Abstract
Background
Depressive symptoms and cardiometabolic diseases (CMDs) are highly prevalent among middle-aged and older adults. However, longitudinal evidence on their reciprocal and time-varying associations remains limited. This study aimed to examine the bidirectional and temporal associations between depressive symptoms and CMDs among Chinese adults aged ≥ 45 years.
Methods
Data were derived from three waves (2015, 2018, 2020) of the nationally representative China Health and Retirement Longitudinal Study (CHARLS). Depressive symptoms were assessed with the 10-item Center for Epidemiologic Studies Depression Scale (CESD). CMDs included hypertension, diabetes, dyslipidemia, stroke, and heart disease, and were operationalised as continuous counts and alternative categorical definitions. The continuous CMDs count was specified as the primary exposure. Random-intercept cross-lagged panel models (RI-CLPM) were used to examine within-person bidirectional associations while separating stable between-person differences.
Results
At baseline (2015, T1), 18,837 participants were included, with 16,832 and 16,204 retained at the 2018 (T2) and 2020 (T3) waves, respectively. Mean depressive symptom scores increased from 7.88 (SD = 6.35) to 8.78 (SD = 6.49), while the mean number of CMDs rose from 0.81 (SD = 1.03) to 1.14 (SD = 1.21) across the study period. RI-CLPM identified strong autoregressive effects for CMDs across both intervals (β = 0.414 and 0.676, p < 0.001), suggesting that CMDs are highly stable across time. A time-dependent asymmetry was observed in the bidirectional associations. During T1–T2, higher depressive symptoms were prospectively associated with increases in CMDs (β = 0.076, p < 0.001), whereas CMDs did not predict subsequent depressive symptoms (β = 0.031, p > 0.05). During T2–T3, CMDs showed a stronger prospective association with later depressive symptoms (β = 0.103, p < 0.001) than the reverse pathway (β = 0.021, p < 0.01). These patterns were consistent across alternative CMDs specifications and sensitivity analyses.
Conclusion
This study provides longitudinal evidence of a dynamic and temporally asymmetric association between depressive symptoms and CMDs among middle-aged and older Chinese adults. The findings underscore the importance of considering time-varying and bidirectional processes and support the need for integrated, stage-specific approaches to mental and cardiometabolic health management in ageing populations.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12889-026-26374-9.
Keywords: Depressive symptoms, Cardiometabolic diseases, Random-intercept cross-lagged panel models, Chinese, Older adults
Introduction
Population ageing has emerged as a critical public health challenge worldwide, with China experiencing particularly rapid demographic shifts. By the end of 2024, an estimated 310 million Chinese citizens (22.0% of the population) were aged ≥ 60 years, including 220 million (15.6%) aged ≥ 65 years [1]. Amid demographic ageing and shifting lifestyles, the prevalence of cardiometabolic diseases (CMDs), including hypertension, diabetes, coronary heart disease, and stroke [2], has been steadily increasing. Cardiometabolic multimorbidity (CMM), defined as the coexistence of two or more CMDs in a single individual [3], represents one of the most prevalent and stable multimorbidity patterns, substantially elevating risks of dementia, disability, premature mortality, and healthcare burden [3–9].
Depression is another leading public health concern and represents the largest contributor to disability among mental disorders worldwide [10]. Recent estimates suggest that over 330 million individuals are affected globally [11]. In China, the prevalence of depressive symptoms among adults aged ≥ 60 years has risen markedly in the past decade, with detection rates increasing from 36.8% in 2011 to 44.5% in 2018 [12]. Late-life depression is independently associated with disability, cognitive decline, and all-cause mortality [13, 14], compounding the challenges of population ageing.
An extensive body of evidence suggests that depression and CMDs are interrelated. Depression has been shown to increase risks of incident hypertension, diabetes complications, stroke, and coronary events [15–17], while CMDs are frequently followed by subsequent onset of depressive symptoms [18, 19]. Moreover, the accumulation of CMDs shows a dose-dependent association with depressive symptom severity [20], and their co-occurrence exacerbates adverse outcomes including premature mortality and economic burden [21, 22]. Therefore, CMDs are increasingly conceptualised as an integrated indicator of cardiometabolic disease burden, operationalised as the count of CMDs and cardiometabolic multimorbidity (CMM; ≥2 CMDs), rather than as isolated chronic diseases. This approach better reflects the real-world clustering of cardiometabolic conditions that share common behavioural and pathophysiological pathways and are often managed concurrently in clinical practice and primary care settings. Prior studies have consistently shown that a greater cardiometabolic burden is associated with increased depressive symptom severity and adverse health trajectories [20, 23, 24]. Evidence from both cross-sectional and longitudinal studies suggests that higher cardiometabolic burden predicts subsequent depressive symptoms [20, 23], while depressive symptoms in turn are prospectively associated with developing CMDs [25, 26]. However, longitudinal evidence simultaneously capturing the reciprocal and time-varying dynamics between depressive symptoms and cardiometabolic burden remains limited.
Despite accumulating evidence linking depressive symptoms with individual cardiometabolic conditions, research focusing on cardiometabolic disease burden, defined as the number of CMDs or CMM, and its reciprocal, time-dependent relationship with depressive symptoms remains insufficient. Many studies have examined depression in relation to single CMDs or have treated CMDs as isolated exposures, thereby overlooking the fact that these diseases frequently co-occur and jointly shape individuals’ mental health in midlife and later life. Cardiometabolic burden better captures the cumulative physiological strain, shared pathophysiological mechanisms, and clinical complexity associated with clustered cardiometabolic conditions than individual diseases considered in isolation. Moreover, prior research has rarely investigated whether and how the direction and strength of the association between depressive symptoms and cardiometabolic disease burden may vary over time. This limitation constrains our understanding of the dynamic interplay between mental and cardiometabolic health across the ageing process and highlights the need for longitudinal approaches that explicitly model reciprocal and time-varying associations. Such evidence is particularly relevant in light of national policy priorities, such as China’s Healthy Ageing Initiative [27], which explicitly calls for integrated management of cardiovascular–metabolic and mental health conditions.
