Association of the C-reactive protein-triglyceride-glucose (CTI) index and its central obesity-modified derivatives with incident cardiometabolic multimorbidity: results from two prospective cohorts in China and England Authors: Chuhui Yue, Xinyu Wang, Yaona Li, Sujun Yuan, Zhu Wang, Hongli Liu, Huiyu Yang Published: 2026 ABSTRACT AbstractBackgroundCentral obesity exacerbates cardiometabolic multimorbidity (CMM) susceptibility through insulin resistance and inflammation. We investigated the longitudinal association of C-reactive protein-triglyceride-glucose (CTI)-related indices—which integrate these mechanisms with abdominal obesity measurements—with incident CMM in middle-aged and older adults.MethodsThis study included 6,498 participants from the China Health and Retirement Longitudinal Study and 2,455 from the English Longitudinal Study of Ageing, all free of diabetes, heart disease, and stroke at baseline. Incident CMM was defined as the occurrence of at least two of these diseases during follow-up. CTI-waist circumference (CTI-WC) and CTI-waist-to-height ratio (CTI-WHtR) were prespecified as principal exposures, with CTI, triglyceride-glucose index (TyG), and corresponding obesity-related derivatives as comparators. Death before CMM was treated as a competing event. Associations and prediction were assessed using Fine-Gray models, cumulative incidence functions, restricted cubic splines, time-dependent area under the curve, C-statistic, net reclassification improvement, integrated discrimination improvement, and calibration metrics.ResultsDuring median follow-up periods of 7.00 and 9.58 years, 303 and 255 participants developed incident CMM in the Chinese and English cohorts, respectively. In the Chinese cohort, each 1-standard deviation increase in CTI-WC and CTI-WHtR was associated with higher CMM risk (CTI-WC: subdistribution HR = 1.59, 95% CI 1.44–1.75; CTI-WHtR: subdistribution HR = 1.59, 95% CI 1.44–1.76), and the highest quartile had approximately fourfold higher risk than the lowest quartile. In the English cohort, associations were weaker but directionally consistent, with CTI-WC remaining significant per 1-standard deviation increment (1.18, 1.03–1.35). In the Chinese cohort, CTI-WC and CTI-WHtR yielded the highest 7-year area under the curve values (0.693 and 0.692), and CTI-WC showed the largest reclassification improvement. Calibration and sensitivity analyses supported model stability.ConclusionsPrespecified CTI-WC and CTI-WHtR were associated with incident CMM after accounting for competing mortality. Integrating inflammatory-metabolic burden with central adiposity may improve CMM risk stratification, especially in the Chinese cohort.Graphic abstract Supplementary InformationThe online version contains supplementary material available at 10.1186/s12933-026-03238-5.Keywords: Cardiometabolic multimorbidity, CTI-related indices, Abdominal obesity, CHARLS, ELSA What is currently known about this topic? CMM is rising in ageing populations. CMM is rising in ageing populations. Insulin resistance and inflammation contribute to CMM. Insulin resistance and inflammation contribute to CMM. Central obesity improves cardiometabolic risk stratification. Central obesity improves cardiometabolic risk stratification. What is the key research question? Do CTI-WC and CTI-WHtR predict incident CMM beyond TyG-related indices? What is new? CTI-WC and CTI-WHtR predicted CMM after accounting for competing mortality. CTI-WC and CTI-WHtR predicted CMM after accounting for competing mortality. CTI-WC showed the strongest reclassification gain in CHARLS. CTI-WC showed the strongest reclassification gain in CHARLS. External validation in ELSA showed weaker but consistent associations. External validation in ELSA showed weaker but consistent associations. How might this study influence clinical practice? CTI-WC may help identify older adults needing closer CMM prevention. Background Cardiometabolic multimorbidity (CMM) refers to the presence of two or more cardiometabolic diseases, such as diabetes, heart disease, and stroke, within the same individual [1]. Recent large-scale cohort studies have indicated a major global increase in the prevalence of CMM, currently affecting approximately 3–6% of the adult population worldwide [2, 3]. In China, evidence from a nationwide cohort including more than one million persons showed that the prevalence of CMM increased from 2.4 to 5.9% between 2010 and 2016 [4], indicating a rapid escalation of cardiometabolic disease clustering. Large population-based studies further indicate that CMM is associated with markedly higher mortality compared with single cardiometabolic diseases, with the coexistence of two conditions conferring approximately a twofold increase in mortality risk and three conditions a nearly fourfold increase [1]. Beyond mortality, CMM has been repeatedly associated with functional decline, disability, and poor quality of life [3, 5, 6], consequently placing a large burden on individuals and healthcare systems [7, 8]. Because CMM develops through the stepwise accumulation of cardiometabolic diseases over time, identifying individuals at high risk before overt disease clustering—or even before the first clinically recognized cardiometabolic disease—remains an important prevention gap [9]. Taken together, these data indicate the urgent need for early risk classification and appropriate management to reduce the growing health burden associated with CMM. Insulin resistance is widely regarded as a shared upstream mechanism linking multiple cardiometabolic disorders [10], such as type 2 diabetes, dyslipidemia, and obesity, and may exert a crucial impact on the onset and progression of CMM. While the hyperinsulinemic euglycemic clamp test is regarded as the reference standard to evaluate IR [11], its limited feasibility in large population-based cohort studies has spurred the adoption of proxy markers such as the triglyceride-glucose (TyG) index and its obesity-associated variants [12]. However, CMM is increasingly viewed as a clinical phenotype reflecting long-term systemic imbalance, involving multiple interrelated pathological processes beyond metabolic dysregulation, including chronic low-grade inflammation, abnormal adiposity distribution, and vascular dysfunction [13–15]. Therefore, reliance on the TyG index alone may be insufficient to capture inflammatory burden, underscoring the rationale for integrating metabolic and inflammatory information into composite risk indicators. Against this background, Ruan et al. proposed the C-reactive protein-triglyceride-glucose index (CTI), a metric that combines C-reactive protein (CRP) with the TyG index to assess both systemic inflammatory status and the degree of IR, thus offering a more comprehensive characterization of metabolic-inflammatory dysregulation [16]. Later investigations have demonstrated that CTI is associated with several adverse cardiometabolic outcomes, including nonalcoholic fatty liver disease, stroke, and coronary heart disease [17–19]. Beyond cardiovascular outcomes, CTI has also been applied in metabolic risk stratification and was shown to identify incident diabetes in the general middle-aged and elderly population, with improved prognostic precision compared with the TyG index or CRP alone [20]. A recent large-scale multi-state study further suggested that CTI is associated with the incidence and progression trajectory of CMM, supporting its relevance to cardiometabolic risk stratification [21]. However, whether incorporating abdominal obesity measurements can further refine CTI-based CMM risk assessment remains unclear. Central obesity, especially visceral adiposity, is closely linked to insulin resistance and chronic low-grade inflammatory responses, both of which are important processes in cardiometabolic disease development [22]. Recent cohort evidence further indicates that abdominal adiposity-related phenotypes, such as trajectories of body roundness index, are significantly linked to the risk of incident CMM, highlighting the potential role of long-term central fat accumulation in multimorbidity development [23]. Beyond individual adiposity phenotypes, recent studies have also combined insulin resistance-related indicators with obesity-related anthropometric measures to improve CMM risk assessment, including work on insulin resistance-related indices and a composite obesity-TyG index [24, 25]. This rationale is supported by evidence that visceral adiposity and ectopic fat accumulation are closely related to adipokine dysregulation, cardiometabolic dysfunction, and vascular endothelial abnormalities [26]. Building on this evidence, several research teams have examined the combination of IR or inflammation-related indices with assessments of abdominal adiposity to more comprehensively reflect cardiometabolic risk-related characteristics. Following the established practice of TyG-derived anthropometric indices [27], CTI-WHtR and CTI-WC were constructed as multiplicative composite indicators by multiplying CTI by waist-to-height ratio or waist circumference, respectively, rather than by simply adding components. CTI-WHtR can be interpreted as a dimensionless relative-risk marker because waist-to-height ratio is dimensionless, whereas CTI-WC retains the scale of waist circumference multiplied by a dimensionless composite biomarker. Clinically and biologically, higher standardized or quartile-based values of these indices indicate a greater combined burden of metabolic-inflammatory dysregulation and central adiposity, rather than a direct physical measurement. However, evidence remains limited regarding whether CTI-WC and CTI-WHtR provide additional risk-stratification information beyond unmodified CTI and TyG-related indices within a competing-risk framework for incident CMM assessment. Therefore, leveraging population-based longitudinal data from the two nationally representative aging cohorts—the China Health and Retirement Longitudinal Study (CHARLS) and the English Longitudinal Study of Ageing (ELSA)—the present study evaluates prespecified CTI-WC and CTI-WHtR indices alongside CTI and TyG-related comparators in relation to incident CMM, with the aim of informing risk identification and stratification across different populations. Methods Study design and populations We utilized datasets from two nationally representative longitudinal aging cohorts: the CHARLS and the ELSA. To ensure reliable cross-national comparisons, the study designs and measurement tools of both cohorts were standardized, making cross-country evaluations feasible. Furthermore, previous studies have confirmed the comparability of these cohorts regarding measurement and analytical consistency [28]. The CHARLS baseline survey (2011) employed a multistage probability sampling design to enroll Chinese residents aged ≥ 45 years across 28 provinces [29]. Data on socioeconomic status, health history, and lifestyle were collected via standardized interviews, clinical examinations, and biomarker tests. From the initial 17,708 participants, we sequentially excluded those: (1) aged < 45 or > 110 years, or missing sex data (n = 429); (2) lacking baseline biomarkers or anthropometrics (n = 8,137); (3) with prevalent heart disease, stroke, or diabetes at baseline, or missing diagnosis (n = 2,008); and (4) with invalid follow-up duration or missing status (n = 636). Ultimately, 6,498 participants were included in the final CHARLS cohort (Fig. 1A). Flowchart of participant selection for the analytic cohorts in A CHARLS and B ELSA The ELSA cohort recruited community-dwelling individuals aged ≥ 50 years in England using postcode-based random sampling [30]. For this analysis, wave 2 (2004–2005) served as the baseline, with outcomes tracked until wave 7. Applying consistent criteria to the initial 9,432 participants, we excluded: (1) age < 45 or > 110 years, or missing sex (n = 89); (2) missing baseline biomarkers or anthropometrics (n = 6,048); (3) prevalent heart disease, stroke, or diabetes at baseline, or missing diagnosis (n = 583); and (4) invalid follow-up data (n = 257), yielding a final ELSA cohort of 2,455 (Fig. 1B). Exposures CTI-WHtR and CTI-WC were prespecified as the principal exposures because they combine inflammatory-metabolic burden with central adiposity. CTI and TyG, together with their corresponding obesity-related derivatives, including TyG-WHtR and TyG-WC, were evaluated as comparator indices. The formulas for CTI, TyG, and their obesity-related derivatives are provided below. Outcome The primary outcome was incident CMM, defined as the coexistence of two or more cardiometabolic diseases during follow-up, including diabetes, heart disease, and stroke [3, 31]. The diagnosis of CMM was established when a second cardiometabolic disease was identified [32]. Heart disease and stroke were mainly ascertained through self-reported physician diagnoses, whereas diabetes was defined according to self-reported physician diagnosis, medication use, fasting plasma glucose (FPG), or HbA1c criteria. Because all participants were free of diabetes, heart disease, and stroke at baseline, incident CMM was identified when a participant first met the definition of having two or more of these conditions during follow-up. The onset of CMM was assigned to the first follow-up wave in which the CMM definition was met. In the competing-risk framework, death before incident CMM was treated as a competing event. Covariates Covariates were selected according to prior epidemiological evidence, clinical relevance, and availability across both cohorts. The main multivariable models included age, sex, education