Association between extreme temperature events and intrinsic capacity trajectories among middle-aged and older adults: the China Health and Retirement Longitudinal Study (CHARLS)

Abstract

Background

Global climate change has increased the frequency, intensity, and duration of extreme temperature events, while population aging poses unprecedented public health challenges. However, the association between extreme temperature events and longitudinal trajectories of intrinsic capacity (IC)—a multidimensional construct proposed by the World Health Organization as a core indicator of healthy aging—remains unknown.

Methods

Data were drawn from 1,393 participants aged 45–80 years who had complete IC assessments across three waves (2011, 2013, 2015) of the China Health and Retirement Longitudinal Study (CHARLS). IC was operationalized as a composite score (range 0–10) spanning locomotion, sensory, vitality, psychological, and cognition domains. K-means clustering identified heterogeneous IC trajectories. Extreme temperature events were defined using city-specific daily apparent temperature (heatwave: ≥95th percentile for ≥ 3 days; cold spell: ≤5th percentile for ≥ 3 days). Multinomial logistic regression examined associations between cumulative heatwave days (2011–2015) and IC trajectory membership, adjusting progressively for sociodemographic, lifestyle, and health-related covariates.

Results

Three IC trajectories were identified: High-stable (41.3%), Moderate-declining (26.4%), and Low-declining (32.3%). Each additional heatwave day significantly increased the risk of belonging to the Low-declining trajectory (fully adjusted RRR = 1.035, 95% CI: 1.012–1.058, P = 0.003). This association was more pronounced among participants with primary school education or below (RRR = 1.07, 95% CI: 1.04–1.11) and rural residents (RRR = 1.05, 95% CI: 1.03–1.08), with significant effect modification by education (P for interaction = 0.031) and residence (P for interaction = 0.024). Cold spell days showed no significant association (RRR = 1.02, 95% CI: 0.96–1.08, P = 0.554). Sensitivity analyses supported result robustness.

Conclusions

Cumulative heatwave exposure is associated with an increased risk of belonging to the most vulnerable IC trajectory among Chinese middle-aged and older adults, particularly among socioeconomically disadvantaged groups. These findings underscore the urgent need for targeted climate adaptation strategies to protect functional health in aging populations.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12889-026-29470-y.

Introduction

Global population aging is accelerating at an unprecedented pace [1]. By 2050, the number of people aged 60 years and older is projected to approach 2.1 billion, making aging one of the most significant public health challenges of the 21st century [2]. Concurrently, climate change is driving increases in the frequency, intensity, and duration of extreme temperature events (ETEs), including heatwaves and cold spells [3]. It is projected that global warming will reach 1.5 °C between 2021 and 2040, further intensifying extreme weather events [4]. Against the backdrop of rapid population aging and escalating climate extremes, understanding how environmental factors influence the health and functional trajectories of older adults has become a pressing scientific question.

Middle-aged and older adults are particularly vulnerable to extreme temperatures due to age-related decline in thermoregulation, reduced physiological reserve, and a higher burden of chronic diseases [5]. A growing body of epidemiological evidence has confirmed that ETEs exert adverse effects on multiple health dimensions in this population, including elevated mortality risk, increased incidence of cardiovascular and respiratory diseases, accelerated cognitive decline, and worsening mental health [6–8]. Several studies have demonstrated that both heatwaves and cold spells are positively associated with the risk of multimorbidity among older adults, and that high temperatures are correlated with increased functional disability assessed by activities of daily living (ADL) [9, 10]. However, the existing literature has predominantly focused on single-domain health outcomes—such as cognitive function, specific disease endpoints, or mortality—rather than adopting a holistic, multidimensional perspective to systematically evaluate how ETEs influence the overall functional aging trajectory of older individuals.

To address this gap, the World Health Organization (WHO) introduced the concept of intrinsic capacity (IC) within its framework for healthy aging [11]. IC is defined as “the composite of all the physical and mental capacities that an individual can draw on” and encompasses five core domains: locomotion, vitality, sensory function (vision and hearing), cognition, and psychological well-being [12]. Unlike traditional disease-centered outcomes, IC reflects an individual’s inherent functional reserve and has been placed at the core of the WHO Integrated Care for Older People model, which provides guidance on integrating IC assessment into care delivery for aging societies. Accumulating evidence demonstrates that IC trajectories in older populations are heterogeneous and strongly predictive of adverse health outcomes [13, 14]. Longitudinal studies conducted in China, France, and other countries have identified distinct IC trajectory patterns—such as persistently high, moderately declining, and rapidly declining—and have confirmed that unfavorable IC trajectories are significantly associated with increased risk of disability, reduced quality of life, and higher incidence of cardiovascular disease [15, 16]. Thus, systematic evaluation of IC trajectories provides a more comprehensive perspective for monitoring the aging process and identifying individuals at high risk of functional deterioration.

Despite the growing recognition of IC as a key indicator of healthy aging, research on its environmental determinants remains scarce. A review on the associations among air pollution, climate change, and IC noted that extreme meteorological events, such as high temperatures, are related to functional deterioration in individual IC domains, yet studies focusing on overall IC as a multidimensional construct are notably lacking. To date, no study has examined the association between exposure to ETEs and the longitudinal trajectories of IC among middle-aged and older adults. To fill this research gap, the present study utilizes data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative longitudinal survey of Chinese adults aged 45 years and older, to investigate the association between ETE exposure and IC trajectories in this population. By integrating climate exposure data with a multidimensional assessment of IC within a longitudinal analytical framework, this study aims to provide novel scientific evidence on how environmental factors shape IC trajectories and to offer critical insights for the development of targeted climate adaptation and health intervention policies aimed at preserving the functional well-being of aging populations in the context of accelerating climate change.