The middle-aged population (45–64 years) represents a critical window for prevention, as the burden of chronic diseases in this age group already surpasses that among older adults [28]. However, longitudinal evidence on the reciprocal relationship between depressive symptoms and CMDs across midlife and older age remains limited, particularly with respect to disentangling stable between-person differences from within-person changes over time. To address this gap, the present study utilised random-intercept cross-lagged panel models (RI-CLPM) to examine the bidirectional and time-dependent associations between depressive symptoms and CMDs among adults aged ≥ 45 years. By separating stable trait-like individual differences from within-person fluctuations, RI-CLPM provide a more rigorous framework for assessing temporal ordering and dynamic reciprocity than traditional cross-lagged models. Specifically, we aimed to (1) investigate whether within-person increases in depressive symptoms predict subsequent changes in cardiometabolic disease burden, (2) examine whether changes in cardiometabolic disease burden predict subsequent changes in depressive symptoms, and (3) evaluate whether the direction and strength of these associations vary across follow-up intervals.
Methods
Data source
This study utilised data from the China Health and Retirement Longitudinal Study (CHARLS) collected in the 2015, 2018, and 2020 survey waves. CHARLS is a nationally representative longitudinal survey designed to provide high-quality microdata on households and individuals aged 45 years and older in China, facilitating research on population ageing and related health and socioeconomic issues. The study employs a multistage, stratified probability-proportional-to-size sampling strategy. The national baseline survey was conducted in 2011, encompassing 150 county-level and 450 village-level units, with 10,257 households and 17,708 individuals enrolled [29]. Follow-up surveys have since been conducted biennially or triennially, including rounds in 2013, 2015, 2018, and 2020. Comprehensive details of the CHARLS design and methodology are documented elsewhere [30]. All participants or their guardians provided written informed consent for participation in both the baseline and follow-up surveys. Consent was obtained prior to the administration of the study questionnaire. The study received approval from the Biomedical Ethics Committee of Peking University (IRB00001052-11015).
Sample sizes for the analysis waves were 21,097 (T1, 2015 wave), 19,816 (T2, 2018 wave), and 19,395 (T3, 2020 wave). Following the inclusion and exclusion criteria outlined in Fig. 1, the final analytical sample consisted of 18,837 participants at baseline (T1), 16,832 at T2 (2,005 samples were lost to follow-up), and 16,204 at T3 (2,633 samples were lost to follow-up).
Fig. 1.
Flowchart of sample selection and data availability across three CHARLS waves. Note: The analytical cohort was defined at baseline in 2015 after excluding individuals with no CESD or CMDs measurements across all three waves. Random-intercept cross-lagged panel models were estimated using full information maximum likelihood (FIML), allowing participants with partially missing data at follow-up waves to be retained in the analysis
The measurement of depressive symptoms
Depressive symptoms were assessed using the 10-item Center for Epidemiologic Studies Depression Scale (CESD) [31]. Respondents rated the frequency of symptoms experienced during the preceding week on a 4-point scale: rarely or none of the time (< 1 day; score 0), some or a little of the time (1–2 days; score 1), occasionally or a moderate amount of the time (3–4 days; score 2), and most or all of the time (5–7 days; score 3). Total scores range from 0 to 30. Consistent with established thresholds in a prior study [32], a score ≥ 10 was used to indicate the presence of clinically significant depressive symptoms. The Chinese CESD has demonstrated good reliability and validity [33, 34].
The measurement of cmds
In this study, CMDs encompassed five conditions: hypertension, diabetes, dyslipidemia (elevation of low-density lipoprotein, triglycerides, and total cholesterol, or a low high-density lipoprotein level), stroke, and heart disease (including myocardial infarction, coronary heart disease, angina, congestive heart failure, and other cardiac disorders). Specifically, participants were classified as positive for a given condition if they responded “Yes” to the question, “Have you been diagnosed with (conditions listed below, read one by one) by a doctor?” This doctor-diagnosed, self-reported approach has been widely adopted in epidemiological studies and represents a pragmatic and reliable strategy for identifying chronic diseases in large-scale population surveys [24, 30, 35, 36].
Based on the above, three measures were derived: (1) the number of CMDs as a continuous variable, which was specified as the primary exposure; (2) categorized as none, single, or multiple (CMM); and (3) categorized as none or single, and multiple (CMM).
The measurement of control variables
The fully adjusted models controlled for a range of sociodemographic and health-related covariates, including age (in years, presented as mean with standard deviation), sex (male vs. female), marital status (married vs. others, including never married, divorced, or widowed), place of residence (urban vs. rural), and educational level (below primary school, primary school, junior high school, high school and above). In addition, annual per capita expenditure, health insurance status (yes vs. no), and past-year hospitalization status (yes vs. no) were included to further account for participants’ economic situation, access to healthcare, and recent healthcare utilization. Annual per capita expenditure was calculated by dividing the household’s total expenditure in the past year by the number of household members, and participants were subsequently categorized into four levels (Level 1 [lowest] to Level 4 [highest]) based on quartiles of annual per capita expenditure. Health insurance status was defined as having any form of medical insurance. Past-year hospitalization status was assessed based on self-reported hospitalization during the past year. All covariates were assessed at baseline (T1, 2015 wave).
Statistical analysis
Descriptive analyses were performed to characterize the study population at baseline. Continuous variables were summarized using means and standard deviations, while categorical variables were presented as frequencies and percentages. Descriptive analyses were conducted using Stata MP 17.0.
We examined the bidirectional longitudinal associations between depressive symptoms and CMDs using RI-CLPM within the structural equation modeling framework. Compared with traditional cross-lagged panel models, RI-CLPM explicitly separate stable between-person differences from within-person fluctuations over time, thereby providing a more appropriate framework for disentangling reciprocal longitudinal associations between depressive symptoms and CMDs. The primary model specified both CESD (depressive symptoms) and CMDs as continuous variables. To examine the robustness of the findings to alternative operationalizations of CMDs, two additional models were estimated, in which CMDs were specified as a three-category ordinal variable (0, 1, ≥ 2 CMDs) and as a binary variable (< 2 vs. ≥2 CMDs), respectively. All models were estimated using the same fitting strategy to facilitate comparability across different CMDs specifications.