level, marital status, smoking status, alcohol consumption, and hypertension. Education level and marital status were harmonized across CHARLS and ELSA to improve comparability. Smoking and alcohol consumption were included as lifestyle-related covariates, and hypertension was included as a major cardiometabolic risk factor. Baseline low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), estimated glomerular filtration rate (eGFR), and body mass index (BMI) were not included in the primary fully adjusted models. LDL-C and HDL-C were evaluated in additional sensitivity analyses, whereas BMI was not adjusted for in the primary models to avoid overadjustment because the main exposures already incorporated central adiposity. Statistical analysis All statistical analyses were performed using R software. Continuous variables were summarized as means with standard deviations or medians with interquartile ranges according to their distributions, whereas categorical variables were presented as numbers and percentages. Baseline characteristics were compared between participants with and without incident CMM separately in the CHARLS and ELSA cohorts. Student’s t test was used for normally or approximately normally distributed continuous variables, and the Wilcoxon rank-sum test was used for skewed continuous variables. Categorical variables were compared using chi-square or Fisher’s exact tests, as appropriate. After applying all exclusion criteria, missing baseline covariates were imputed separately in the CHARLS and ELSA cohorts using random forest imputation. Exposure variables, outcome status, and follow-up time were not imputed. The main CTI-related exposures were specified a priori based on biological plausibility and clinical interpretability. CTI-WHtR and CTI-WC were considered the principal exposures because they integrate inflammatory-metabolic burden with central adiposity, while CTI, TyG, TyG-WHtR, and TyG-WC were evaluated as comparator indices. Spearman rank correlation analyses were conducted separately in CHARLS and ELSA to describe the correlation structure among CTI, TyG, their obesity-related derivatives, and conventional anthropometric and metabolic measures. Cumulative incidence functions were plotted to estimate the incidence of CMM while accounting for death before CMM as a competing event. The associations of CTI-WHtR, CTI-WC, and comparator indices with incident CMM were examined using Fine-Gray proportional subdistribution hazards models, and subdistribution hazard ratios (sHRs) with 95% confidence intervals (CIs) were reported. Three hierarchical models were constructed: Model 1 was unadjusted; Model 2 was adjusted for age and sex; and Model 3 was further adjusted for education level, marital status, smoking status, alcohol consumption, and hypertension. Restricted cubic splines with four knots were fitted within the fully adjusted Fine-Gray models to assess potential nonlinear dose–response associations. Multicollinearity among covariates in the fully adjusted models was assessed using generalized variance inflation factors (GVIFs). For categorical variables with more than one degree of freedom, adjusted GVIF values were calculated as GVIF^(1/(2 × Df)). The proportional subdistribution hazards assumption was evaluated using tests for time-dependent effects and Schoenfeld residual-based diagnostic plots derived from the weighted Cox representation of the Fine-Gray models. Prediction performance was evaluated within the competing-risk framework. Time-dependent receiver operating characteristic curves were generated based on predicted cumulative incidence from the Fine-Gray models, and the area under the curve (AUC) was calculated for models incorporating CTI- and TyG-related indices. The incremental predictive value of CTI-related indices was further assessed using net reclassification improvement (NRI) and integrated discrimination improvement (IDI) based on Fine-Gray model-derived predicted risks. Calibration of the competing-risk prediction models was quantified using calibration intercepts, calibration slopes, and Brier scores at the corresponding prediction time points. Several complementary analyses were conducted to evaluate the robustness and interpretability of the main findings. Prespecified subgroup analyses were conducted in CHARLS to examine the association between CTI-WC and incident CMM across strata defined by age, sex, smoking status, drinking status, and hypertension. Exploratory pathway analyses were further conducted in CHARLS to assess whether systolic blood pressure (SBP) and estimated pulse wave velocity (ePWV) were statistically consistent with indirect pathways linking CTI-WC and CTI-WHtR to incident CMM. In both cohorts, cause-specific Cox models were fitted as sensitivity analyses to the primary Fine-Gray framework. Additional sensitivity analyses with further adjustment for LDL-C and HDL-C were also performed in both CHARLS and ELSA. In CHARLS, additive and multiplicative CTI-WC specifications were compared by fitting models that included CTI alone, WC alone, CTI and WC as separate terms, or CTI-WC as a product term. These specifications were compared in terms of association estimates, 7-year discrimination, and 7-year reclassification performance. Additional CHARLS sensitivity analyses included complete-case analysis, exclusion of participants with follow-up duration less than two years, redefinition of hypertension using the ≥ 130/80 mmHg threshold, and further adjustment for depressive symptoms and grip strength. A two-sided P value < 0.05 was considered statistically significant. Results Baseline characteristics and follow-up information The analytic cohorts consisted of participants free of diabetes, heart disease, and stroke at baseline. In CHARLS, 6,498 participants were included, with a median follow-up duration of 7.00 years; 303 participants developed incident CMM, and 83 died before CMM during follow-up. In ELSA, 2,455 participants were included, with a median follow-up duration of 9.58 years; 255 participants developed incident CMM, and 84 died before CMM during follow-up. Baseline characteristics according to incident CMM status are shown in Table 1. In CHARLS, participants who developed CMM were older and had higher BMI, waist circumference, blood pressure, FPG, glycated hemoglobin (HbA1c), triglycerides (TG), total cholesterol (TC), LDL-C, and a higher prevalence of hypertension and depressive symptoms, whereas HDL-C was lower. In ELSA, participants with incident CMM were older, had lower educational attainment, and showed higher HbA1c levels. Across both cohorts, CTI, TyG, and their obesity-related derivatives, including CTI-WHtR, TyG-WHtR, CTI-WC, and TyG-WC, were consistently higher among participants who developed incident CMM. Table 1Baseline characteristics of the study