Methods

This study used data from the CHARLS, a nationally representative prospective cohort study of community-dwelling adults aged 45 years and older in China. CHARLS adopted a multistage probability-proportional-to-size sampling design, recruiting participants from 150 county-level units across 28 provinces. The baseline survey was conducted in 2011, with biennial follow-up waves carried out thereafter; data from the 2013, 2015, 2018, and 2020 waves have been publicly released. Trained interviewers collected comprehensive information through face-to-face structured questionnaires, covering demographic characteristics, socioeconomic status, lifestyle factors, and health conditions. Standardized physical measurements and venous blood sample collection were also performed during the main survey waves. The study protocol was approved by the Biomedical Ethics Review Committee of Peking University (IRB00001052-11015), and written informed consent was obtained from all participants prior to each interview [17].

In the present study, we initially included 17,708 participants from the 2011 baseline survey of CHARLS. Participants were sequentially excluded according to the following criteria: (1) age < 45 or > 80 years (n = 1,153); (2) missing intrinsic capacity data (n = 14,577); (3) missing or implausible height or weight values (n = 5); and (4) lack of follow-up data for at least two years (n = 850). After these exclusions, a total of 1,393 participants remained for the final analysis. The detailed participant selection process is shown in Fig. 1. To assess potential selection bias, we compared baseline characteristics between included and excluded participants (Supplementary Table S1), with additional details on sample selection provided in Supplementary Material Text S1.

Fig. 1.

Flow chart of participant selection

IC

IC was assessed according to the World Health Organization framework for healthy aging, operationalized as a composite score derived from five domains: locomotion, sensory function, vitality, psychological well-being, and cognition. Each domain was scored from 0 to 2, yielding a total IC score ranging from 0 to 10, with higher scores indicating better intrinsic capacity. Locomotion was measured using the Short Physical Performance Battery (SPPB); sensory function combined self-rated vision and hearing; vitality was evaluated through peak expiratory flow and handgrip strength; psychological status was assessed with the 10-item Center for Epidemiologic Studies Depression Scale (CES‑D 10) [18, 19]; and cognition comprised tests of intellectual function and episodic memory. All variables were obtained from the CHARLS 2011, 2013, and 2015 waves. Detailed scoring algorithms are provided in Supplementary Material Text S2 and Table S2.

ETEs

ETEs were defined using city-specific daily apparent temperature, which integrates ambient temperature, relative humidity, and wind speed to capture the thermal stress perceived by the human body. Apparent temperature was calculated using the Steadman formula, with all meteorological data obtained from the China Meteorological Administration Land Data Assimilation System (CLDAS, version 3.0). For each city, percentile thresholds were determined based on the distribution of daily apparent temperature over the 2010–2015 reference period. A heatwave was defined as a period when the daily apparent temperature remained at or above the 95th percentile for at least three consecutive days, and a cold spell was defined as a period when the daily apparent temperature remained at or below the 5th percentile for at least three consecutive days. For each participant and each calendar year from 2011 to 2015, the annual number of heatwave days and cold spell days was calculated and matched according to their city of residence. The primary analysis adopted these moderately strict definitions, while a total of twelve alternative definitions combining different percentile thresholds (90th, 92.5th, 95th, and 97.5th for heatwaves; 10th, 7.5th, 5th, and 2.5th for cold spells) and minimum durations (2, 3, and 4 days) were employed in sensitivity analyses. Detailed definitions and the calculation formula are provided in Table S3 and Supplementary Material Text S3.

Covariates

Covariates were selected based on prior studies and the available information in the CHARLS database. Sociodemographic variables included age (continuous, years), sex (male/female), marital status (married vs. other, including never married, divorced, or widowed), residence (urban/rural), and educational level (categorized as no completion of primary school, Sishu/home school/elementary school, middle school, or high school and above).

Anthropometric and physiological measurements were obtained by trained examiners following standardized procedures. Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters (kg/m²). Waist circumference was measured in centimeters. Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times on the left arm using an automatic electronic sphygmomanometer (OMRON HEM-7112), and the average of the three readings was used for analysis. Pulse rate was recorded in beats per minute. Venous blood samples were collected after overnight fasting and assayed for serum creatinine (SCR), total cholesterol (TC), triglycerides, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and C-reactive protein (CRP); all biomarkers were expressed in mg/dL. Sleep duration was assessed by self-report as the average hours of sleep per night.

Chronic disease history was based on self-reported physician diagnoses and included hypertension, diabetes, cancer, lung disease, heart disease, and stroke. Hypertension was defined as SBP ≥ 140 mmHg and/or DBP ≥ 90 mmHg, or a self-reported physician diagnosis. Diabetes was defined as fasting blood glucose ≥ 126 mg/dL, or a self-reported physician diagnosis. Other chronic conditions were assessed solely by self-report.

Functional status was evaluated using self-reported difficulties in basic activities of daily living (ADL) and instrumental activities of daily living (IADL). Self-rated health was recorded on a five-point scale (excellent, very good, good, fair, or poor). Lifestyle factors included smoking status (never, former, current) and alcohol consumption (never, former, current).

Statistical analysis

Baseline characteristics of the study participants were summarized according to the three IC trajectory groups. Continuous variables were presented as median with interquartile range (Q1, Q3) and compared using the Kruskal-Wallis rank sum test; categorical variables were expressed as frequencies (percentages) and compared using Pearson’s chi-squared test or Fisher’s exact test, as appropriate.