The conceptual RI-CLPM is illustrated in Fig. 2. In this model, each observed CESD and CMDs score was decomposed into a stable between-person component, captured by a random intercept, and a within-person component representing deviations from the individual’s expected level at each wave. Within-person CESD deviations at each wave were regressed on their prior within-person CESD deviations and prior CMDs deviations, and within-person CMDs deviations were regressed on their prior CMDs deviations and prior CESD deviations. Concurrent within-person associations between CESD and CMDs at the same wave were modeled by allowing correlations between their within-person residuals. Random intercepts for CESD and CMDs were allowed to correlate, capturing stable between-person associations across the three waves.
Fig. 2.
Conceptual Random-Intercept (RI) Cross-Lagged Panel Model of CESD and CMDs. Note: Circles labeled T1, T2, and T3 represent within-person latent deviations from individuals’ stable trait levels captured by the random intercepts. Parameters d1–d3 represent concurrent within-person associations between CESD and CMDs at the same wave, and parameter e denotes the correlation between the random intercepts
Baseline covariates, including age (mean-centered), sex, marital status, place of residence, educational level, annual per capita expenditure, health insurance status, and past-year hospitalization status, were specified as predictors of the random intercepts for CESD and CMDs. This specification allowed covariates to account for stable between-person differences across all three waves, rather than adjusting outcomes at a single time point.
We conducted two sensitivity analyses to evaluate the consistency of the results: (1) specifying depressive symptoms as a binary variable (CESD score ≥ 10) while retaining CMDs as a continuous count; and (2) re-estimating the primary model (continuous CESD and continuous CMDs) in a complete-case subsample (n = 9,988) after excluding all participants with any missing data in depression or CMDs across all waves. These sensitivity analyses were conducted within the same analytical framework as the primary analyses.
All RI-CLPM were estimated using robust maximum likelihood (MLR) in Mplus version 8.3, which provides robust standard errors and accommodates missing data under the missing-at-random assumption via full information maximum likelihood. Model fit was evaluated using the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), and Root Mean Square Error of Approximation (RMSEA). CFI and TLI values greater than 0.90 were considered acceptable, and greater than 0.95 indicated excellent fit; RMSEA values below 0.08 indicated good fit. Standardized coefficients (STDYX) were reported for all paths to facilitate interpretation.
Results
Baseline characteristics
At baseline (T1), a total of 18,837 participants were included, with 16,832 and 16,204 remaining at T2 and T3, respectively (Table 1). The mean (SD) age of participants was similar across waves. Approximately half of the sample was women (47.6%–48.6% across waves). Most participants were married (87.4%–88.6%), and the majority resided in rural areas (60.4%–62.3%). Educational attainment was generally low, with more than 40% reporting education below primary school.
Table 1.
Demographic Characteristics, CESD, and CMDs of in each wave
| Characteristic | T1 (n = 18837) | T2 (n = 16832) | T3 (n = 16204) |
|---|---|---|---|
| Age, mean (SD), years | 60.06 (9.81) | 59.84 (9.64) | 59.43 (9.42) |
| Sex, No. (%) | |||
| Male | 9690 (51.4) | 8754 (52.0) | 8496 (52.4) |
| Female | 9147 (48.6) | 8078 (48.0) | 7708 (47.6) |
| Marital status, No. (%) | |||
| Others* | 2370 (12.6) | 2018 (12.0) | 1850 (11.4) |
| Married | 16,467 (87.4) | 14,799 (88.0) | 14,354 (88.6) |
| Place of residence, No. (%) | |||
| Urban | 7456 (39.6) | 6435 (38.2) | 6103 (37.7) |
| Rural | 11,381 (60.4) | 10,397 (61.8) | 10,101 (62.3) |
| Educational level, No. (%) | |||
| Below primary school | 7779 (41.3) | 6974 (41.4) | 6702 (41.4) |
| Primary school | 5235 (27.8) | 4730 (28.1) | 4543 (28.0) |
| Junior high school | 3687 (19.6) | 3284 (19.5) | 3189 (19.7) |
| High school and above | 2136 (11.3) | 1844 (11.0) | 1770 (10.9) |
| Annual per capita expenditure, No. (%) | |||
| Level 1 (lowest) | 4709 (25.0) | 4288 (25.5) | 4106 (25.3) |
| Level 2 | 4710 (25.0) | 4262 (25.3) | 4083 (25.2) |
| Level 3 | 4710 (25.0) | 4157 (24.7) | 3999 (24.7) |
| Level 4 (highest) | 4708 (25.0) | 4125 (24.5) | 4016 (24.8) |
| Health insurance, No. (%) | |||
| No | 1555 (8.3) | 1358 (8.1) | 1262 (7.8) |
| Yes | 17,282 (91.7) | 15,474 (91.9) | 14,942 (92.2) |
| Past-year hospitalization status, No. (%) | |||
| No | 16,256 (86.3) | 14,598 (86.7) | 14,128 (87.2) |
| Yes | 2581 (13.7) | 2234 (13.3) | 2076 (12.8) |
| CESD score, mean (SD)** | 7.88 (6.35) | 8.50 (6.47) | 8.78 (6.49) |
| % CESD score ≥ 10** | 32.7 | 37.3 | 38.6 |
| CMDs count, mean (SD)** | 0.81 (1.03) | 1.05 (1.17) | 1.14 (1.21) |
| CMDs count, No. (%) | |||
| 0 | 7729 (41.0) | 7113 (42.3) | 6412 (39.6) |
| 1 | 4132 (21.9) | 4812 (28.6) | 4560 (28.1) |
| ≥2 | 3295 (17.5) | 4884 (29.0) | 5217 (32.2) |
| Missing | 3681 (19.5) | 23 (0.1) | 15 (0.1) |
| % CMDs count ≥ 2** | 21.7 | 29.1 | 32.2 |
CESD Center for Epidemiologic Studies Depression Scale, CMDs cardiometabolic diseases. * Includes never married, divorced, and widowed. ** Percentages and mean (SD) values were calculated among participants with non-missing data
The mean CESD score increased over time, from 7.88 (6.35) at T1 to 8.78 (6.49) at T3, and the proportion scoring above the threshold of 10 rose from 32.7% to 38.6%. Meanwhile, the average number of CMDs also increased, from 0.81 (1.03) at T1 to 1.14 (1.21) at T3. The percentage of participants with two or more CMDs rose markedly from 21.7% to 32.2%, indicating a progressive burden of multimorbidity.
