The CHARLS is a nationwide, representative longitudinal survey focusing on individuals aged 45 and older. The baseline survey was conducted between 2011 and 2012, with follow-up assessments in 2013, 2015, 2018, and 2020. The CHARLS employs computer-assisted personal interviewing technology to systematically collect data on respondents’ demographics, health status, lifestyle, socioeconomic conditions, and family support. During each follow-up, researchers perform physical measurements and health function tests, including blood pressure, lung capacity, grip strength, and balance ability. Additionally, blood samples are collected for biochemical analysis to assess important health indicators such as blood glucose and cholesterol levels (24). All participants provided written informed consent, and the study received ethical approval from the Biomedical Ethics Committee of Peking University (IRB 00001052-11015). This study strictly adheres to the STROBE guidelines for reporting observational epidemiological research, ensuring scientific rigor in study design and data collection (25).

For this research, the 2011 CHARLS data was utilized as baseline data, with follow-ups conducted in 2013, 2015, and 2018 to assess study outcomes. To ensure data integrity and research rigor, the following exclusion criteria were established: (1) participants younger than 45 years; (2) individuals with missing CMM data or already diagnosed with CMM during the 2011 baseline survey; (3) participants with missing data on HbA1c, HDL-C, or any covariates included at baseline; and (4) individuals with missing CMM data during follow-ups in 2013, 2015, or 2018. Ultimately, 7,256 participants who met the study criteria were included, with the specific inclusion and exclusion processes detailed in Figure 1.

Trained staff from the Chinese Center for Disease Control and Prevention collected three tubes of venous blood from each respondent, requiring overnight fasting; however, if fasting was not achieved, blood samples were still collected, and fasting status was recorded. The collected blood samples were preserved by freezing and transported to the Clinical Laboratory of Capital Medical University for centralized testing. Specific tests included TG, TC, LDL-C, and HDL-C, assessed using enzymatic colorimetric methods. HbA1c was measured using boronate affinity high-performance liquid chromatography; creatinine was assessed using the rate-blanked and compensated Jaffe method; uric acid (UA) was measured with the UA Plus method; and C-reactive protein (CRP) was evaluated using immunoturbidimetric assays. All results were merged and verified by two staff members from the Chinese Center for Disease Control and Prevention to ensure accuracy and consistency (26). Height was measured with a stadiometer, with participants standing upright, heels together, and weight evenly distributed. Weight was assessed using a calibrated scale with participants barefoot. BMI = weight (kg)/height (m)² (9). The HbA1c/HDL-C ratio was calculated using the following formula (12):

Depressive symptoms were assessed using the Chinese version of the 10-item Center for Epidemiological Studies Depression Scale (CES-D) (Supplementary Table S1). Each item was scored on a 4-point scale ranging from 0 to 3, with the total score ranging from 0 to 30. A total score of 10 or above was used to indicate the presence of depression, with higher scores reflecting greater severity of depressive mood (27).

The primary outcome of this study was the occurrence of CMM, defined as the coexistence of two or more CMDs, including stroke, heart disease, and diabetes (28). Heart disease and stroke diagnoses were based on standardized survey questions: “Have you ever been diagnosed by a doctor with heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems?” and “Have you ever been diagnosed by a doctor with stroke?” Diabetes was diagnosed if any of the following criteria were met: fasting blood glucose level ≥126 mg/dL (7 mmol/L), random blood glucose level ≥200 mg/dL (11.1 mmol/L), HbA1c level ≥6.5%, currently using diabetes medication, or self-reported diagnosis of “Have you ever been diagnosed by a doctor with diabetes or high blood sugar?” based on the standards published by the American Diabetes Association in 2024 (29). Although the current diagnostic criteria were published in 2024, the core thresholds have remained unchanged since 2010 (30). Thus, our classification of baseline data (2011–2012) is consistent with both historical and current standards. The event time was defined as the interval between the baseline assessment date and the date of the first reported occurrence of CMM. Once diagnosed with CMM, participants were no longer followed up. For individuals who did not report CMM during the follow-up period, the duration of follow-up was determined from the baseline assessment date to the final follow-up date.