These trajectory groups are shown in Fig. 2. To identify heterogeneous trajectories of IC, we applied K-means clustering (unsupervised learning) based on three repeated measurements of IC total scores (range 0–10) in 2011, 2013, and 2015. The five IC domains were equally weighted when calculating the overall IC score. This approach was adopted because no consensus weighting scheme has been established for the WHO intrinsic capacity framework, and equal weighting provides a transparent and comparable method that has been widely used in previous IC-related studies.

Fig. 2.

Trajectories of intrinsic capacity among middle-aged and older Chinese adults from 2011 to 2015

K-means clustering, an unsupervised machine learning technique, groups participants according to the similarity of their observed multivariate profiles by minimizing the total within-cluster sum of squares. In the present study, each participant’s three repeated IC measurements were treated as a longitudinal profile, and K-means clustering was used as a data-driven approach to identify subgroups with similar observed IC patterns across the three survey waves [18]. The three time-point values were standardized (Z-score) to give equal weight to each wave. Candidate solutions with two to five clusters were evaluated using the total within-cluster sum of squares (WCSS), the average silhouette coefficient, and the Calinski–Harabasz index. The WCSS values for the two-, three-, four-, and five-cluster solutions were 2567.78, 2105.40, 1759.64, and 1456.53, respectively; the corresponding average silhouette coefficients were 0.345, 0.272, 0.285, and 0.295, and the Calinski–Harabasz indices were 871.19, 683.57, 636.12, and 644.69, respectively. Although the two-cluster solution yielded the highest silhouette coefficient and Calinski–Harabasz index, the final number of clusters was selected based on a comprehensive assessment of the cluster validity indices, the distinctiveness of the resulting trajectories, and their clinical interpretability. The three-cluster solution was retained because it provided three distinct and clinically interpretable longitudinal patterns. Detailed cluster validity indices are presented in Supplementary Table S4. The elbow plot based on total WCSS across candidate cluster numbers (2–5 clusters) is presented in Supplementary Figure S1. The stability of the three-cluster K-means solution was assessed using 1,000 bootstrap resamples. The overall stability was 0.705, with cluster-specific stability values ranging from 0.695 to 0.714 (Supplementary Table S5). According to the mean IC scores of each cluster at the three time points, the groups were named: High-stable, Moderate-declining, and Low-declining. Because K-means is an algorithm-based rather than model-based clustering method, the resulting groups were interpreted as data-driven trajectory subgroups rather than latent trajectory classes. Figure S2 displays the mean scores of the five dimensions (locomotion, sensory, vitality, psychological, cognition) for the three trajectory groups. The area and shape of each polygon reflect the relative strengths and weaknesses of each group.

The association between cumulative heatwave days (2011–2015) and IC trajectory membership was examined using multinomial logistic regression models, with the High-stable group serving as the reference category. We constructed four nested models with progressive adjustment: Model 1 was unadjusted; Model 2 additionally adjusted for age and sex; Model 3 further adjusted for education, residence, household registration type (hukou), smoking status, and alcohol consumption; and Model 4 (fully adjusted) additionally included body mass index and the number of chronic diseases. Results were expressed as relative risk ratios (RRR) with 95% confidence intervals (CI). Subgroup analyses were performed to assess potential effect modification by sociodemographic and health-related characteristics; multiplicative interaction terms were added to the fully adjusted model, and P values for interaction were calculated using Wald tests. All statistical analyses were performed using R software (version 4.4.0). Model performance was assessed using Akaike information criterion (AIC), Bayesian information criterion (BIC), and McFadden pseudo-R² for the progressively adjusted multinomial logistic regression models. A two-sided P value < 0.05 was considered statistically significant.

Results

Baseline characteristics

Table 1 presents the baseline characteristics of the study participants according to the three intrinsic capacity trajectory groups (High stable, Moderate declining, Low declining). Significant differences across groups were observed for several sociodemographic and health related variables. Participants in the Low declining group were older (median age 68.0 years) compared with those in the High stable (64.0 years) and Moderate declining groups (64.0 years, P < 0.001). The proportion of females was highest in the Low declining group (50%) and lowest in the High stable group (42%, P = 0.022). Married participants were least prevalent in the Low declining group (83%) compared with the other two groups (87 90%, P = 0.005). Educational level differed markedly: 69% of the Low declining group had no primary school completion, versus 35% in the High stable group (P < 0.001).

Table 1.

