Biological Age Acceleration Associated with the Progression Trajectory of Cardio-Renal–Metabolic Multimorbidity: A Prospective Cohort Study

We used data from the UK Biobank, a large-scale cohort study that enrolled more than 500,000 UK residents aged 37 to 73 years. Detailed descriptions of UK Biobank measurement methods and data collection are available in previous studies [26]. Briefly, between 2006 and 2010, approximately 9.2 million individuals near 22 health assessment centers in England, Scotland, and Wales were invited to join the cohort. About 5.5% agreed to participate and provided written consent before enrollment. At baseline and during follow-up, participants provided data on lifestyle factors, health information, and physical measurements. Participants were excluded if they withdrew consent, had T2DM, CVD, or CKD before baseline, or lacked BA information at baseline. In total, 278,927 participants were included in the main analysis (Figure S1). The study was approved by the North West Multicenter Research Ethics Committee (reference number 11/NW/0382). This research was conducted using UK Biobank data under application number 88589.

Incident cases of CRMDs (i.e., T2DM, CVD, and CKD) and death were ascertained using self-reports, primary care records, hospital admission records, and death registry records [1,2,27]. FCRMD was defined as the first occurrence of any CRMD [2,5], with diagnosis time based on the earliest recorded diagnosis among these conditions. CRMM was defined as the coexistence of at least two CRMDs [2,3,28], with diagnosis time based on the recorded diagnosis of a second CRMD during follow-up. Further details on CRMD diagnoses are provided in Table S1.

We assessed BA using widely validated algorithms: PhenoAge [13] and KDMAge [14]. With reference to similar UKB studies [25,29], we identified 14 clinical biomarkers to calculate BA, covering multiple systems like immunity, inflammation, organ homeostasis, and metabolism. PhenoAge focused on immunity and inflammation, using nine biochemical parameters: lymphocytes, mean cell volume, serum glucose, red cell distribution width, white blood cell count, albumin, creatinine, C-reactive protein, and alkaline phosphatase [25,29]. KDMAge emphasized pulmonary function and cardiovascular metabolism, using forced expiratory volume in one second (FEV1), systolic blood pressure (SBP), and seven biochemical parameters: albumin, creatinine, C-reactive protein, alkaline phosphatase, total cholesterol, glycated hemoglobin, and blood urea nitrogen [25,29]. We excluded participants with missing data for any BA component (Table S2).

We trained the algorithm parameters in a reference sample and applied them to the UK Biobank data. The reference sample consisted of individuals aged 30 to 75 years from the National Health and Nutrition Examination Survey (NHANES) III cohort. PhenoAge was developed based on factors related to mortality risk. PhenoAge represents the predicted BA matching the average mortality risk in the reference sample [13]. KDMAge was derived from regression analyses of clinical biomarkers and CA, reflecting a predicted BA corresponding to normal physiological function in the reference sample [14]. BA acceleration was defined as the residual from a linear regression of BA on CA, reflecting the rate of biological aging [25,29]. We analyzed BA acceleration as a continuous variable and further dichotomized it, defining “biologically older” as KDMAge acceleration or a PhenoAge acceleration >0 (indicating faster aging) and “biologically younger” as ≤0. Table 1 summarizes the baseline characteristics of BA, its components, and BA acceleration. The algorithms for BA and the corresponding R code are available in the “BioAge” R package [30].

Based on prior knowledge [2,5], the baseline covariates included age (years), sex (male/female), ethnicity (White/non-White), the Townsend deprivation index (≤median/>median), education level (low/high), body mass index (BMI, kg/m²), smoking status (never/former/current), alcohol consumption (none/moderate/heavy), physical activity (low/moderate/high), and dietary behaviors (unhealthy/healthy). In this study, a high educational level refers to a college or university degree or professional qualifications like nursing or teaching. Moderate alcohol consumption is defined as a daily intake of ≤14 g for females and ≤28 g for males. Physical activity is assessed using the International Physical Activity Questionnaire (IPAQ). Using the duration of activities (walking/moderate/vigorous), physical activity was converted into metabolic equivalents (MET) and categorized as low (<600 MET-minutes/week), moderate (600–3000 MET-minutes/week), or high (>3000 MET-minutes/week). Dietary behaviors were assessed according to recent cardiovascular health dietary recommendations [31]. Healthy dietary behaviors were defined as meeting at least five items (Table S3).