Cross-lagged models with CESD and cmds as continuous variables
Figure 3 presents the results of the RI-CLPM examining the within-person bidirectional associations between depressive symptoms and CMDs, with both variables specified as continuous measures. The model demonstrated an excellent fit to the data (CFI = 0.995; TLI = 0.990; RMSEA = 0.019; SRMR = 0.011) after adjusting for covariates. Associations between sociodemographic and health-related covariates and the stable trait components (random intercepts) of depressive symptoms and CMDs are reported in eTable 1. With respect to within-person autoregressive effects, the path from T1 to T2 depressive symptoms was not statistically significant (standardized β = 0.005; p > 0.05), whereas depressive symptoms showed a significant autoregressive effect from T2 to T3 (β = 0.107; p < 0.001). CMDs exhibited strong and statistically significant autoregressive effects across both intervals (β = 0.414 and 0.676; p < 0.001), indicating substantial temporal persistence over time. Regarding within-person cross-lagged associations, a clear directional pattern emerged across the two intervals. During the earlier interval (T1–T2), higher depressive symptoms at T1 were significantly associated with higher CMDs at T2 (standardized β = 0.076, p < 0.001), whereas CMDs at T1 were not significantly associated with subsequent depressive symptoms at T2 (β = 0.031, p > 0.05). During the later interval (T2–T3), bidirectional associations between depressive symptoms and CMDs were observed. Depressive symptoms at T2 were positively associated with CMDs at T3 (β = 0.021, p < 0.01), and CMDs at T2 were also significantly associated with higher depressive symptoms at T3, with a substantially larger magnitude (β = 0.103, p < 0.001). Notably, the prospective association from CMDs to depressive symptoms during this later interval was markedly stronger than the reverse pathway.
Fig. 3.
Results of Random-Intercept (RI) Cross-Lagged Panel Model of CESD and CMDs. Note: Dashed lines indicate non-significant associations. The model controlled for baseline age, sex, marital status, place of residence, educational level, annual per capita expenditure, health insurance status, and past-year hospitalization status
Taken together, these findings indicate a time-dependent asymmetry in the within-person bidirectional associations, whereby depressive symptoms preceded subsequent changes in CMDs during the earlier interval, whereas CMDs showed stronger prospective associations with depressive symptoms during the later interval.
Table 2 presents the results of the RI-CLPM examining within-person associations between depressive symptoms and CMDs across three waves under alternative categorical specifications of CMDs. Both models controlled for baseline age, sex, marital status, place of residence, educational level, annual per capita expenditure, health insurance status, and past-year hospitalization status. Both models demonstrated excellent fit to the data (Model 1: CFI = 0.996, TLI = 0.992, RMSEA = 0.018, SRMR = 0.011; Model 2: CFI = 0.994, TLI = 0.989, RMSEA = 0.018, SRMR = 0.011). Overall, the pattern of results was highly consistent across the ordinal and binary CMDs specifications. Both models revealed a time-dependent asymmetry in the within-person associations. Specifically, during the earlier interval (T1–T2), depressive symptoms were prospectively associated with subsequent CMDs (all p < 0.01), whereas the reverse pathway from CMDs to depressive symptoms was not significant (all p > 0.05). During the later interval (T2–T3), CMDs showed prospective associations with subsequent depressive symptoms in both models (all p < 0.001). However, the pathway from depressive symptoms to CMDs in this later interval (T2–T3) was significant in the ordinal CMDs model (β = 0.021, p = 0.009) but was not significant in the binary CMDs model (β = 0.016, p = 0.090).
Table 2.
RI-CLPM results for CESD and cmds using ordinal and binary cmds definitions
| Path labela | Path relationship | Model 1 | Model 2 | ||
|---|---|---|---|---|---|
| Standardized β | p value | Standardized β | p value | ||
| a1 | T1 CESD → T2 CESD | 0.013 | 0.512 | −0.006 | 0.771 |
| a2 | T2 CESD → T3 CESD | 0.103 | < 0.001 | 0.106 | < 0.001 |
| b1 | T1 CMDs → T2 CMDs | 0.437 | < 0.001 | 0.346 | < 0.001 |
| b2 | T2 CMDs → T3 CMDs | 0.685 | < 0.001 | 0.565 | < 0.001 |
| c1 | T1 CESD → T2 CMDs | 0.087 | < 0.001 | 0.042 | 0.006 |
| c2 | T2 CESD → T3 CMDs | 0.021 | 0.009 | 0.016 | 0.090 |
| c3 | T1 CMDs → T2 CESD | −0.013 | 0.651 | −0.034 | 0.119 |
| c4 | T2 CMDs → T3 CESD | 0.111 | < 0.001 | 0.064 | < 0.001 |
| d1 | T1 CESD WITH T1 CMDs | 0.030 | 0.301 | −0.030 | 0.156 |
| d2 | T2 CESD WITH T2 CMDs | 0.110 | < 0.001 | 0.054 | 0.001 |
| d3 | T3 CESD WITH T3 CMDs | 0.069 | < 0.001 | 0.046 | < 0.001 |
| e | RI-CESD WITH RI-CMDs | 0.215 | < 0.001 | 0.209 | < 0.001 |
CESD Center for Epidemiologic Studies Depression Scale, CMDs cardiometabolic diseases, RI Random-Intercept
aThe path labels correspond one-to-one with those in Figure 2
Model 1 treated CESD as a continuous variable and CMDs as an ordinal variable categorized as 0, 1, and ≥2 CMDs. Model 2 treated CESD as a continuous variable and CMDs as a binary variable categorized as <2 and ≥2 CMDs
All models controlled for baseline age, sex, marital status, place of residence, educational level, annual per capita expenditure, health insurance status, and past-year hospitalization status
Sensitivity analyses
Table 3 presents the results of sensitivity analyses conducted to assess the robustness of the main findings under alternative model specifications. In Model 3, depressive symptoms were specified as a binary variable (CESD score ≥ 10), while CMDs were retained as a continuous count variable. In Model 4, the primary RI-CLPM was re-estimated in a complete-case subsample after excluding participants with missing depressive symptoms or CMDs data across any wave. Both models controlled for baseline age, sex, marital status, place of residence, educational level, annual per capita expenditure, health insurance status, and past-year hospitalization status. Both sensitivity models demonstrated excellent fit to the data (Model 3: CFI = 0.996, TLI = 0.993, RMSEA = 0.015, SRMR = 0.010; Model 4: CFI = 0.996, TLI = 0.992, RMSEA = 0.020, SRMR = 0.011). Overall, the pattern of results from both sensitivity analyses was highly consistent with the primary models. Both models reproduced the time-dependent asymmetry in the within-person associations, whereby depressive symptoms were prospectively associated with subsequent CMDs during the earlier interval (T1–T2), whereas CMDs showed stronger prospective associations with subsequent depressive symptoms during the later interval (T2–T3). These findings further support the robustness of the main conclusions to alternative outcome definitions and sample restrictions.