We selected covariates based on previously published literature and the availability of variables in the CHARLS database (9, 10, 31, 32). Basic information was collected from participants via structured questionnaires, including demographic factors such as gender, age, location (categorized as urban and rural), marital status (categorized as single and married, with married defined as living with a spouse), and education level (categorized as illiteracy, primary school, middle school, and high school and above). Additionally, various health-related indicators were collected, including WC, BMI, smoking (categorized as non-smoker, ex-smoker, and smoker), drinking (defined as drinking more than once a month), systolic blood pressure (SBP), diastolic blood pressure (DBP), sleep duration (indicating actual nighttime sleep time), hypertension, kidney disease, antidiabetic, antidyslipidemic, LDL-C, TC, TG, creatinine, UA, and CRP. Hypertension was defined as SBP ≥ 140 mmHg or DBP ≥ 90 mmHg, self-reported diagnosis of hypertension, or currently using antihypertensive medication (33). The diagnosis of kidney disease primarily relied on self-reported history of prior kidney disease.

We first assessed the proportion of missing data for each selected variable and excluded those with more than 20% missing values. To evaluate multicollinearity, we then calculated the generalized variance inflation factor (GVIF) for the remaining variables. Variables with GVIF values exceeding 10 were removed, and the rest were retained for inclusion in the final analysis (34). This study first described baseline data according to CMM status, HbA1c/HDL-C ratio quartiles, and depression status. The normality of continuous variables was assessed using the Anderson-Darling test and Q-Q plots. Due to the non-normal distribution of the data, continuous variables were expressed as medians and interquartile ranges (IQR), with group comparisons conducted using the Mann–Whitney U test. Categorical variables were presented as counts and percentages, with comparisons between groups using the chi-squared test or Fisher’s exact test. Given the lack of clear clinical cutoff values for the HbA1c/HDL-C ratio, the median was utilized as the threshold based on prior research (12). To explore the joint effect of the HbA1c/HDL-C ratio and depression on the risk of CMM, participants were divided into four groups: (1) HbA1c/HDL-C < median without depression, (2) HbA1c/HDL-C ≥ median without depression, (3) HbA1c/HDL-C < median with depression, and (4) HbA1c/HDL-C ≥ median with depression. Kaplan–Meier survival analysis was employed to estimate the cumulative incidence of CMM for each group, with cumulative risk curves plotted and intergroup differences compared using the log-rank test.

Multivariable-adjusted Cox proportional hazards models were employed to examine the associations of baseline HbA1c/HDL-C ratio with CMM, both independently and jointly with depression. The HbA1c/HDL-C ratio was analyzed as quartiles and as a continuous variable. For all analyses, three models were specified: Model 1 without adjustment, Model 2 adjusted for demographic and lifestyle factors (gender, age, location, marital status, education, BMI, WC, smoking, drinking, SBP, DBP, and sleep duration), and Model 3 further adjusted for clinical variables (hypertension, kidney disease, antidiabetic, antidyslipidemic, LDL-C, TC, TG, creatinine, UA, and CRP).

Furthermore, to explore the potential nonlinear relationship between the HbA1c/HDL-C ratio and the risk of CMM, restricted cubic splines (RCS) were constructed based on Cox regression models. To compare the predictive abilities of the HbA1c/HDL-C ratio, depression, and their combination on CMM risk, receiver operating characteristic (ROC) curves were constructed at two, four, and seven-year time points to assess predictive performance.