Baseline characteristics of participants by intrinsic capacity trajectory groups

Characteristic	High-stable N = 5751	Moderate-declining N = 3681	Low-declining N = 4501	p-value2
Age, years	64.0 (61.0, 68.0)	64.0 (62.0, 68.0)	68.0 (63.0, 72.0)	< 0.001
Gender, n(%)				0.022
Female	240 (42%)	158 (43%)	225 (50%)
Male	335 (58%)	210 (57%)	225 (50%)
Marital status, n(%)				0.005
Others	60 (10%)	48 (13%)	78 (17%)
Married	515 (90%)	320 (87%)	372 (83%)
Residence, n(%)				0.11
Urban	211 (37%)	144 (39%)	145 (32%)
Rural	364 (63%)	224 (61%)	305 (68%)
Educational status, n(%)				< 0.001
No completion of primary school	201 (35%)	155 (42%)	311 (69%)
Sishu/home school/elementary school	221 (38%)	122 (33%)	100 (22%)
Middle school	103 (18%)	64 (17%)	30 (6.7%)
High school and above	48 (8.3%)	27 (7.3%)	9 (2.0%)
BMI, kg/m2	23.6 (21.2, 25.8)	22.7 (20.7, 25.6)	22.3 (20.2, 24.8)	< 0.001
Waist circumference, cm	86 (79, 93)	85 (78, 93)	84 (77, 91)	0.060
SBP, mmHg	128 (116, 143)	131 (120, 146)	131 (117, 150)	0.041
DBP, mmHg	73 (66, 81)	74 (67, 82)	73 (66, 83)	0.3
Pulse rate, beats/min	71 (64, 77)	70 (65, 78)	73 (65, 80)	0.016
SCR, mg/dL	0.79 (0.68, 0.92)	0.79 (0.68, 0.92)	0.77 (0.66, 0.89)	0.3
TC, mg/dL	196 (170, 217)	194 (170, 219)	192 (166, 219)	0.5
Triglycerides, mg/dL	109 (80, 156)	107 (73, 161)	106 (74, 143)	0.2
HDL-C, mg/dL	48 (40, 58)	49 (40, 60)	51 (41, 61)	0.019
LDL-C, mg/dL	121 (99, 140)	120 (95, 145)	114 (92, 140)	0.12
CRP, mg/dL	1.04 (0.57, 1.95)	1.30 (0.72, 2.63)	1.16 (0.64, 2.28)	0.005
Sleep, hours	6.00 (5.00, 7.50)	6.00 (5.00, 8.00)	6.00 (5.00, 8.00)	0.023
Heatwave days	81 (77, 84)	81 (76, 84)	81 (79, 85)	0.002
Cold spell days	84 (82, 86)	84 (82, 86)	84 (82, 86)	0.7
Hypertension, n(%)				0.2
No	276 (48%)	159 (43%)	192 (43%)
Yes	297 (52%)	209 (57%)	255 (57%)
Diabetes, n(%)				0.7
No	478 (84%)	310 (84%)	368 (82%)
Yes	94 (16%)	58 (16%)	80 (18%)
Cancer, n(%)				0.5
No	563 (98%)	363 (99%)	447 (99%)
Yes	9 (1.6%)	5 (1.4%)	3 (0.7%)
Lunge disease, n(%)				0.7
No	498 (87%)	317 (86%)	382 (85%)
Yes	77 (13%)	51 (14%)	68 (15%)
Heart disease, n(%)				0.2
No	473 (83%)	299 (81%)	384 (86%)
Yes	100 (17%)	68 (19%)	63 (14%)
Stroke, n(%)				0.2
No	565 (98%)	360 (98%)	434 (96%)
Yes	10 (1.7%)	8 (2.2%)	16 (3.6%)
IADL				< 0.001
No	480 (88%)	306 (89%)	300 (80%)
Yes	66 (12%)	37 (11%)	77 (20%)
ADL				< 0.001
No	501 (90%)	303 (89%)	303 (79%)
Yes	54 (9.7%)	39 (11%)	80 (21%)
Smoking status, n(%)				0.3
Never	307 (53%)	184 (50%)	247 (55%)
Ever	79 (14%)	45 (12%)	47 (10%)
Current	189 (33%)	139 (38%)	156 (35%)
Drinking status, n(%)				0.14
Never	298 (52%)	194 (53%)	264 (59%)
Ever	72 (13%)	49 (13%)	59 (13%)
Current	205 (36%)	125 (34%)	127 (28%)
Self-rated Health				< 0.001
Poor	13 (2.3%)	10 (2.7%)	26 (5.8%)
Fair	119 (21%)	87 (24%)	150 (33%)
Good	346 (60%)	216 (59%)	228 (51%)
Very Good	72 (13%)	40 (11%)	35 (7.8%)
Excellent	25 (4.3%)	15 (4.1%)	11 (2.4%)

Abbreviations: BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, SCR serum creatinine, TC total cholesterol, HDL-C highdensity lipoprotein cholesterol, LDL-C low density lipoprotein cholesterol, CRP C reactive protein, IADL instrumental activities of daily living, ADL activities of daily living

¹Median (Q1, Q3); n (%);²Kruskal-Wallis rank sum test; Pearson’s Chi-squared test; Fisher’s exact test

Metabolic and inflammatory markers also varied significantly. The Low declining group had lower BMI (median 22.3 kg/m²) and higher HDL-C (median 51 mg/dL) but also higher C reactive protein (CRP, median 1.16 mg/dL) and shorter sleep duration (median 6.0 h, all P < 0.05). Systolic blood pressure and pulse rate were slightly higher in the Low declining and Moderate declining groups (P = 0.041 and 0.016, respectively). Cumulative heatwave days (2011 2015) were highest in the Low declining group (median 81 days, P = 0.002), whereas cold spell days did not differ (P = 0.7).

Functional status at baseline showed strong associations with trajectory membership: the prevalence of IADL disability (20% vs. 11 12%, P < 0.001) and ADL disability (21% vs. 9 11%, P < 0.001) were substantially higher in the Low declining group. Consistently, participants in the Low-declining group reported poorer self-rated health, with higher proportions rating their health as Fair or Poor and lower proportions rating it as Excellent or Very good (P < 0.001). No significant differences were observed for waist circumference, diastolic blood pressure, serum creatinine, total cholesterol, triglycerides, LDL-C, hypertension, diabetes, cancer, lung disease, heart disease, stroke, smoking, or drinking status (all P > 0.05).