participants stratified by incident CMM statusVariablesCHARLS (n = 6,498)ELSA (n = 2,455)Non-CMM (n = 6,195)CMM (n = 303)P-valueNon-CMM (n = 2,200)CMM (n = 255)P-valueAge, mean (SD), years58.72 (9.24)60.23 (7.87)0.00163.07 (7.22)64.06 (7.01)0.034Female, n (%)3272 (52.8%)173 (57.1%)0.1621225 (55.7%)146 (57.3%)0.680Education, n (%)0.8950.005 Below high school3032 (48.9%)150 (49.5%)1414 (64.3%)187 (73.3%) High school or above3163 (51.1%)153 (50.5%)786 (35.7%)68 (26.7%)Marital status, n (%)0.8400.315 Married or partnered5882 (94.9%)289 (95.4%)1614 (73.4%)179 (70.2%) Other313 (5.1%)14 (4.6%)586 (26.6%)76 (29.8%)Current smoker, n (%)2440 (39.4%)119 (39.3%)1.000295 (13.4%)37 (14.5%)0.697Current drinker, n (%)2415 (39.0%)124 (40.9%)0.5381675 (76.1%)190 (74.5%)0.618BMI, mean (SD), kg/m223.08 (3.50)24.84 (3.78) < 0.00127.45 (4.46)27.91 (4.65)0.134WC, mean (SD), cm84.18 (9.77)89.25 (10.33) < 0.00193.93 (12.30)95.14 (13.18)0.162Grip strength, mean (SD), kg31.87 (10.49)30.90 (10.62)0.12332.91 (11.22)32.05 (11.60)0.259SBP, mean (SD), mmHg128.06 (21.04)135.92 (23.00) < 0.001133.52 (17.92)134.81 (17.12)0.258DBP, mean (SD), mmHg74.80 (12.21)78.10 (13.00) < 0.00176.80 (10.35)77.01 (10.11)0.754FPG, mean (SD), mg/dL100.86 (13.28)104.50 (14.64) < 0.00188.23 (9.64)89.00 (9.95)0.244HbA1c, mean (SD), %5.10 (0.39)5.23 (0.42) < 0.0015.41 (0.33)5.47 (0.35)0.007TG, median (Q1-Q3), mg/dL100.89 (72.57–145.14)109.74 (85.40–163.73) < 0.001124.00 (88.57–168.28)124.00 (97.43–186.00)0.004TC, mean (SD), mg/dL192.28 (37.51)198.93 (38.81)0.004236.93 (43.35)241.32 (46.44)0.152HDL-C, mean (SD), mg/dL52.39 (15.09)47.83 (14.33) < 0.00161.26 (14.76)60.04 (14.74)0.212LDL-C, mean (SD), mg/dL116.51 (34.09)124.45 (35.50) < 0.001148.24 (36.14)150.38 (36.61)0.377CRP, median (Q1-Q3), mg/L0.95 (0.52–1.99)1.42 (0.76–2.86) < 0.0011.80 (0.80–3.70)2.00 (1.00–4.65)0.033Hypertension, n (%)1636 (26.4%)122 (40.3%) < 0.001894 (40.6%)119 (46.7%)0.074Depressive symptoms, n (%)2153 (34.8%)151 (49.8%) < 0.001227 (10.3%)33 (12.9%)0.238CTI, mean (SD)8.61 (0.74)8.91 (0.71) < 0.0018.84 (0.73)9.02 (0.76) < 0.001TyG index, mean (SD)8.56 (0.56)8.74 (0.55) < 0.0018.60 (0.49)8.70 (0.52)0.002CTI-WHtR, mean (SD)4.61 (0.78)5.07 (0.82) < 0.0015.02 (0.88)5.17 (0.96)0.016TyG-WHtR, mean (SD)4.58 (0.70)4.96 (0.73) < 0.0014.87 (0.73)4.98 (0.81)0.037CTI-WC, mean (SD)726.88 (120.61)797.66 (126.26) < 0.001834.66 (152.15)862.27 (163.74)0.011TyG-WC, mean (SD)722.42 (108.00)781.14 (112.97) < 0.001809.86 (129.88)830.70 (141.67)0.026Abbreviations: CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; CMM, cardiometabolic multimorbidity; BMI, body mass index; WC, waist circumference; SBP, systolic blood pressure; DBP, diastolic blood pressure; FPG, fasting plasma glucose; HbA1c, glycated hemoglobin; TG, triglycerides; TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; CRP, C-reactive protein; CTI, C-reactive protein-triglyceride-glucose index; TyG, triglyceride-glucose index; WHtR, waist-to-height ratio Baseline characteristics of the study participants stratified by incident CMM status Abbreviations: CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; CMM, cardiometabolic multimorbidity; BMI, body mass index; WC, waist circumference; SBP, systolic blood pressure; DBP, diastolic blood pressure; FPG, fasting plasma glucose; HbA1c, glycated hemoglobin; TG, triglycerides; TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; CRP, C-reactive protein; CTI, C-reactive protein-triglyceride-glucose index; TyG, triglyceride-glucose index; WHtR, waist-to-height ratio Correlation structure of CTI- and TyG-related indices Spearman correlation analyses showed generally similar correlation patterns in the CHARLS and ELSA cohorts (Fig. S1 and Fig. S2). CTI and TyG were strongly correlated, reflecting their shared metabolic components. Obesity-related derivatives also showed strong pairwise correlations, particularly between CTI-WC and CTI-WHtR and between TyG-WC and TyG-WHtR. These findings were expected given their shared mathematical components and supported the use of parallel and comparative models rather than interpreting highly correlated indices as mutually independent predictors. Fine-Gray model diagnostics The proportional subdistribution hazards assumption was assessed for the Fine-Gray models of CTI-WHtR and CTI-WC in both cohorts. As shown in Table S1, all individual and global tests yielded P values greater than 0.05 in CHARLS and ELSA, providing no evidence of violation of the proportional subdistribution hazards assumption. Schoenfeld residual-based diagnostic plots for the primary CTI-WHtR models also showed no obvious time-dependent patterns in either cohort (Fig. S3 and Fig. S4). Multicollinearity among covariates included in the fully adjusted models was assessed using GVIF and adjusted GVIF values (Table S2). In CHARLS, the highest adjusted GVIF was observed for sex (1.6006), while in ELSA the highest adjusted GVIF was observed for age (1.0877). These low values suggested that covariate collinearity was unlikely to materially affect the stability of the fully adjusted models. Cumulative incidence and Fine-Gray associations Cumulative incidence function analyses showed increasing cumulative incidence of CMM across quartiles of CTI-WHtR and CTI-WC after accounting for death before CMM as a competing event (Fig. 2). In CHARLS, clear separation of the cumulative incidence curves was observed for both CTI-WHtR and CTI-WC, with the highest incidence consistently seen in participants in Q4 (Gray’s test: both P < 0.001). Similar but less pronounced patterns were observed in ELSA, where higher quartiles of CTI-WHtR and CTI-WC were also associated with higher cumulative incidence of CMM (Gray’s test: P = 0.011 and P = 0.009, respectively). Cumulative incidence functions for incident CMM accounting for death as a competing event In the fully adjusted Fine-Gray models (Table 2), CTI-WC and CTI-WHtR were strongly associated with incident CMM in CHARLS. Each 1-SD increase in CTI-WC and CTI-WHtR was associated with a 59% higher subdistribution hazard of CMM (CTI-WC: sHR = 1.59, 95% CI 1.44–1.75; CTI-WHtR: sHR = 1.59, 95% CI 1.44–1.76). Compared with Q1, participants in Q4 had approximately fourfold higher CMM risk for both CTI-WC and CTI-WHtR, with significant trends across quartiles. TyG-WC and TyG-WHtR showed comparable associations in CHARLS, whereas the basic CTI and TyG indices had relatively smaller effect estimates. Table 2Fine-Gray associations of CTI- and TyG-related indices with incident CMMExposureCHARLS (n = 6,498)ELSA (n = 2,455)per SD sHR (95% CI)Q4 vs. Q1 sHR (95% CI)P for trendPer 1-SD sHR (95% CI)Q4 vs. Q1 sHR (95% CI)P for trendPrimary CTI indicesCTI-WC1.59 (1.44–1.75)4.06 (2.72–6.07) < 0.0011.18 (1.03–1.35)1.40 (0.98–2.00)0.098CTI-WHtR1.59 (1.44–1.76)4.03 (2.71–5.97) < 0.0011.13 (0.99–1.29)1.24 (0.88–1.75)0.142TyG