To explore the effect of the individual HbA1c/HDL-C ratio and its joint effect with depression on the risk of CMM in different subgroups, we performed subgroup and interaction analyses. The calculation of interactions was performed using product terms in the main analysis. We further conducted sensitivity analyses to enhance the reliability and robustness of our findings. First, participants with a history of heart disease, stroke, or diabetes at baseline were excluded to reduce potential confounding from pre-existing conditions. Second, to address potential reverse causality, individuals who developed CMM during the 2013 and 2015 follow-up waves were excluded, ensuring a more accurate temporal relationship between exposure and outcome. Finally, to handle missing data in a methodologically sound manner, we employed multiple imputation using chained equations (MICE). This approach allowed for the preservation of statistical power and minimized bias associated with missing values.

We conducted counterfactual-based mediation analyses using the cmest function to examine the direct and indirect relationships among the HbA1c/HDL-C ratio, depression, and the risk of developing CMM. First, we assessed whether depression mediated the association between the HbA1c/HDL-C ratio and CMM using a Cox model for the outcome and a logistic model for the binary mediator, adjusting for all covariates. Second, we reversed the model to evaluate whether the HbA1c/HDL-C ratio mediated the relationship between depression and CMM, applying a linear model for the continuous mediator and a Cox model for the outcome, with the same covariate adjustments. Both analyses used imputation-based estimation with 1,000 bootstrap resamples to derive the total effect, total natural direct effect, and total natural indirect effect with 95% confidence intervals (CIs). A p-value < 0.05 for the TNIE was considered evidence of significant mediation. Hazard ratios (HRs) and 95% CIs were reported for all effects. All statistical analyses were performed using R software (version 4.3.2), with a two-sided significance level set at p < 0.05.

Taking into account the proportion of missing values and the results of the GVIF assessment, a total of 22 covariates were ultimately included in the final analysis (Supplementary Tables S2, S3). A total of 7,256 participants from the 2011 to 2018 waves of the CHARLS were included in this study. The mean age was 58 years, 52.83% were female, and 65.99% resided in rural areas. During the 7-year follow-up period, 419 incident cases of CMM were identified, corresponding to an incidence proportion of 5.77%. Compared with participants without CMM, those with CMM were older, more often female, had higher BMI and WC, were more likely to have depression and hypertension, and exhibited higher use of antidiabetic and antidyslipidemic medications as well as elevated lipid and inflammatory markers (Table 1). Stratification by HbA1c/HDL-C quartiles showed that higher ratios were associated with male sex, urban residence, higher BMI, smoking, hypertension, and increased use of metabolic medications (Supplementary Table S4). Participants with depression were more likely to be female, older, rural residents, single, and have lower educational attainment and HbA1c/HDL-C ratios (Supplementary Table S5).

Figure 2A displays Kaplan–Meier curves showing that CMM incidence increased with higher HbA1c/HDL-C ratio quartiles. To further evaluate the association between the HbA1c/HDL-C ratio and the risk of developing CMM, Cox regression analyses were conducted. In the unadjusted model, individuals in the highest quartile of the HbA1c/HDL-C ratio had a significantly elevated risk of CMM compared to those in the lowest quartile (HR = 3.51, 95% CI = 2.60–4.74). This association remained statistically significant after adjustment for all potential confounders (HR = 2.19, 95% CI = 1.14–3.32). The Schoenfeld residuals test indicated no violation of the proportional hazards assumption (p > 0.05) across all models, regardless of whether covariates were adjusted. Furthermore, when the HbA1c/HDL-C ratio was modeled as a continuous variable, each IQR increase in the ratio was associated with a 19% higher risk of CMM after full adjustment (HR = 1.19, 95% CI = 1.12–1.27) (Supplementary Table S6).

Figure 2B shows the Kaplan–Meier curves for cumulative CMM incidence by HbA1c/HDL-C ratio and depression. The highest risk was observed in participants with both a high ratio (above the median) and depression. Cox regression models were then used to evaluate the joint effects of HbA1c/HDL-C and depression on CMM risk. In the fully adjusted model, individuals with both a higher HbA1c/HDL-C ratio and depression had the highest risk (HR = 2.59, 95% CI = 1.82–3.68), followed by those with depression but a lower HbA1c/HDL-C ratio (HR = 2.05, 95% CI = 1.45–2.90), and then those with a higher ratio but without depression (HR = 1.57, 95% CI = 1.11–2.21) (Table 2). Across all models, the Schoenfeld residuals test indicated no violation of the proportional hazards assumption (p > 0.05).