Intrinsic capacity trajectories

Based on the three repeated measurements of intrinsic capacity (2011, 2013, 2015), we identified three distinct trajectory groups using K means clustering (Table 2). The High-stable group (n = 575, 41.3%) had the highest baseline IC score (8.56 ± 0.93) and maintained a stable high level over the 4 year follow up (ΔIC= -0.57), representing the most favorable health trajectory. The Moderate-declining group (n = 368, 26.4%) started with a moderate baseline IC (8.35 ± 0.81), experienced a sharp initial decline followed by partial recovery (ΔIC= -0.67), indicating some physiological resilience. The Low-declining group (n = 450, 32.3%) consistently had the lowest IC scores across all time points (baseline 6.68 ± 1.07) and exhibited a progressive, irreversible decline (ΔIC= -1.24), representing the most vulnerable trajectory.

Table 2.

Characteristics of the three identified intrinsic capacity trajectory groups

Trajectory Group	N (%)	Baseline IC Score (Mean ± SD)	ΔIC (2011→2015)	Key Feature Description
High-stable	575 (41.3%)	8.56 ± 0.93	-0.57	Highest baseline IC level; remains stable and high over time, representing the most favorable health trajectory.
Moderate-declining	368 (26.4%)	8.35 ± 0.81	-0.67	Moderate baseline IC; experiences initial sharp decline followed by partial recovery, indicating some physiological resilience.
Low-declining	450 (32.3%)	6.68 ± 1.07	-1.24	Lowest baseline IC level; shows progressive and irreversible decline across all time points, representing the most vulnerable health trajectory.

IC scores range from 0 to 10, with higher scores indicating better intrinsic capacity. ΔIC (2011→2015) was calculated as IC-2015 minus IC-2011. The three trajectory groups were identified using K‑means clustering based on repeated IC measurements in 2011, 2013, and 2015. Group labels were assigned according to baseline IC level and subsequent change pattern

Association between ETEs and intrinsic capacity trajectories

Table 3 presents the results of multinomial logistic regression examining the association between cumulative heatwave days (2011–2015) and the three intrinsic capacity trajectories, using the High‑stable group as the reference. In the fully adjusted model (Model 4), each additional heatwave day was associated with a significantly increased risk of belonging to the Low‑declining trajectory (RRR = 1.035, 95% CI: 1.012–1.058, P = 0.003). The association remained robust across all four models with progressive adjustment (Model 1: RRR = 1.037, 95% CI: 1.016–1.058, P = 0.0005; Model 2: RRR = 1.043, 95% CI: 1.021–1.065, P < 0.001; Model 3: RRR = 1.036, 95% CI: 1.013–1.059, P = 0.0017).

Table 3.

Multinomial logistic regression analysis of cumulative heatwave exposure (2011–2015) and intrinsic capacity trajectories among middle-aged and older Chinese adults

Model	Trajectory	RRR (95% CI)	P value
Model 1	High-stable	Ref
	Moderate-declining	1.000 (0.980–1.021)	0.996
	Low-declining	1.037 (1.016–1.058)	0.0005
Model 2	High-stable	Ref
	Moderate-declining	1.001 (0.980–1.022)	0.938
	Low-declining	1.043 (1.021–1.065)	< 0.0001
Model 3	High-stable	Ref
	Moderate-declining	1.002 (0.981–1.023)	0.860
	Low-declining	1.036 (1.013–1.059)	0.0017
Model 4	High-stable	Ref
	Moderate-declining	1.002 (0.981–1.024)	0.839
	Low-declining	1.035 (1.012–1.058)	0.003

Data are presented as Relative Risk Ratio (RRR) with 95% confidence intervals (CI). Model 1 was unadjusted. Model 2 was adjusted for age and gender. Model 3 further adjusted for education, rural residence, household registration type (hukou), smoking and drinking status. Model 4 additionally adjusted for body mass index (BMI) and number of chronic diseases. The reference trajectory group is High stable. RRR > 1 indicates a higher risk of belonging to the specified trajectory (Moderate declining or Low declining) compared to the reference group. All models were fitted using the same complete case sample (n = 1,393). P values were calculated using Wald tests

To provide a more interpretable measure of cumulative exposure, we additionally quantified the effect associated with an interquartile range (IQR) increase in cumulative heatwave days. The IQR was 6 days (Q1 = 78 days; Q3 = 84 days), and a 6-day increase in cumulative heatwave exposure was associated with an RRR of 1.229 (95% CI: 1.074–1.403) for belonging to the Low-declining trajectory. The proportion of participants with high cumulative heatwave exposure, defined as exposure at or above the 75th percentile (≥ 84 days), was 28.5%, and the estimated population-attributable fraction (PAF) was 5.8% (Supplementary Table S6).

Consistent with these findings, the predicted probability of belonging to the Low-declining trajectory increased progressively across the observed range of cumulative heatwave exposure in the fully adjusted model. This exposure–response pattern is illustrated in Supplementary Figure S3, with other covariates held at their reference or median values.

No significant association was observed for the Moderate‑declining trajectory in any of the models (Model 4: RRR = 1.002, 95% CI: 0.981–1.024, P = 0.839). In separate analyses, cold spell days showed no statistically significant association with intrinsic capacity trajectories (see Supplementary Table S7). The full covariate estimates from the fully adjusted model (Model 4) are provided in Supplementary Table S8.These findings indicate that cumulative heatwave exposure specifically increases the risk of falling into the most vulnerable (Low‑declining) trajectory, whereas it does not significantly influence the Moderate‑declining trajectory. Model fit statistics for the progressively adjusted multinomial logistic regression models are presented in Supplementary Table S9. The AIC values decreased from Model 1 to Model 4, indicating improved model fit with progressive adjustment for covariates.

Subgroup analyses

As shown in Table 4, we performed subgroup analyses to explore whether the association between cumulative heatwave exposure (2011–2015) and the Low-declining trajectory differed across population strata.

Table 4.