comparisonsTyG-WC1.56 (1.41–1.73)3.75 (2.54–5.54) < 0.0011.17 (1.02–1.34)1.19 (0.83–1.69)0.105TyG-WHtR1.56 (1.41–1.74)3.81 (2.57–5.65) < 0.0011.11 (0.97–1.27)1.16 (0.82–1.64)0.109Basic indicesCTI1.42 (1.28–1.56)3.04 (2.07–4.47) < 0.0011.22 (1.06–1.39) 1.53 (1.05–2.22)0.060TyG index1.30 (1.17–1.44)2.26 (1.57–3.26) < 0.0011.21 (1.06–1.37)1.48 (1.04–2.12)0.013All results are based on Model 3, which was adjusted for age, sex, educational level, marital status, smoking status, drinking status, and hypertension. Abbreviations: sHR, subdistribution hazard ratio; CI, confidence interval; SD, standard deviation; Q, quartile. Fine-Gray associations of CTI- and TyG-related indices with incident CMM All results are based on Model 3, which was adjusted for age, sex, educational level, marital status, smoking status, drinking status, and hypertension. Abbreviations: sHR, subdistribution hazard ratio; CI, confidence interval; SD, standard deviation; Q, quartile. In ELSA, the associations were generally attenuated but remained directionally consistent. CTI-WC was significantly associated with incident CMM when modeled per 1-SD increment (sHR = 1.18, 95% CI 1.03–1.35), whereas the corresponding association for CTI-WHtR was weaker (sHR = 1.13, 95% CI 0.99–1.29). Among comparator indices, TyG-WC, CTI, and TyG showed significant per-SD associations with incident CMM. Overall, these findings suggest that CTI-WC and CTI-WHtR were robustly associated with incident CMM in CHARLS, while the external ELSA cohort showed weaker but generally concordant association patterns. Dose–response analyses Restricted cubic spline analyses based on the fully adjusted Fine-Gray models further characterized the dose–response patterns between CTI-related indices and incident CMM. In CHARLS, CTI-WHtR and CTI-WC showed significant positive overall associations with incident CMM (both P overall < 0.001), with no statistically significant evidence of nonlinearity (P nonlinear = 0.093 and 0.081, respectively; Fig. 3A and B). The estimated subdistribution hazard increased progressively at higher levels of both indices, consistent with the per-SD and quartile-based associations observed in the Fine-Gray models. CTI also showed a significant overall association with incident CMM in CHARLS (P overall < 0.001), with evidence of nonlinearity (P nonlinear = 0.036; Fig. 3C). Restricted cubic spline analyses of CTI-related indices with incident CMM in CHARLS. All results are based on Model 3, which was adjusted for age, sex, educational level, marital status, smoking status, drinking status, and hypertension In ELSA, the spline analyses showed broadly consistent but attenuated patterns (Fig. S5). CTI-WHtR, CTI-WC, and CTI were all associated with incident CMM in the overall spline tests (P overall = 0.049, 0.011, and 0.001, respectively), whereas no statistically significant nonlinear associations were observed (P nonlinear = 0.121, 0.129, and 0.058, respectively). These findings supported the overall dose–response relationship between CTI-related indices and incident CMM, while indicating weaker effect sizes in the external ELSA cohort. Predictive performance and reclassification improvement Seven-year time-dependent ROC analyses were performed to compare the discrimination performance of CTI- and TyG-related prediction models based on Fine-Gray model-derived predicted cumulative incidence (Fig. 4). In CHARLS, all augmented models improved discrimination compared with the base model. The models incorporating CTI-WC and CTI-WHtR showed the highest AUC values, with 7-year AUCs of 0.693 (95% CI 0.663–0.723) and 0.692 (95% CI 0.662–0.722), respectively. TyG-WC and TyG-WHtR also improved prediction, although their AUCs were slightly lower. The basic CTI model outperformed the basic TyG model, with AUCs of 0.675 (95% CI 0.645–0.705) and 0.655 (95% CI 0.625–0.686), respectively. In ELSA, the discrimination improvement was more modest, with AUCs ranging from 0.590 to 0.609 across the augmented models, compared with 0.581 for the base model. Seven-year time-dependent ROC curves for competing-risk prediction models in CHARLS and ELSA. AUC values were calculated using Fine-Gray model-derived predicted cumulative incidence at 7 years. Asterisks indicate significant improvement compared with the base model (P < 0.05). Reclassification analyses showed a similar pattern (Table 3). In CHARLS, all CTI- and TyG-related indices significantly improved reclassification and discrimination compared with the base model. CTI-WC showed the largest improvement, with a continuous NRI of 0.422 (95% CI 0.299–0.538; P < 0.001) and an IDI of 0.013 (95% CI 0.009–0.017; P < 0.001). CTI-WHtR also showed significant improvement, with an NRI of 0.374 (95% CI 0.284–0.482; P < 0.001) and an IDI of 0.012 (95% CI 0.009–0.016; P < 0.001). Compared with the basic TyG index, the basic CTI index showed higher C-statistic and greater NRI and IDI values in CHARLS. Table 3C-statistic, NRI, and IDI of CTI- and TyG-related indices for incident CMMExposureC-statisticNRI (95% CI)NRI pIDI (95% CI)IDI pCHARLS cohortBase + CTI-WC0.6870.422 (0.299–0.538) < 0.0010.013 (0.009–0.017) < 0.001Base + CTI-WHtR0.6840.374 (0.284–0.482) < 0.0010.012 (0.009–0.016) < 0.001Base + TyG-WC0.6790.413 (0.296–0.523) < 0.0010.011 (0.008–0.016) < 0.001Base + TyG-WHtR0.6760.377 (0.264–0.492) < 0.0010.010 (0.007–0.014) < 0.001Base + CTI0.6610.356 (0.224–0.456) < 0.0010.007 (0.004–0.010) < 0.001Base + TyG0.6420.230 (0.114–0.342) < 0.0010.003 (0.002–0.005) < 0.001ELSA cohortBase + CTI-WC0.5960.124 (-0.047–0.284)0.1600.002 (-0.000–0.004)0.087Base + CTI-WHtR0.5920.126 (-0.054–0.291)0.1600.001 (-0.001–0.002)0.320Base + TyG-WC0.5940.093 (-0.059–0.267)0.2930.002 (-0.000–0.004)0.127Base + TyG-WHtR0.5900.091 (-0.060–0.267)0.2200.001 (-0.001–0.002)0.260Base + CTI0.6020.152 (-0.033–0.316)0.0930.003 (0.001–0.006)0.013Base + TyG0.6020.148 (-0.030–0.315)0.1430.003 (0.001–0.006)0.013NRI, net reclassification improvement; IDI, integrated discrimination improvement; CMM, cardiometabolic multimorbidity. The base model was adjusted for age, sex, education level, marital status, smoking status, drinking status, and hypertension. C-statistic, NRI and IDI were calculated by comparing each augmented model with the base model using Fine-Gray model-derived predicted risks C-statistic, NRI, and IDI of CTI- and TyG-related indices for incident CMM NRI, net reclassification improvement; IDI, integrated discrimination improvement; CMM, cardiometabolic multimorbidity. The base model was adjusted for age, sex, education level, marital status, smoking status, drinking status, and hypertension. C-statistic, NRI and IDI were calculated by comparing each augmented model with the base model using Fine-Gray model-derived predicted risks In ELSA, reclassification improvement was weaker and did not