Table 3.
Sensitivity analyses of the RI-CLPM between CESD and cmds across three waves
| Path labela | Path relationship | Model 3 | Model 4 | ||
|---|---|---|---|---|---|
| Standardized β | p value | Standardized β | p value | ||
| a1 | T1 CESD → T2 CESD | 0.005 | 0.749 | 0.023 | 0.329 |
| a2 | T2 CESD → T3 CESD | 0.068 | < 0.001 | 0.107 | < 0.001 |
| b1 | T1 CMDs → T2 CMDs | 0.420 | < 0.001 | 0.443 | < 0.001 |
| b2 | T2 CMDs → T3 CMDs | 0.678 | < 0.001 | 0.689 | < 0.001 |
| c1 | T1 CESD → T2 CMDs | 0.060 | < 0.001 | 0.080 | < 0.001 |
| c2 | T2 CESD → T3 CMDs | 0.021 | 0.007 | 0.032 | 0.001 |
| c3 | T1 CMDs → T2 CESD | −0.010 | 0.710 | −0.012 | 0.700 |
| c4 | T2 CMDs → T3 CESD | 0.069 | < 0.001 | 0.107 | < 0.001 |
| d1 | T1 CESD WITH T1 CMDs | −0.010 | 0.710 | −0.002 | 0.957 |
| d2 | T2 CESD WITH T2 CMDs | 0.073 | < 0.001 | 0.100 | < 0.001 |
| d3 | T3 CESD WITH T3 CMDs | 0.052 | < 0.001 | 0.062 | < 0.001 |
| e | RI-CESD WITH RI-CMDs | 0.188 | < 0.001 | 0.174 | < 0.001 |
CESD Center for Epidemiologic Studies Depression Scale, CMDs cardiometabolic diseases, RI Random-Intercept
aThe path labels correspond one-to-one with those in Fig. 2
Model 3 and Model 4 present sensitivity analyses assessing the robustness of the main findings. In Model 3, CESD were specified as a binary variable (CESD score ≥ 10), while CMDs was retained as a continuous count variable. Model 4 re-estimated the primary model with continuous CESD and continuous CMDs in a complete-case subsample (n = 9,988), excluding all participants with missing data on CESD or CMDs across any wave. Standardized coefficients (β) are reported
All models controlled for baseline age, sex, marital status, place of residence, educational level, annual per capita expenditure, health insurance status, and past-year hospitalization status
Discussion
This large population-based longitudinal study provides evidence for a dynamic and temporally asymmetric association between depressive symptoms and CMDs. Depressive symptoms were prospectively associated with subsequent CMDs during the earlier interval, whereas CMDs showed stronger prospective associations with depressive symptoms during the later interval, indicating a time-dependent asymmetry rather than a static bidirectional relationship. The magnitude and direction of these associations varied across successive observation intervals, highlighting that the interplay between depressive symptoms and CMDs evolves over time.
Our results extend prior research in several important ways. Previous studies have demonstrated reciprocal associations between depression and individual CMDs, such as hypertension [37], metabolic syndrome [38, 39], and type 2 diabetes [40]. However, much of the existing literature has focused on unidirectional pathways or single diseases, with inconsistent evidence regarding CMDs as predictors of depressive symptoms [41, 42]. By applying RI-CLPM in a nationally representative cohort, our study explicitly disentangled stable between-person differences from within-person temporal dynamics, demonstrating that depressive symptoms and CMDs are prospectively associated with each other in a time-dependent and asymmetric manner. Depressive symptoms were more strongly associated with subsequent changes in CMDs during the earlier interval, whereas CMDs showed stronger prospective associations with depressive symptoms during the later interval. Our findings are consistent with evidence from studies across different stages of the life course, which shows that depression often emerges in adolescence or early adulthood [43], cardiometabolic disorders usually manifest later in life [44]. Although our sample comprised middle-aged and older adults, accumulating evidence suggests that depressive symptoms often have their origins earlier in the life course and may persist into midlife, thereby shaping subsequent cardiometabolic health. Thus, in midlife, depression may be more closely linked to subsequent changes in CMDs through behavioral and physiological pathways, whereas in later life, the accumulating burden of CMDs may become more salient for mental health. This dynamic pattern advances beyond prior cross-sectional and conventional longitudinal studies and underscores the importance of considering temporal ordering when examining the interplay between mental and cardiometabolic health.