Subgroup analysis for the association between cumulative heatwave exposure (2011–2015) and the low declining trajectory

Subgroup	RRR(95%CI)	P-value	P for Interaction
Overall	1.03 (1.01,1.06)	0.005
Age			0.844
< 65	1.04 (1.00,1.07)	0.066
≥ 65	1.03 (1.00,1.06)	0.078
Gender			0.146
Female	1.06 (1.03,1.10)	0.001
Male	1.02 (0.99,1.05)	0.318
Education			0.031
No completion of primary school	1.07 (1.04,1.11)	< 0.001
Sishu/home school/elementary school	0.99 (0.96,1.03)	0.678
Middle school	1.00 (0.93,1.09)	0.91
High school and above	1.06 (0.93,1.22)	0.384
Marriage			0.551
Unmarried	1.00 (0.93,1.07)	0.99
Married	1.04 (1.01,1.06)	0.005
Residence			0.024
Urban	0.99 (0.95,1.03)	0.472
Rural	1.05 (1.03,1.08)	< 0.001
Smoking			0.992
Never	1.04 (1.01,1.07)	0.01
Former	0.99 (0.90,1.09)	0.896
Current	1.03 (1.00,1.07)	0.085
Drinking			0.831
Never	1.04 (1.01,1.07)	0.009
Former	1.06 (0.97,1.16)	0.169
Current	1.01 (0.97,1.06)	0.53
Hypertension			0.311
No	1.02 (0.99,1.06)	0.237
Yes	1.04 (1.01,1.08)	0.006
Diabetes			0.923
No	1.03 (1.01,1.06)	0.013
Yes	1.05 (0.99,1.11)	0.134
Chronic diseases			0.106
No	1.04 (0.99,1.08)	0.111
Yes	1.03 (1.00,1.06)	0.022

Significant effect modification was observed for education level (P for interaction = 0.031) and residence (P for interaction = 0.024). Specifically, among participants with primary school education or below, each additional heatwave day was associated with a 7% higher risk of belonging to the Low declining trajectory (RRR = 1.07, 95% CI: 1.04–1.11, P < 0.001), whereas no significant associations were found for those with middle school, high school, or college education. Rural residents exhibited a significantly increased risk (RRR = 1.05, 95% CI: 1.03–1.08, P < 0.001), while urban residents did not (RRR = 0.99, 95% CI: 0.95–1.03, P = 0.472).No other subgroup variable showed statistically significant interaction (all P for interaction > 0.05).

Sensitivity analyses

We conducted several sensitivity analyses to evaluate the robustness of the association between cumulative heatwave exposure and the Low-declining trajectory (Supplementary Table S7). The primary analysis was based on cumulative heatwave days during 2011–2015. Overall, the association remained consistent when heatwave events were used as the exposure metric, whereas no significant association was observed for cold spell days or when the outcome was redefined as a two-year trajectory. Additional sensitivity analyses using different exposure windows (2011–2013 and 2013–2015) showed that the association was stronger for heatwave exposure during 2011–2013 (RRR = 1.092, 95% CI: 1.061–1.124, P < 0.001), whereas the 2013–2015 exposure window showed no significant association (RRR = 0.982, 95% CI: 0.961–1.002, P = 0.077). Detailed results are presented in Supplementary Table S10.

Discussion

In this study, we employed K-means clustering to identify three distinct intrinsic capacity trajectories over a 4-year period—High-stable, Moderate-declining, and Low-declining. Our findings demonstrated that cumulative heatwave exposure from 2011 to 2015 was significantly and specifically associated with an increased risk of belonging to the most vulnerable Low-declining trajectory. In the fully adjusted model, each additional heatwave day was associated with a 3.5% higher risk of following the Low-declining trajectory (RRR = 1.035, 95% CI: 1.012–1.058, P = 0.003), and this association remained robust across all progressively adjusted models. In contrast, cold spell exposure showed no significant association with IC trajectories. Subgroup analyses revealed that the detrimental effect of heatwaves was more pronounced among participants with lower educational attainment and those residing in rural areas. A series of sensitivity analyses using alternative exposure definitions and outcome metrics consistently supported the robustness of these findings. Collectively, this study provides novel evidence that cumulative heatwave exposure may accelerate the decline of intrinsic capacity in aging populations, with the most vulnerable subgroups bearing the greatest burden.

Our findings are consistent with previous evidence on the adverse health impacts of ETEs among older adults. Prior research has shown that heatwaves are positively associated with the risk of multimorbidity, with HRs ranging from 1.008 (95% CI: 1.002–1.013) to 1.021 (95% CI: 1.007–1.035). For cold spells, each additional day of exposure was associated with a 1.8% (HR: 1.018, 95% CI: 1.012–1.024) to 2.4% (HR: 1.024, 95% CI: 1.016–1.032) increased risk of multimorbidity [5]. In terms of frailty, Wang et al. reported among 4,955 middle-aged and older adults that cold spell exposure was significantly associated with frailty index progression; under the definition of the 5th percentile with ≥ 4 days duration, both exposure days (β = 0.802, 95% CI: 0.490–1.114) and events (β = 5.635, 95% CI: 3.276–7.995) were associated with accelerated frailty progression [19]. Regarding kidney function, studies have demonstrated that increased heatwave exposure was associated with a higher risk of rapid kidney function decline [20]. Furthermore, heatwave exposure has been linked to persistent or worsening depressive symptoms, with multimorbidity amplifying this effect [21]. Zhou et al. further found that increased heatwave exposure was associated with reduced sleep duration among 7,240 older adults, with females, urban residents, and individuals with chronic diseases identified as vulnerable populations [4]. Additionally, Yang et al. reported that specific heatwave definitions were significantly associated with incident diabetes (e.g., OR = 1.160 for ≥ 90th percentile over 3 days; OR = 1.224 for ≥ 97.5th percentile over 4 days), while improved air quality and higher green space coverage exhibited protective effects [22]. Although the estimated effect size for each additional heatwave day in our study was modest (RRR = 1.035), this finding should be interpreted in the context of cumulative environmental exposure rather than as the impact of a single isolated exposure day. Heatwaves are recurrent climate events, and older adults may experience repeated exposure over multiple seasons and years. Consequently, even relatively small increases in risk associated with each additional heatwave day may accumulate over time, resulting in a meaningful population-level burden of intrinsic capacity decline, particularly among vulnerable groups with limited adaptive capacity. Therefore, our findings underscore the importance of reducing cumulative heat exposure and strengthening long-term heat adaptation strategies for aging populations, rather than focusing solely on the marginal effect of a single additional heatwave day. In addition, sensitivity analyses using different exposure windows suggested that the timing of heatwave exposure may influence its association with IC trajectories, although further studies with more frequent assessments are needed to clarify these temporal patterns.