reach statistical significance for most obesity-related derivatives. CTI-WC and CTI-WHtR had positive but non-significant NRI values, whereas the basic CTI and TyG indices showed significant but small IDI improvements. Overall, these findings suggest that CTI-related indices, particularly CTI-WC and CTI-WHtR, provided clear incremental predictive value in CHARLS, while the external ELSA cohort showed more modest discrimination and reclassification gains. Calibration of the competing-risk prediction models is summarized in Table S3. Overall, the models showed acceptable calibration in both cohorts, with calibration intercepts close to zero and calibration slopes close to one. In CHARLS, the calibration slopes of models incorporating CTI-WC and CTI-WHtR were closer to the ideal value of one than those of the base model and the models incorporating basic CTI or TyG. The Brier scores were similar across models in CHARLS (0.048), indicating no apparent deterioration in overall prediction error after adding CTI-related indices. In ELSA, calibration intercepts were also close to zero, and Brier scores ranged from 0.063 to 0.064 across models, suggesting broadly stable calibration performance despite modest discrimination and reclassification gains. Subgroup, sensitivity, and model specification analyses In CHARLS, subgroup analyses showed that the positive association between CTI-WC and incident CMM was generally consistent across strata defined by age, sex, smoking status, and drinking status (Fig. 5). A significant interaction was observed for hypertension status (P for interaction = 0.010), with a stronger association among participants without hypertension (sHR = 1.80, 95% CI 1.58–2.06) than among those with hypertension (sHR = 1.32, 95% CI 1.15–1.51). No significant interactions were observed for age, sex, smoking status, or drinking status. Subgroup analyses of the association between CTI-WC and incident CMM in the CHARLS cohort Sensitivity analyses supported the robustness of the main findings. Cause-specific Cox models yielded results broadly consistent with the primary Fine-Gray analyses in both cohorts (Table S4). In analyses with additional adjustment for LDL-C and HDL-C, the associations of CTI-WC and CTI-WHtR with incident CMM remained significant in CHARLS, although the effect estimates were attenuated compared with the primary models (CTI-WC: sHR = 1.48, 95% CI 1.31–1.67; CTI-WHtR: sHR = 1.47, 95% CI 1.30–1.66; both P < 0.001; Table S5). In ELSA, the corresponding estimates were weaker after additional lipid adjustment, but the directions of association were generally consistent with the primary analyses. Additional CHARLS sensitivity analyses, including complete-case analysis, exclusion of participants with follow-up duration less than two years, redefinition of hypertension using the ≥ 130/80 mmHg threshold, and further adjustment for depressive symptoms and grip strength, also showed persistent positive associations for CTI-WC and CTI-WHtR (Tables S6, S7, S8, S9). We further compared additive and multiplicative CTI-WC specifications in CHARLS (Tables S10, S11, S12). When CTI and WC were entered as separate terms, both remained associated with incident CMM in the Fine-Gray model. The multiplicative CTI-WC specification showed a comparable association estimate to the additive CTI + WC model. Similarly, 7-year discrimination and reclassification performance were nearly identical between the additive CTI + WC model and the multiplicative CTI-WC model, with AUCs of 0.694 and 0.695, NRIs of 0.422 and 0.427, and IDIs of 0.013 and 0.013, respectively. These findings indicate that CTI-WC captured risk information comparable to the additive specification while retaining a parsimonious composite-index structure. Exploratory pathway analyses Exploratory pathway analyses were conducted in the CHARLS cohort to evaluate whether vascular traits were statistically consistent with indirect pathways linking CTI-derived central obesity indices to incident CMM. For CTI-WC, the indirect effect through SBP was 0.001 (95% CI 0.001–0.002; P = 0.002), accounting for 6.5% of the total association, while the indirect effect through ePWV was 0.002 (95% CI 0.001–0.003; P = 0.002), accounting for 8.6% (Fig. 6). Similar patterns were observed for CTI-WHtR (Fig. S6), with indirect effects of 0.002 through SBP and ePWV, corresponding to proportions of 8.0% and 9.1%, respectively. In all models, the direct effects remained statistically significant, indicating that SBP and ePWV accounted for only a modest proportion of the observed associations. Given the observational design and the baseline assessment of vascular traits, these pathway analyses should be interpreted as exploratory and hypothesis-generating rather than as evidence of definitive causal mediation. Exploratory pathway analyses of CTI-WC and incident CMM via SBP and ePWV in CHARLS Discussion In this prospective cohort study across Asian and European populations, the prespecified CTI-WC and CTI-WHtR indices were associated with incident CMM after accounting for death before CMM as a competing event. In CHARLS, these waist-based CTI derivatives showed graded cumulative incidence patterns, robust associations in Fine-Gray models, and measurable improvement in risk prediction. In ELSA, the associations were weaker but generally in the same direction, suggesting partial external validation in a European ageing cohort. These findings suggest that integrating inflammation, dysglycemia, and central adiposity may improve early identification of individuals at high risk of CMM. They also indicate that the predictive gain from CTI-derived central obesity indices may depend on population context, especially differences in body-fat distribution and cardiometabolic susceptibility. Recent longitudinal evidence has extended CTI research from single-time cardiometabolic risk assessment to cumulative exposure, dynamic trajectories, and CMM-related outcomes [33, 34]. In parallel, CTI-WHtR has recently been evaluated as a composite marker for incident CMM and major cardiovascular outcomes, supporting the clinical relevance of integrating inflammatory-metabolic burden with central adiposity [35, 36]. Building on this evidence, the present study focused on two waist-based CTI derivatives, CTI-WC and CTI-WHtR, among adults initially free of diabetes, heart disease, and stroke. We further assessed whether these indices provided risk-stratification information beyond unmodified CTI, TyG, and TyG-derived waist-based indicators by using Fine-Gray competing-risk models and comparing association strength, discrimination, reclassification, and calibration performance. Beyond association estimates, the value of CTI-WC and CTI-WHtR lies in whether they improve risk stratification over simpler or established indices. In this study, we