Notably, depressive symptoms remain substantially under-recognised in primary care settings [45, 46]. In routine clinical practice, common depressive symptoms such as sleep problems, fatigue, and low energy often overlap with the physical symptoms of CMDs and are therefore frequently attributed to somatic illness rather than depression, particularly in primary care settings [47–49]. Previous studies have shown that this symptom overlap contributes to delayed or missed diagnosis of depression among individuals with chronic physical conditions, especially in primary care settings where consultation time is limited and clinical encounters are mainly focused on physical disease management [49]. As a result, depression may remain undetected and untreated in populations at high cardiometabolic risk, thereby posing a major barrier to early identification and timely intervention.
This study demonstrates a dynamic, time-dependent, and asymmetric coupling between depressive symptoms and CMDs/CMM, with important implications for clinical practice and public health. Building on these findings, our results support a three-pronged approach centred on screening, integrated diagnosis and treatment, and risk-stratified monitoring to address the psychometabolic vicious cycle and promote healthy ageing. First, routine depression screening could be integrated into national primary public health services for adults aged ≥ 45 years, with regular mental health assessments conducted through primary care facilities to enable early detection of psychometabolic comorbidities. Given the frequent under-recognition of depression and the overlap between depressive and somatic symptoms in primary care, depression screening in primary care should move beyond nonspecific symptom inquiry and incorporate brief, standardised tools that are feasible for routine use. Targeted screening protocols may prioritise CMDs patients presenting with persistent sleep problems, unexplained fatigue, or functional decline, even in the absence of overt psychological complaints. Embedding short, validated screening instruments within chronic disease follow-up visits, together with basic training for primary care providers to distinguish affective symptoms from somatic disease burden, could substantially improve recognition and early intervention for depression in this high-risk population. Second, a community-based integrated care pathway for “cardiovascular–cerebral–psychiatric” multimorbidity may be beneficial, leveraging family physician teams with embedded psychiatric support to deliver coordinated pharmacological and behavioural interventions for patients with CMM. Third, a risk-stratified depression monitoring system could be considered based on baseline CMDs burden, recognizing that sustained depressive trajectories significantly elevate both diabetes incidence and subsequent cardiovascular events [50]. This proposed framework aligns closely with the temporal patterns observed in the present findings, whereby depressive symptoms are more strongly associated with subsequent changes in CMDs during earlier stages, whereas the cumulative burden of CMDs appears to be more salient for depressive symptoms at later stages. Interventions tailored to these temporal dynamics may help mitigate the harmful cycle between mental and physical health, offering a potentially scalable model for ageing societies in China and worldwide.
Several limitations should be considered. First, CMDs were largely self-reported, potentially leading to underestimation of prevalence and associations. Second, although cardiometabolic disease burden was quantified, we were unable to examine specific disease combinations, severity, or progression patterns, which may differentially influence depressive symptoms. Third, depressive symptoms were assessed using a screening scale rather than clinical diagnostic interviews, potentially introducing measurement error. Fourth, as an observational study, causal inferences cannot be established despite the longitudinal design and cross-lagged modeling approach. Fifth, RI-CLPM assumes that within-person processes are relatively stable across observation intervals, which may not fully capture non-linear or abrupt changes in depression or cardiometabolic status. In addition, residual confounding from unmeasured factors, such as health-related behaviours, psychosocial stressors, and social support, cannot be fully excluded and may have influenced the observed associations. Finally, potential biological mediators, including inflammatory pathways, were not measured. Future studies incorporating clinical diagnoses, biomarker data, detailed treatment information, and intervention designs are warranted to further elucidate causal mechanisms and inform integrated prevention strategies.
Conclusions
This study provides longitudinal evidence of a dynamic and temporally asymmetric association between depressive symptoms and CMDs among middle-aged and older Chinese adults. The relative direction of influence also varied over time, with depressive symptoms showing stronger prospective associations with subsequent CMDs progression in the earlier period and CMDs becoming a stronger predictor of depressive symptoms in the later period. These findings underscore the dynamic interplay between mental and physical health and highlight the need for integrated, stage-specific, and risk-stratified interventions to promote healthy ageing.
Supplementary Information
Acknowledgements
We express our gratitude to the National School of Development of Peking University and China Social Science Survey Center of Peking University for providing the CHARLS data. We would also like to express our gratitude to all the individuals and researchers involved in the project.
Authors’ contributions
Xiquan Wang : Conceptualization, Data curation, Formal analysis, Methodology, Resources, Software, Visualization, Writing – original draft. Afei Qin : Conceptualization, Data curation, Project administration, Resources, Validation, Writing – original draft, Writing – review & editing.
Funding
Horizontal project 2025 of Nanjing University of traditional Chinese medicine: support project for market prediction and brand design of aging products based on the elderly with chronic diseases. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.
Data availability
The raw data supporting the conclusions of this article can be found here: https://charls.pku.edu.cn/.
Declarations
Ethics approval and consent to participate
This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by the biomedical ethics committee of Peking University (IRB approval number: IRB00001052-11015). All participants or their guardians provided written informed consent for participation in both the baseline and follow-up surveys. Consent was obtained prior to the administration of the study questionnaire.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Clinical trial number
Not applicable.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Xiquan Wang and Afei Qin contributed equally to this work.
References
- 1.National Bureau of Statistics of China. Statistical Bulletin of the National Economic and Social Development of the People’s Republic of China in 2024. Beijing: National Bureau of Statistics; 2025. Accessed 01 June 2025 https://www.stats.gov.cn/sj/zxfb/202502/t20250228_1958817.html.