Our study found a significant and specific positive association between heatwave exposure and the risk of belonging to the most vulnerable Low-declining trajectory. Several mechanisms may underlie this finding. Exposure to extreme heat disrupts protein homeostasis, leading to the accumulation of misfolded proteins and activation of the heat shock response, which is primarily mediated by heat shock factor 1 and accompanied by the upregulation of molecular chaperones such as HSP70 and HSP90 [23]. Although this response is initially protective, chronic or repeated activation can exhaust the cellular stress response machinery, impair proteostasis, and contribute to progressive dysfunction in multiple organ systems [24]. Concurrently, heat stress activates the hypothalamic-pituitary-adrenal axis, resulting in elevated circulating cortisol levels, which over time promote muscle protein catabolism through the ubiquitin-proteasome pathway and inhibit protein synthesis via suppression of the mTOR signaling pathway, thereby accelerating the loss of muscle mass and strength—a potential contributor to declines in the locomotion and vitality domains of intrinsic capacity [25, 26]. These processes may be particularly relevant to IC domains related to locomotion and vitality, because preserved muscle function, physical capacity, and energy regulation are essential determinants of mobility and resilience in older adults.

Moreover, heat stress upregulates nuclear factor-kappa B signaling, a master transcriptional regulator of pro-inflammatory cytokines including interleukin-6 and tumor necrosis factor-alpha [27]. These inflammatory mediators can cross the blood-brain barrier and activate microglia, triggering neuroinflammation that leads to synaptic dysfunction, impaired neurogenesis, and accelerated cognitive decline [28, 29]. Therefore, inflammation-related neural injury may provide a potential biological explanation for the involvement of cognition and psychological dimensions within the broader decline of intrinsic capacity.

Importantly, heat exposure induces mitochondrial dysfunction and excessive production of reactive oxygen species (ROS) [30]; excess ROS can directly damage cellular macromolecules, including lipid peroxidation of cell membranes, oxidative DNA damage, and protein carbonylation [31]. In the central nervous system, mitochondrial dysfunction and oxidative damage are particularly detrimental to hippocampal neurons and dopaminergic pathways, which are essential for episodic memory and executive function—key components of the cognition domain of intrinsic capacity [32]. In addition, oxidative stress and vascular dysfunction may also affect sensory and cognitive functions, potentially influencing older adults’ ability to interact with and adapt to their environment. The elevated C-reactive protein levels observed in the Low-declining group in our study provide empirical support for this inflammatory pathway. Additionally, chronic inflammation accelerates skeletal muscle catabolism, further contributing to progressive loss of physical function. Together, these interconnected pathways provide a plausible mechanistic framework linking cumulative heatwave exposure with overall intrinsic capacity decline, potentially involving multiple IC domains rather than a single functional system.

In the present study, cold spell days showed no statistically significant association with intrinsic capacity trajectories (RRR = 1.02, 95% CI: 0.96–1.08, P = 0.554). However, the relatively wide confidence interval indicates that a modest association cannot be completely excluded, and the limited analytical sample size may have reduced statistical power to detect smaller effects of cold exposure. This null finding may reflect fundamentally different biological pathways engaged by cold versus heat stress. In addition, the cold spell definition used in this study may not fully capture region-specific cold stress, as populations living in subtropical areas may exhibit different patterns of cold adaptation compared with those in temperate regions. Cold stress primarily triggers acute cardiovascular and respiratory events through peripheral vasoconstriction, elevated blood pressure, and increased thrombotic risk, rather than the progressive, cumulative multisystem deterioration characteristic of heat-related damage [33]. This does not negate the established adverse health effects of cold spells; rather, it suggests that their impact on multidimensional intrinsic capacity trajectories may operate through mechanisms distinct from those of cumulative heat exposure, warranting further investigation. Our subgroup analyses showed that the association between heatwave exposure and the Low-declining trajectory was significantly modified by education level (P for interaction = 0.031) and residence (P for interaction = 0.024). These findings are consistent with the differential vulnerability hypothesis, which posits that the health impacts of environmental hazards are disproportionately borne by socioeconomically disadvantaged groups. The heightened vulnerability among participants with lower educational attainment may reflect inadequate health literacy regarding heat adaptation, a greater likelihood of engaging in outdoor manual labor with prolonged and unavoidable heat exposure, and economic constraints limiting access to air conditioning and cooling resources—all of which compound physiological susceptibility to heat stress [34]. The elevated risk among rural residents may be attributable to disparities in infrastructure and healthcare access, including less developed infrastructure and limited capacity to manage heat-related illnesses in rural areas [35]. Collectively, these environmental and structural factors result in rural residents bearing a disproportionately high burden of heat-related functional decline, underscoring the importance of geographically equitable climate adaptation strategies.

