cost-effective tool for large-scale screening in primary care settings. Targeted interventions for high-risk subgroups identified by PCAge could reduce age-related disease burden in low-resource regions [36], particularly advantageous for resource-limited regions where advanced diagnostics remain unavailable.

There are several limitations in this study. First, PCAge was derived from a limited set of biomarkers available in primary health care examinations. Thus, PCAge might not capture the full spectrum of biological aging. Second, reliance on self-reported lifestyle data introduces potential measurement bias. Future work should integrate wearable device data (e.g., sleep tracking) to enhance precision. Third, PCAge was trained and tested in different datasets from the same cohort of elderly Chinese. Although the unity of PCAge had been externally validated in the SCC, further validation in diverse cohorts is warranted. Fourth, age-stratified analysis of PCAge focused only on the 55–64 and 65–74 age groups. Future studies with larger sample sizes and longer follow-ups are warranted to further evaluate the utility of PCAge in younger and more advanced elderly.

Conclusion

PCAge represents an innovative and feasible tool for quantifying biological aging and early identification of individuals susceptible to accelerated age-related decline. These metrics hold significant promise for advancing targeted preventive strategies aimed at promoting healthy longevity.

Supplementary information

Below is the link to the electronic supplementary material.

Funding

This study was supported by the National R&D Project (2022YFC3600600).

Data availability

The SCC data are available through a reasonable request to the corresponding authors. For the CHARLS dataset, researchers registered with the official websites of CHARLS (http://charls.pku.edu.cn) can apply for access to the CHARLS Resource.

Code availability

Code can be made available upon request by contacting the corresponding author.

Declarations

Ethics approval and consent to participate

The China Health and Retirement Longitudinal Study was approved by the Ethics Review Committee of Peking University. The South China Cohort was approved by the Ethics Committee of the School of Public Health, Sun Yat-Sen University (2017–001).

Consent for publication

None.

Competing interests

The authors declare no competing interests.