Participant characteristics, grouped by incident disease states (CRMD-free, FCRMD, and CRMM), were summarized with mean and standard deviations (SDs) for continuous variables and the frequency and percentage for categorical variables. For covariates with missing data (0.12–21.69%,), multiple imputations by chained equations were applied to reduce inferential bias. Follow-up for each participant began at recruitment and continued until the earliest occurrence out of CRMD onset, death, loss to follow-up, or 19 December 2022. Table S4

First, we used the Cox regression model to evaluate associations between BA acceleration and FCRMD, CRMM, and death. Second, to explore the association between the two BA accelerations and progression from baseline health to FCRMD, and then to CRMM and death, we applied the multi-state model. This model, an extension of the Cox regression model, estimates associations between risk factors and multiple disease states while accounting for competing risks [32,33]. Schoenfeld residuals verified the proportional hazards assumption. No violations were detected. Following previous research [2], five transition phases (transition pattern A, Figure 1A) were established: (1) baseline to FCRMD; (2) baseline to no N-CRMD death; (3) FCRMD to CRMM; (4) FCRMD to all-cause death; and (5) CRMM to all-cause death. If participants entered different states on the same date, we adjusted the prior state’s entry date to 0.5 days before the later state’s entry date, referencing prior studies [2,29,34,35]. Additionally, analyses specific to the various types of FCRMD were performed. In this analysis, eleven transitions (transition pattern B, Figure 1B) were established based on the three FCRMD types (T2DM, CVD, and CKD). In transition pattern B, participants with at least two CRMD diagnoses on the same date (n = 2147) were excluded as the sequence of CRMD events could not be determined. In total, 276,780 participants were included. To explore the dose–response relationship, we used a restricted cubic spline (RCS) analysis in transition pattern A, with knots at the 10th, 50th, and 90th percentiles.

Subsequently, using the “elect” R package and referring to previous studies [36,37,38], we calculated differences in CRMD-free and total life expectancy, using age (in years) as the time scale. Given notable sex differences in lifespan, we estimated life expectancy separately for males and females. To maintain estimate reliability and avoid unrealistic extrapolation [37], life expectancy predictions were restricted to ages 40, 50, and 60, covering the majority of the UK Biobank sample (99.62% aged <70 years) and representing middle age, late–middle age, and early old age.

To identify subgroups sensitive to BA acceleration, we performed subgroup analysis in transition pattern A based on sociodemographic characteristics (age, sex, Townsend deprivation index, and education) and lifestyle factors (BMI, smoking status, alcohol consumption, physical activity, and diet). Multiplicative interactions were tested using the likelihood ratio test.

To evaluate the stability of the results, we conducted several sensitivity analyses for transition pattern A: (1) for participants entering different states on the same date, the entry date for the prior state was calculated using four additional time intervals (0.5, 1, 3, and 5 years); (2) a transition directly from baseline to CRMM was added; (3) only White participants were included; (4) participants diagnosed with any CRMD within two years after enrollment were excluded; (5) further adjusting for the use of cholesterol-lowering and antihypertensive medications was conducted to assess their association with progression trajectory; (6) blood glucose was excluded from PhenoAge and glycated hemoglobin and creatinine from KDMAge to remove biomarkers directly tied to T2DM and CKD diagnoses; (7) participants with any of the following at baseline were excluded to mitigate the confounding effect of baseline glucose levels and kidney function: glucose ≥7.0 mmol/L with a fasting time ≥8 h, glucose ≥11.1 mmol/L with a fasting time <8 h, glycated hemoglobin ≥48 mmol/mol, an estimated glomerular filtration rate (eGFR) [39] <60 mL/min/1.73m²; or albuminuria ≥3 mg/mmol; (8) the association of BA acceleration per interquartile range (IQR) and increase in progression trajectory were evaluated; and (9) in the multi-state model, three states—specific FCRMD, specific two CRMM (coexistence of two CRMDs), and three CRMM (coexistence of three CRMDs)—were considered to investigate their associations with disease progression. This transition pattern is presented in Figure S2 (transition pattern C). All analyses were performed using R version 4.4.1.

Table 1 outlines the characteristics of the total participants, CRMD-free participants, FCRMD patients, and CRMM patients. Patients with one or more CRMDs tended to be older, more often male and had a higher Townsend deprivation index, lower educational level, higher BMI, higher smoking rate, lower rate of moderate alcohol consumption, an older BA, and higher BA acceleration.