This study has several notable strengths. First, we used data from CHARLS, a nationally representative longitudinal survey of Chinese middle-aged and older adults, which employs rigorous multistage probability-proportional-to-size sampling, standardized data collection protocols, and comprehensive assessments spanning sociodemographic, anthropometric, lifestyle, and biomarker domains, thereby enhancing the generalizability of our findings to the Chinese aging population. Second, by employing trajectory modeling, we captured heterogeneous longitudinal patterns of intrinsic capacity over a 4-year period, rather than relying on single-time-point or change-score approaches, providing a more nuanced understanding of functional aging dynamics. Third, the integration of city-specific meteorological data with individual-level intrinsic capacity assessments within a longitudinal framework represents a novel contribution to the climate-health-aging literature, extending prior research that has predominantly focused on single-domain outcomes such as mortality, specific diseases, or individual functional measures.

At the same time, several limitations should be acknowledged. First, although K-means clustering is a data-driven approach that has been applied in previous longitudinal studies and showed acceptable reproducibility in our bootstrap stability analysis, it is an algorithm-based rather than model-based classification method. Unlike group-based trajectory modeling (GBTM) or latent class mixed models, K-means does not explicitly model within-individual temporal correlation, measurement error, or uncertainty in cluster membership. Therefore, the identified groups should be interpreted as data-driven trajectory subgroups rather than definitive latent trajectory classes. The resulting trajectory patterns may also be influenced by the selected clustering algorithm and methodological specifications. Future studies with more repeated measurements should further validate these trajectory classifications using complementary model-based approaches. In addition, cumulative heatwave exposure was modeled as a continuous linear variable, and potential non-linear exposure–response relationships were not examined. Future studies using flexible statistical approaches, such as restricted cubic splines, are warranted to further characterize the shape and potential thresholds of the association. Second, intrinsic capacity and several covariates, including chronic disease history, functional status, and lifestyle behaviors, were assessed through self-reported measures, which may be subject to recall or reporting bias. Another limitation is that the overall IC score was calculated by equally weighting the five domains. Although this approach is consistent with previous IC studies and facilitates interpretation, it assumes that each domain contributes equally to overall intrinsic capacity, which may not fully capture differences in their clinical relevance. Future studies should explore alternative weighting approaches and evaluate whether domain-specific weighting improves the assessment of IC. Third, of the 17,708 participants in the 2011 baseline survey, only 1,393 were included in the final analysis after sequential exclusions, primarily due to missing intrinsic capacity data. Participants with complete IC measurements differed from those excluded in several demographic and health-related characteristics, suggesting that selection bias cannot be completely excluded. However, the observed differences were not uniformly indicative of better or worse health status among included participants, and therefore the direction and magnitude of potential selection bias remain uncertain. Although we adjusted for several major sociodemographic and health-related factors, residual selection bias due to unmeasured determinants of participation and missingness may persist. Therefore, caution is warranted when extrapolating our findings to the broader older population. Fourth, meteorological data were assigned at the city level based on participants’ residential locations, which may not fully capture individual-level heat exposure variability. Specifically, this approach does not account for individual differences in indoor cooling conditions (e.g., access to air conditioning or other cooling facilities), time spent outdoors, occupational exposure, or personal adaptive behaviors during heatwave periods. In addition, within-city microclimate variations and potential seasonal migration may have resulted in exposure misclassification for some participants. Such non-differential exposure measurement errors may have attenuated the observed association between heatwave exposure and IC trajectory membership, leading to an underestimation of the observed association. However, if individuals with greater susceptibility to heat exposure systematically experienced different exposure patterns than those captured by city-level estimates, some degree of exposure-related distortion cannot be completely excluded. Future studies incorporating individual-level temperature monitoring, indoor environmental measurements, mobility information, or more refined spatial exposure assessment are needed to improve exposure characterization.

Abbreviations

ICl

Intrinsic capacity

CHARLS

China Health and Retirement Longitudinal Study

ETEs

Extreme temperature events

ADL

Activities of daily living

SPPB

Short Physical Performance Battery

BMI

Body mass index

SBP

Systolic blood pressure

DBP

Diastolic blood pressure

SCR

Serum creatinine

TC

Total cholesterol

HDL-C

Highdensity lipoprotein cholesterol

LDL-C

Low density lipoprotein cholesterol

CRP-C

Reactive protein

IADL

Instrumental activities of daily living

GBTM

Group-based trajectory modeling

Authors’ contributions

Lei Huang: Writing – original draft. Yu Huang: Writing – original draft. Bin Qian: Methodology, Formal analysis. Changlin Zhou: Data curation. YuNaHuang: Visualization. YuChen Wang: Investigation. Yong Sun: Resources. XingChun Huang: Software. Weilong Yao: Validation. YuHao Chen: Formal analysis. YuHui Wang: Conceptualization. Zhiqiang Luo: Conceptualization, Writing – review & editing, Supervision.

Funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This work was supported by the Natural Science Foundation of Gansu Province (No. 26JRRA366).

Data availability

The data used in this study were obtained from the China Health and Retirement Longitudinal Study (CHARLS), which is publicly available at http://charls.pku.edu.cn/. Researchers can access the data after registering and submitting a data use agreement to the CHARLS team.

Declarations

Ethics approval and consent to participate

This study was approved by the Biomedical Ethics Review Committee of Peking University (IRB00001052-11015). The study was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.