Figure 3A displays the results of the unadjusted RCS model, in which four knots were selected based on the lowest Akaike Information Criterion (AIC) value. The specific knot locations—2.45, 3.51, 4.44, and 6.74—were determined automatically using the datadist function. Figures 3B,C show the RCS results after adjusting for covariates, where three knots were selected based on the same AIC criterion. The corresponding knot locations were 3.23, 3.94, and 4.88. The nonlinear relationship between the HbA1c/HDL-C ratio and CMM risk remained significant after adjustment. Based on the Cox proportional hazards model, the HbA1c/HDL-C ratio corresponding to an HR of 1 was identified as 3.94. Notably, we observed that the risk of developing CMM increased significantly when the HbA1c/HDL-C ratio exceeded 3.94, regardless of whether the model was adjusted for covariates (Figure 3).

After adjusting for all covariates, we evaluated the predictive performance of HbA1c/HDL-C, depression, and their combination for CMM using time-dependent ROC curves at 2-, 4-, and 7-year follow-ups (Figure 4). At 2 years, the AUCs were 0.781 for HbA1c/HDL-C, 0.795 for depression, and 0.796 for the combined model. At 4 years, the AUCs were 0.752, 0.760, and 0.765, respectively; at 7 years, they were 0.747, 0.753, and 0.758. These results indicate that the joint model consistently outperformed either predictor alone. DeLong’s test showed no significant difference between HbA1c/HDL-C and depression alone (p = 0.215), but the combined model significantly outperformed HbA1c/HDL-C alone (p = 0.012) and depression alone (p = 0.001). The joint model also exhibited superior sensitivity and specificity at optimal cut-offs and higher overall prognostic discrimination as measured by the concordance index (Supplementary Table S7).

Supplementary Figure S1 shows the forest plot of subgroup and interaction analyses for the association between HbA1c/HDL-C ratio and CMM risk. A significant interaction was observed only for age (P for interaction <0.05), with the association remaining positive across all age subgroups. In most other subgroups, the association was also generally positive. Figure 5 illustrates similar analyses for the joint effect of higher HbA1c/HDL-C ratio and depression. Again, age was the only significant effect modifier, while the combined effect remained significant across most subgroups, except among single participants and former smokers. Sensitivity analyses further supported the robustness of our findings. First, excluding participants with pre-existing heart disease, stroke, or diabetes at baseline yielded results consistent with the primary analysis (Supplementary Table S8). Second, excluding individuals who developed CMM during the 2013 or 2015 follow-ups also produced similar results (Supplementary Table S9). Finally, using MICE to handle missing data, rather than simple deletion, did not materially alter the findings (Supplementary Table S10).

Supplementary Figure S2 illustrates the results of the mediation analyses exploring the potential mediating roles of depression and the HbA1c/HDL-C ratio in relation to CMM. In Supplementary Figure S2A, when depression was evaluated as a mediator in the association between the HbA1c/HDL-C ratio and CMM, the indirect effect was not statistically significant (HR = 0.997, 95% CI: 0.984–1.012, p = 0.776), despite significant total and direct effects. This suggests that depression does not mediate the relationship between the HbA1c/HDL-C ratio and CMM. Similarly, in Supplementary Figure S2B, when the HbA1c/HDL-C ratio was tested as a mediator between depression and CMM, the indirect effect also failed to reach statistical significance (HR = 0.998, 95% CI: 0.992–1.004, p = 0.422). Taken together, these findings indicate that while both depression and the HbA1c/HDL-C ratio are independently associated with increased CMM risk, neither acts as a significant mediator in the pathway of the other.