- 2.Mechanick JI, Farkouh ME, Newman JD, Garvey WT. Cardiometabolic-Based chronic Disease, adiposity and dysglycemia drivers: JACC State-of-the-Art review. J Am Coll Cardiol. 2020;75(5):525–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Di Angelantonio E, Kaptoge S, Wormser D, Willeit P, Butterworth AS, Bansal N, O’Keeffe LM, Gao P, Wood AM, Burgess S, et al. Association of cardiometabolic Multimorbidity with mortality. JAMA. 2015;314(1):52–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Kivimäki M, Kuosma E, Ferrie JE, Luukkonen R, Nyberg ST, Alfredsson L, Batty GD, Brunner EJ, Fransson E, Goldberg M, et al. Overweight, obesity, and risk of cardiometabolic multimorbidity: pooled analysis of individual-level data for 120 813 adults from 16 cohort studies from the USA and Europe. Lancet Public Health. 2017;2(6):e277–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Barnett K, Mercer SW, Norbury M, Watt G, Wyke S, Guthrie B. Epidemiology of Multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet (London England). 2012;380(9836):37–43. [DOI] [PubMed] [Google Scholar]
- 6.Kuo SC, Lai SW, Hung HC, Muo CH, Hung SC, Liu LL, Chang CW, Hwu YJ, Chen SL, Sung FC. Association between comorbidities and dementia in diabetes mellitus patients: population-based retrospective cohort study. J Diabetes Complicat. 2015;29(8):1071–6. [DOI] [PubMed] [Google Scholar]
- 7.Jin Y, Liang J, Hong C, Liang R, Luo Y. Cardiometabolic multimorbidity, lifestyle behaviours, and cognitive function: a multicohort study. Lancet Healthy Longev. 2023;4(6):e265–73. [DOI] [PubMed] [Google Scholar]
- 8.He H, Tu R, Chen H, Wang C, Wu S, Wang S. Longitudinal trajectories of disability among Chinese adults: the role of cardiometabolic Multimorbidity. Aging Clin Exp Res. 2024;36(1):79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Einarson TR, Acs A, Ludwig C, Panton UH. Economic burden of cardiovascular disease in type 2 diabetes: A systematic review. Value Health: J Int Soc Pharmacoeconomics Outcomes Res. 2018;21(7):881–90. [DOI] [PubMed] [Google Scholar]
- 10.Herrman H, Kieling C, McGorry P, Horton R, Sargent J, Patel V. Reducing the global burden of depression: a Lancet-World psychiatric association commission. Lancet (London England). 2019;393(10189):e42–3. [DOI] [PubMed] [Google Scholar]
- 11.Rong J, Wang X, Cheng P, Li D, Zhao D. Global, regional and National burden of depressive disorders and attributable risk factors, from 1990 to 2021: results from the 2021 global burden of disease study. Br J Psychiatry 2025;227(4):688-97. [DOI] [PubMed]
- 12.Cui L, Ding D, Chen J, Wang M, He F, Yu S. Factors affecting the evolution of Chinese elderly depression: a cross-sectional study. BMC Geriatr. 2022;22(1):109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Friedrich MJ. Depression is the leading cause of disability around the world. JAMA. 2017;317(15):1517–1517. [DOI] [PubMed] [Google Scholar]
- 14.Walker ER, McGee RE, Druss BG. Mortality in mental disorders and global disease burden implications: a systematic review and meta-analysis. JAMA Psychiatry. 2015;72(4):334–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Meng L, Chen D, Yang Y, Zheng Y, Hui R. Depression increases the risk of hypertension incidence: a meta-analysis of prospective cohort studies. J Hypertens. 2012;30(5):842–51. [DOI] [PubMed] [Google Scholar]
- 16.Fang J, Zhang Z, Greenlund KJ. Association of depressive symptoms and hypertension prevalence, awareness, treatment and control among USA adults. J Hypertens. 2022;40(9):1658–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Maalouf E, Hallit S, Salameh P, Hosseini H. Depression, anxiety, insomnia, stress, and the way of coping emotions as risk factors for ischemic stroke and their influence on stroke severity: A case-control study in Lebanon. Front Psychiatry. 2023;14:1097873. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Robinson RG, Jorge RE. Post-stroke depression: a review. Am J Psychiatry. 2016;173(3):221–31. [DOI] [PubMed] [Google Scholar]
- 19.Nouwen A, Winkley K, Twisk J, Lloyd CE, Peyrot M, Ismail K, Pouwer F. Type 2 diabetes mellitus as a risk factor for the onset of depression: a systematic review and meta-analysis. Diabetologia. 2010;53(12):2480–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Huang ZT, Luo Y, Han L, Wang K, Yao SS, Su HX, Chen S, Cao GY, De Fries CM, Chen ZS, et al. Patterns of cardiometabolic Multimorbidity and the risk of depressive symptoms in a longitudinal cohort of middle-aged and older Chinese. J Affect Disord. 2022;301:1–7. [DOI] [PubMed] [Google Scholar]
- 21.Prigge R, Wild SH, Jackson CA. Depression, diabetes, comorbid depression and diabetes and risk of all-cause and cause-specific mortality: a prospective cohort study. Diabetologia. 2022;65(9):1450–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Chong KS, Chang YH, Yang CT, Chou CK, Ou HT, Kuo S. Longitudinal economic burden of incident complications among metabolic syndrome populations. Cardiovasc Diabetol. 2024;23(1):246. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Maschke V, Lohner V, Mons U. Linking cardiometabolic Multimorbidity to depressive symptoms in the oldest-old: results from a cross-sectional study in Germany. BMC Public Health. 2025;25(1):1720. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Zhang Y, Zhou Y, Kivimäki M, Holt-Lunstad J, Carrillo-Larco RM, Xu X. Cardiometabolic multimorbidity, social activity, and joint trajectories of physical disability, depressive symptom, and cognitive function in mid-to-late life: a multicohort study. BMC Med. 2025;23(1):577. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Wang M, Su W, Chen H, Li H. Depressive symptoms and risk of incident cardiometabolic Multimorbidity in community-dwelling older adults: the China health and retirement longitudinal study. J Affect Disord. 2023;335:75–82. [DOI] [PubMed] [Google Scholar]
- 26.Qiao Y, Ding Y, Li G, Lu Y, Li S, Ke C. Role of depression in the development of cardiometabolic multimorbidity: findings from the UK biobank study. J Affect Disord. 2022;319:260–6. [DOI] [PubMed] [Google Scholar]
- 27.Central People’s Government of the People’s Republic of China. 14th Five-Year Plan for Healthy Aging. Beijing: State Council of the People’s Republic of China; 2022. Accessed 11 Mar 2025. https://www.gov.cn/gongbao/content/2022/content_5692863.htm
- 28.Chen Y, Dai X. Research on risk protection of middle-aged and older adults in China. Beijing: Chinese Insurance Association; 2023.