compared these two waist-based CTI derivatives with CTI, TyG, and TyG-derived obesity indicators using discrimination, reclassification, and calibration metrics. In CHARLS, CTI-WC and CTI-WHtR provided additional predictive information beyond conventional risk factors, with CTI-WC showing the most favorable reclassification performance among the evaluated indices. In ELSA, the improvement was weaker and less consistent, suggesting that these indices should be interpreted as population-sensitive indicators for risk assessment rather than universally superior predictors. These findings should be viewed alongside recent work on insulin-resistance and adiposity-related scores. In Chinese middle-aged and older adults, indices such as TyG, TyG-BMI, TyG-WC, CVAI, METS-IR, AIP, and eGDR have been compared for incident CMM prediction, with heterogeneous performance across discrimination and reclassification metrics [37]. Evidence from hypertensive populations also suggests that TyG-WHtR and TyG-WC carry prognostic information for CMM risk stratification [38]. Unlike these insulin-resistance-centered indices, CTI-WC and CTI-WHtR additionally incorporate CRP, thereby reflecting an inflammatory-metabolic-central adiposity profile. This distinction is relevant because CMM arises from overlapping metabolic, inflammatory, adiposity-related, and vascular processes. Therefore, our findings should be interpreted as extending, rather than replacing, existing insulin-resistance and adiposity scores. From a pathophysiological perspective, CMM is not driven by a single dominant risk factor but reflects the long-term accumulation of multiple adverse states, including metabolic dysregulation, chronic low-grade inflammation, and central obesity. Previous evidence suggests that surplus visceral fat deposition or central obesity may elevate CMM susceptibility via a wide array of direct or indirect biological mechanisms [39]. The core biological basis underlying this association lies in adipose tissue dysfunction [39], which may induce IR, systemic low-grade inflammation, and lipotoxic responses, thereby accelerating metabolic disturbances and substantially increasing cardiovascular disease risk [40, 41]. Dysregulated energy metabolism and mitochondrial dysfunction in adipose tissue may promote inflammatory activation and disturbances in glucose homeostasis, thereby accelerating cardiometabolic disease progression [42, 43]. Against this background, indices capturing only a single metabolic dimension may be insufficient to fully characterize the complex pathophysiological processes involved in the development of CMM. Because all participants were free of diabetes, heart disease, and stroke at baseline, this framework is particularly relevant to identifying metabolic-inflammatory and adiposity-related risk before overt cardiometabolic disease clustering becomes clinically apparent. By integrating IR-related metabolic burden, systemic inflammatory status, and abdominal adiposity within a single construct, CTI-abdominal obesity-based indices may more closely reflect, at the phenotypic level, the multi-pathway accumulation characteristic of CMM. In the present study, CTI-WHtR and CTI-WC should be interpreted as two waist-based specifications of the same CTI modification strategy, rather than as indices with clearly distinct biological meanings. Their broadly similar performance in CHARLS also argues against claiming a decisive advantage of either derivative. Although the biological mechanisms linking CTI to adverse outcomes remain incompletely understood, accumulating evidence suggests a mutually reinforcing interplay between IR and inflammation, which may lead to endothelial dysfunction, accelerate atherosclerotic progression, and increase plaque vulnerability, thereby increasing the risk of cardiovascular disease and cerebrovascular incidents [44–47]. This link is driven by increased metabolic demand, which stimulates glycolytic flux and results in the excessive generation of reactive oxygen species (ROS) [48]. Oxidative stress may further contribute to endothelial impairment, chronic inflammation, and mitochondrial dysfunction, processes that are closely linked to insulin resistance and atherosclerosis [49–51]. Consistent with this concept, prior studies have proposed that combining obesity-related anthropometric measures with metabolic indices such as TyG can enhance risk discrimination and predictive performance [19, 24, 52–54]. Our use of CTI-WC and CTI-WHtR follows the same general construction logic as established TyG-derived anthropometric indices, such as TyG-BMI, TyG-WC, and TyG-WHtR. These multiplicative composite markers should be interpreted as relative risk indicators reflecting the combined burden of metabolic-inflammatory dysregulation and waist-based central adiposity, rather than as direct physical measurements or evidence that the anthropometric component is an independent causal mediator. Previous studies have linked CTI-related metabolic-inflammatory indices with cardiometabolic outcomes, supporting the biological plausibility of integrating inflammatory and metabolic information for risk stratification [19, 20]. To further explore potential downstream pathways, we performed exploratory analyses focusing on vascular traits rather than components mathematically embedded in the CTI construct. In these analyses, SBP and ePWV accounted for modest proportions of the observed association between CTI-related indices and incident CMM, with ePWV showing a relatively larger contribution. These findings are consistent with the possibility that chronic metabolic-inflammatory burden may be linked to vascular load, arterial stiffness, and multisystem cardiometabolic deterioration during the development of CMM [27]. Overall, these exploratory findings are consistent with the multifactorial nature of CMM and suggest that CTI-related indices may reflect a metabolic-inflammatory phenotype related to vascular risk [14, 55]. Given the observational design and baseline assessment of vascular traits, these pathway analyses should be interpreted as exploratory and hypothesis-generating rather than as evidence of definitive causal mediation. The stronger performance of CTI-WC and CTI-WHtR in CHARLS may reflect population differences in body-fat distribution and cardiometabolic susceptibility rather than a uniform advantage of these indices across all settings. This interpretation is biologically plausible, as cardiometabolic risk associated with general and abdominal adiposity differs across world regions, and ancestry-related differences in body-fat distribution have been reported [56, 57]. Accordingly, multiplying CTI by WC or WHtR may be particularly informative in populations where central adiposity contributes substantially to cardiometabolic risk. In contrast, the