During a median follow-up of 13.79 years (IQR: 13.10–14.44), 64,093 participants (23.0%) experienced at least one CRMD. Among them, 9512 (14.8%) progressed to CRMM. During follow-up, 18,065 participants died: 8701 after FCRMD and 2192 after CRMM (Figure 1A). In the specific FCRMD analysis (excluding 2147 participants), 8171 (3.0%) patients developed incident T2DM; 47,794 (17.3%) developed incident CVD; and 5981 (2.2%) developed incident CKD (Figure 1B).

The mean (SD) values of PhenoAge acceleration and KDMAge acceleration were 6.51 (4.26) and −3.31 (9.76) years, respectively. A weak positive correlation was found between PhenoAge acceleration and KDMAge acceleration (with a Pearson correlation coefficient = 0.22; p < 0.001).

PhenoAge acceleration was associated with the risks of FCRMD and CRMM using the Cox regression model (). After full adjustment, individuals classified as biologically older by PhenoAge acceleration showed a 64% increased risk of FCRMD (HR [95% CI]: 1.64 [1.59, 1.68]), a 137% increased risk of CRMM (HR [95% CI]: 2.37 [2.24, 2.50]), and a 107% increased risk of all-cause death (HR [95% CI]: 2.07 [1.99, 2.17]) compared with those classified as biologically younger. For every per 1-SD increase in PhenoAge acceleration, the HRs (95% CI) were 1.18 (1.17, 1.19) for FCRMD, 1.37 (1.35, 1.39) for CRMM, and 1.27 (1.26, 1.28) for all-cause death. Similarly, KDMAge acceleration was positively associated with the risks of FCRMD, CRMM, and all-cause death. Table S5

Multi-state analysis showed that BA acceleration was associated with all five transitions in CRMM progression (Table 2). Fully adjusted HRs (95% CI) per 1-SD increase in PhenoAge acceleration were 1.18 (1.17, 1.19) for baseline to FCRMD; 1.24 (1.22, 1.26) for FCRMD to CRMM; 1.25 (1.22, 1.27) for baseline to death; 1.13 (1.11, 1.15) for FCRMD to death; and 1.09 (1.06, 1.12) for CRMM to death. KDMAge acceleration showed positive associations across all five transitions, though the association with progression from CRMM to death was weaker (HR [95% CI]: 1.04 [1.00, 1.08]). Further analysis showed that participants with above-mean BA acceleration showed a dose–response relationship across all five transitions (Figure 2). Associations remained consistent when analyzed per IQR increase (Figure S3), except for the transition from CRMM to death for KDMAge acceleration, which was not significant (HR [95% CI]: 1.01 [0.97, 1.06]).

Further analysis using the multi-state model divided FCRMD into T2DM, CVD, and CKD (Table 3), indicating that a per 1-SD increase in PhenoAge acceleration was associated with all eleven transitions. A per 1-SD increase in KDMAge acceleration was associated with most transitions but not with certain transitions related to death, including the transition from T2DM to death (HR [95% CI]: 0.96 [0.88, 1.05]), from CKD to death (HR [95% CI]: 0.94 [0.86, 1.04]), and from CRMM to death (HR [95% CI]: 1.03 [0.99, 1.07]). When BA acceleration was dichotomized, the results remained largely unchanged. For PhenoAge acceleration, compared with biologically younger individuals, biologically older individuals showed higher risks across all eleven transitions, except for the transition from CKD to death (HR [95% CI]: 1.27 [0.97, 1.65]), possibly due to a limited sample size. For KDMAge acceleration, most transitions showed higher risks, including the transition from CRMM to death; however, the transitions from T2DM to death (HR [95% CI]: 0.97 [0.81, 1.17]) and from CKD to death (HR [95% CI]: 0.85 [0.70, 1.05]) lacked statistical significance. Furthermore, in the disease progression analysis of CRMM (transition pattern C, Figure S4), BA acceleration was associated with all transitions, except for the transitions from T2DM and CKD to three CRMMs for PhenoAge acceleration, potentially due to a small sample size.

Both males and females demonstrated reductions in CRMD-free and total life expectancy associated with higher BA acceleration, with PhenoAge acceleration linked to greater losses (Figure 3). Among the males classified as biologically older based on PhenoAge acceleration compared with biologically younger males, CRMD-free life expectancy was associated with a reduction (3.53 [3.38, 3.70] years at age 60 and 5.43 [5.18, 5.69] years at age 40) and total life expectancy was associated with a reduction (4.10 [3.82, 4.40] years at age 60 and 5.26 [4.91, 5.62] years at age 40). Similarly, among males classified as biologically older based on KDMAge acceleration compared with biologically younger males, CRMD-free life expectancy was associated with a reduction (2.21 [2.11, 2.32] years at age 60 and 3.25 [3.10, 3.42] years at age 40) and total life expectancy was associated with a reduction (2.16 [1.93, 2.40] years at age 60 and 2.55 [2.31, 2.81] years at age 40). The same pattern was observed in females.