- 29.Zhao Y, Strauss J, Yang G, Giles J, Hu P, Hu Y, Lei X, Park A, Smith JP, Wang Y. China health and retirement longitudinal study–2011–2012 National baseline users’ guide. Beijing: Natl School Dev Peking Univ. 2013;2:1–56. [Google Scholar]
- 30.Zhao Y, Hu Y, Smith JP, Strauss J, Yang G. Cohort profile: the China health and retirement longitudinal study (CHARLS). Int J Epidemiol. 2014;43(1):61–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Björgvinsson T, Kertz SJ, Bigda-Peyton JS, McCoy KL, Aderka IM. Psychometric properties of the CES-D-10 in a psychiatric sample. Assessment. 2013;20(4):429–36. [DOI] [PubMed] [Google Scholar]
- 32.Wu Y, Lei P, Ye R, Sunil TS, Zhou H. Prevalence and risk factors of depression in middle-aged and older adults in urban and rural areas in china: a cross-sectional study. Lancet. 2019;394:S53. [Google Scholar]
- 33.Chen H, Mui AC. Factorial validity of the center for epidemiologic studies depression scale short form in older population in China. Int Psychogeriatr. 2014;26(1):49–57. [DOI] [PubMed] [Google Scholar]
- 34.Andresen EM, Malmgren JA, Carter WB, Patrick DL. Screening for depression in well older adults: evaluation of. Prev Med. 1994;10:77–84. [PubMed] [Google Scholar]
- 35.Yan C, Jing Y, Ruan Q, Liu P, Chen G. Socioeconomic disparities in physiological dysregulation: a longitudinal mediation analysis of cardiometabolic Multimorbidity among middle-aged and elderly Chinese. BMC Public Health. 2025;25(1):3462. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Zhao L, Zhao C, Sun W, Zheng H, Gao Y, Wa CK, Wang Q, Liu Q, Wang Y, Wang Z. Long-term air pollution exposure and cardiovascular disease risk across cardiovascular-renal-metabolic stages: a nationwide study. BMC Public Health. 2025;25(1):2179. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Xu Z, Wu X, Xiao C, Zhang W, Yan P, Yang C, Zhang L, Cui H, Tang M, Wang Y, et al. Observational and genetic analyses of the bidirectional relationship between depression and hypertension. J Affect Disord. 2024;348:62–9. [DOI] [PubMed] [Google Scholar]
- 38.Moorehead NR, Goodie JL, Krantz DS. Prospective bidirectional relations between depression and metabolic health: 30-year follow-up from the National Heart, Lung, and blood Institute (NHLBI) coronary artery disease in young adults (CARDIA) study. Health Psychology: Official J Div Health Psychol Am Psychol Association. 2024;43(4):259–68. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Pan A, Keum N, Okereke OI, Sun Q, Kivimaki M, Rubin RR, Hu FB. Bidirectional association between depression and metabolic syndrome: a systematic review and meta-analysis of epidemiological studies. Diabetes Care. 2012;35(5):1171–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Chae WR, Kohring C, Rohde C, Köhler-Forsberg O, Otte C, Holstiege J. Eight-year nationwide study of the bidirectional association between type 2 diabetes and depression in nearly 8 million German outpatients. BMJ Open Diabetes Res Care. 2024;12(3):e003903. [DOI] [PMC free article] [PubMed]
- 41.Wang Z, Du Z, Lu R, Zhou Q, Jiang Y, Zhu H. Causal relationship between diabetes and depression: A bidirectional Mendelian randomization study. J Affect Disord. 2024;351:956–61. [DOI] [PubMed] [Google Scholar]
- 42.Maina JG, Balkhiyarova Z, Nouwen A, Pupko I, Ulrich A, Boissel M, Bonnefond A, Froguel P, Khamis A, Prokopenko I, et al. Bidirectional Mendelian randomization and multiphenotype GWAS show causality and shared pathophysiology between depression and type 2 diabetes. Diabetes Care. 2023;46(9):1707–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Solmi M, Radua J, Olivola M, Croce E, Soardo L, Salazar de Pablo G, Il Shin J, Kirkbride JB, Jones P, Kim JH, et al. Age at onset of mental disorders worldwide: large-scale meta-analysis of 192 epidemiological studies. Mol Psychiatry. 2022;27(1):281–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Bruce DG, Davis WA, Hunter ML, Peters KE, Davis TM, Starkstein SE. Lifetime depression history and depression risk in type 2 diabetes: A case-control study. J Diabetes Complicat. 2016;30(1):38–42. [DOI] [PubMed] [Google Scholar]
- 45.Mitchell AJ, Vaze A, Rao S. Clinical diagnosis of depression in primary care: a meta-analysis. Lancet (London England). 2009;374(9690):609–19. [DOI] [PubMed] [Google Scholar]
- 46.Fekadu A, Demissie M, Birhane R, Medhin G, Bitew T, Hailemariam M, Minaye A, Habtamu K, Milkias B, Petersen I, et al. Under detection of depression in primary care settings in low and middle-income countries: a systematic review and meta-analysis. Syst Rev. 2022;11(1):21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Kapfhammer HP. Somatic symptoms in depression. Dialog Clin Neurosci. 2006;8(2):227–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Li X, Zhang H, Han X, Guo L, Ceban F, Liao Y, Shi J, Wang W, Liu Y, Song W, et al. Predictive potential of somatic symptoms for the identification of subthreshold depression and major depressive disorder in primary care settings. Front Psychiatry. 2023;14:999047. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Ferenchick EK, Ramanuj P, Pincus HA. Depression in primary care: part 1-screening and diagnosis. BMJ (Clinical Res ed). 2019;365:l794. [DOI] [PubMed] [Google Scholar]
- 50.Peter RS, Jaensch A, Mons U, Schöttker B, Schmucker R, Koenig W, Brenner H, Rothenbacher D. Prognostic value of long-term trajectories of depression for incident diabetes mellitus in patients with stable coronary heart disease. Cardiovasc Diabetol. 2021;20(1):108. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The raw data supporting the conclusions of this article can be found here: https://charls.pku.edu.cn/.