weaker improvement observed in ELSA suggests that the additional value of waist-based CTI derivatives may be context-dependent. Moreover, the broadly comparable performance of CTI-WHtR and CTI-WC in CHARLS suggests that minor differences between these two derivatives should not be overinterpreted. Rather, both indices appear to reflect the combined burden of metabolic-inflammatory dysregulation and abdominal obesity. In subgroup analyses, the association between CTI-WC and new-onset CMM remained broadly consistent across age, sex, smoking status, and drinking status. A statistically significant interaction was observed only for hypertension status, with a stronger association among participants without hypertension. Evidence from large prospective analyses suggests that blood pressure is an important metabolic mediator linking excess adiposity to cardiovascular outcomes, including stroke [58]. In this context, once hypertension has already become established, the relative contribution of CTI-WC may be attenuated by dominant vascular pathways. In the present study, the stronger association among participants without hypertension may suggest that CTI-WC is more informative before overt vascular risk factors become established. However, this subgroup result should be interpreted cautiously, because the direction of association was consistent across strata and the interaction finding requires confirmation in further studies. The robustness of the main findings was supported by multiple sensitivity analyses. Cause-specific Cox models yielded results broadly consistent with the primary Fine-Gray framework. Additional adjustment for LDL-C and HDL-C also produced generally similar findings, although lipid adjustment attenuated several estimates, as expected given the potential for over-adjustment when evaluating lipid-derived indices. Further sensitivity analyses in CHARLS, including complete-case analysis, exclusion of participants with short follow-up, application of an alternative hypertension definition, and additional adjustment for depressive symptoms and grip strength, yielded materially similar estimates. Because death before CMM was treated as a competing event in the primary analyses, the reported associations and prediction results were interpreted within a competing-risk framework, which is particularly important in ageing cohorts. Limitations Several limitations should also be noted. First, owing to the observational design, causality cannot be inferred. Second, some outcomes were partly ascertained from self-reported medical history, which may have introduced misclassification, although prior validation studies in CHARLS support the reliability of self-reported diabetes [59]. Third, despite adjustment for multiple covariates, residual confounding cannot be excluded. Finally, although we included two independent cohorts from Asian and European populations, both comprised middle-aged and older adults, which may limit generalizability to younger populations or other ethnic groups. Further external validation is warranted. In addition, CTI-WC and CTI-WHtR are mathematically constructed from CTI and waist-based anthropometric measures; therefore, the findings should be interpreted as markers for risk assessment rather than evidence of causal effects of individual components. Because CTI incorporates fasting glucose and triglycerides, it may partly capture subclinical metabolic dysfunction before formal cardiometabolic disease diagnosis; therefore, residual pathological overlap or lead-time bias cannot be fully excluded. The pathway analyses involving SBP and ePWV were exploratory, as the observational design and baseline assessment of vascular traits preclude firm conclusions regarding temporal sequence or causality. Conclusion In conclusion, CTI-WC and CTI-WHtR were associated with incident CMM after accounting for competing mortality, with the clearest predictive improvement observed in the Chinese cohort. These findings suggest that combining inflammatory-metabolic burden with waist-based central adiposity may improve early CMM risk stratification, particularly in populations where cardiometabolic risk is closely linked to central fat distribution. Future studies should clarify population-specific thresholds, evaluate repeated measurements, and determine whether incorporating CTI-derived indices into preventive strategies can reduce progression to CMM. Supplementary Information Additional file1 (DOCX 4436 KB) Acknowledgements We thank the original data creators, depositors, copyright holders, and the funders of the data collections for the use of data from the English Longitudinal Study of Ageing and the China Health and Retirement Longitudinal Study. De-identified data of these cohorts are publicly available. Abbreviations Cardiometabolic multimorbidity Triglyceride-glucose index C-reactive protein-triglyceride-glucose index China Health and Retirement Longitudinal Study English Longitudinal Study of Ageing CTI-waist circumference TyG-waist circumference CTI-waist-to-height ratio TyG-waist-to-height ratio Area under the curve Low-density lipoprotein cholesterol High-density lipoprotein cholesterol Estimated glomerular filtration rate Generalized variance inflation factors Body mass index Net reclassification improvement Integrated discrimination improvement Fasting plasma glucose Glycated hemoglobin Total cholesterol Triglycerides Standard deviation Estimated pulse wave velocity Systolic blood pressure Author contributions CHY conceived and designed the study. XYW and YNL performed the statistical analyses. SJY, ZW, HLL assisted with data curation and interpretation. HYY supervised the project. CHY drafted the manuscript, and all authors critically reviewed, revised, and approved the final version. Funding This study was funded by the Basic Research Program of Shanxi Province (Natural Science General Project) (No. 202303021221207). Data availability Original survey datasets from the ELSA and the CHARLS are freely available to all bonafide researchers. Access to the data can be obtained by visiting their respective websites (ELSA: https://ukdataservice.ac.uk; CHARLS: http://charls.pku.edu.cn). Declarations Ethics approval and consent to participate The ELSA and the CHARLS were approved by the London Multicentre Research Ethics Committee and the Peking University Institutional Review Board, respectively. All participants provided written informed consent. Competing interest The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Additional file1 (DOCX 4436 KB) Data Availability Statement Original survey datasets from the ELSA and the CHARLS are freely available to all bonafide researchers. Access to the data can be obtained by visiting their respective websites (ELSA: https://ukdataservice.ac.uk; CHARLS: http://charls.pku.edu.cn).