We observed modification effects using sociodemographic characteristics (age, economic status, education) and lifestyle factors (smoking status, alcohol consumption, physical activity, diet) on the relationship between BA acceleration and one or more transitions (). When exposed to the same increase in BA acceleration, older individuals faced a higher risk of progressing from FCRMD to CRMM; healthy individuals with low economic status, low education, non-moderate alcohol consumption, low physical activity, or unhealthy dietary behaviors demonstrated an elevated risk of developing FCRMD; smoking history modified the transitions from baseline to FCRMD, baseline to death, and FCRMD to death. Findings from the multiplicative interaction model were consistent with those of the stratified analysis (). Specifically, older age, low physical activity, smoking history, and unhealthy dietary behaviors showed a synergistic effect, whereas moderate alcohol consumption, high economic status, and high education exhibited an antagonistic effect. Tables S6–S14 Tables S6–S14

The sensitivity analysis results are summarized in. The PhenoAge acceleration results remained stable, but KDMAge acceleration results lost statistical significance in transitions from CRMM to death when excluding those diagnosed with any CRMD within two years after enrollment or participants with abnormal baseline levels of glucose, glycated hemoglobin, eGFR, or albuminuria. Table S15 and Figure S5

Using prospective cohort data from approximately 270,000 UK Biobank adults, we first examined how accelerated biological aging, measured by two novel integrative BA algorithms (PhenoAge and KDMAge), was related to the progression from health to FCRMD and then to CRMM and death. We found that BA acceleration was associated with all five transitions, with PhenoAge acceleration showing stronger associations. After splitting FCRMD into T2DM, CVD, and CKD, PhenoAge acceleration consistently exhibited stronger associations. Additionally, biologically older individuals were correlated with reduced CRMD-free and total life expectancy at ages 40, 50, and 60, with larger reductions in those defined as biologically older by PhenoAge acceleration. We also identified subgroups more vulnerable to BA acceleration in certain transitions.

Consistent with prior studies, we found that higher BA acceleration was correlated with single CRMDs, including T2DM [21,22], CVD [22,23,24], and CKD [25]. For example, a Korean study found that BA acceleration, calculated using principal component analysis and based on clinical parameters, was associated with the incidence of diabetes, heart disease, and stroke [22]. A UK Biobank study showed that individuals with higher PhenoAge or KDMAge acceleration had an increased risk of CKD, independent of CA [25]. T2DM, CVD, and CKD share pathophysiological mechanisms, often co-occur, and are proposed to be managed as a syndrome through unified prevention and treatment [1]. However, to our knowledge, no study has yet explored the temporal association between BA acceleration and the progression from a single CRMD to CRMM and death. The multi-state model effectively accounts for various disease stages and competing risks [29,32,33], supporting the dynamic monitoring of progression within a single framework.

In this study, we employed a multi-state model to assess, for the first time, the association between BA acceleration and the risk of the subsequent progression of a single CRMD. We found that baseline BA acceleration was correlated with an increased risk of transition from FCRMD to CRMM. Our findings are consistent with evidence linking BA acceleration to cardio-renal–metabolic health. For instance, a prospective cohort study indicated that patients with diabetes have a higher risk of progressing to ischemic heart disease or stroke [29]. Another study found that in CKD patients, biological aging—measured by telomere length, KDMAge, and PhenoAge—was associated with an elevated risk of CVD [40]. We further categorized CRMM into two coexisting CRMDs (two CRMMs) and three coexisting CRMDs (three CRMMs), revealing that BA acceleration was associated with the progression from a single specific CRMD to two specific CRMMs and three CRMMs, thus extending existing research. The scarcity of comparable studies may stem from two main factors: first, although the interplay between T2DM, CVD, and CKD is well-recognized, their integration as a unified syndrome is a recent concept [1]; second, CRMD patients often face poor prognosis and high mortality [8].

Furthermore, our study showed BA acceleration to be associated with mortality risk in participants with or without CRMD. PhenoAge acceleration was associated with increased risk across all mortality-related transitions, whereas KDMAge acceleration showed no significant association with certain transitions, such as those from T2DM or CKD to death. Additionally, individuals defined as biologically older by PhenoAge were linked to a greater loss in life expectancy. These differences might partly stem from the selection of biomarkers and their degrees of contribution: PhenoAge primarily focuses on biomarkers of inflammation and immune function [25,29], potentially capturing a multi-system decline in disease progression more comprehensively. In CKD patients [41] and T2DM patients [42], inflammation levels have served as an independent predictor of mortality risk. By comparison, KDMAge concentrates on heart and lung function [25,29]. While it accurately represents cardiovascular and respiratory health, it is less sensitive to inflammation and other significant pathological processes. In addition, KDMAge only tracks CA with a linear increase [30,43], while PhenoAge captures both CA and mortality risk with an exponential increase [30,43], allowing it to capture nonlinear increases in mortality risk. This might account for PhenoAge’s greater sensitivity to mortality risk and lifespan loss. Our findings offer important implications for clinical practice. The blood, biochemical, and clinical parameters used for the BA assessment can be easily obtained in clinical environments, allowing for the timely identification and prioritized monitoring of high-risk individuals with higher BA acceleration. Our results show that higher PhenoAge acceleration is more closely associated with CRMM progression risk, warranting the prioritization of secondary prevention measures. If interventions (e.g., active lifestyles or aging-targeted drugs) improve clinical parameters in BA algorithms, this might lead to the prevention, control, and delay of age-related CRMM progression and healthy lifespans might be extended.

We identified subgroups more susceptible to BA acceleration. Specifically, FCRMD patients aged 65 years or older were more vulnerable to progressing to CRMM. This may be due to the poorer health of older populations [44,45]. Healthy individuals at baseline with unhealthy lifestyle factors (e.g., non-moderate alcohol consumption, low physical activity, or unhealthy dietary behaviors) were found to face a higher risk of developing FCRMD. This was likely associated with the cumulative impact of long-term unhealthy behaviors on metabolic disorders [46]. Moreover, BA acceleration was more closely linked to the transition from health to FCRMD among individuals with lower socioeconomic status or education, possibly associated with limited healthcare access [47,48] and increased exposure to stressful social conditions [49]. Notably, smoking history modified the most transitions, including those from health to FCRMD, health to death, and FCRMD to death. This aligns with a study showing how smoking is most strongly associated with CRMM progression among seven lifestyle factors [2]. This underscores the urgent need for smoking control measures to prevent CRMM progression. More research is needed to explore how biological aging and sociodemographic and lifestyle factors contribute to CRMM progression.

Clinical parameter-based BA may affect CRMM progression through interconnected mechanisms. Accelerated biological aging is often associated with altered adipokine expression [16,17], while dysregulated lipid metabolism may drive CRMD onset and subsequent progression [50]. BA acceleration is associated with the activation of pro-inflammatory [18] and pro-oxidative pathways [19]. These disruptions in inflammation and oxidative stress are associated with increased pro-fibrotic factor production, which may lead to pathological changes like tubulointerstitial fibrosis and glomerulosclerosis, which are key to CKD initiation [51,52]. Furthermore, inflammation is linked to atherosclerosis, endothelial damage, and myocardial injury [53,54], which correlate with CVD and mortality risks. Additionally, BA acceleration is linked to insulin resistance [20], which may contribute to a harmful feedback loop between T2DM, CVD, and CKD [50]. Further research is needed to clarify these complex mechanisms.

The key strengths of this study include the use of the multi-state model, data from a large, prospective cohort, and BA algorithms with the potential for broader clinical application. Nevertheless, several limitations warrant consideration. First, biomarkers in the BA algorithms were measured only at baseline, limiting our ability to track BA acceleration over time; however, relying on baseline BA acceleration reduces reverse causality after the onset of outcomes. Second, the association between BA acceleration and CRMM progression might be related to cardio-renal–metabolic health biomarkers incorporated into the algorithms; however, this association was observed even after excluding these biomarkers. Third, our study population, primarily of European descent, might have introduced healthy volunteer bias. Participants generally presented better health, higher education, and socioeconomic status than other cohorts, which limited the generalizability of our findings. Caution is advised when applying these results to broader populations. Future studies should validate these findings in more diverse and representative groups. Fourth, our analyses were conducted in a general population, and the clinical implementation of BA acceleration still requires further study. Finally, the observational nature of this study limits the ability to establish causality.

The availability of this data is restricted. The data were sourced from the UK Biobank and, with permission from UK Biobank, can be accessed at https://www.ukbiobank.ac.uk↗.