2.1 Data source

The China Health and Retirement Longitudinal Study (CHARLS) is a tracking survey of the health and retirement status of a representative population aged 45 and older in mainland China. It has collected follow-up data from 2011 to 2020. It is the first nationally representative survey of the middle-aged and elderly population in China, serving as a high-quality public micro-database nationally. This study utilizes data from CHARLS collected between 2013 (Wave 2) and 2020 (Wave 5). During the survey, each respondent who agreed to participate signed an informed consent form.

2.2 Study cohort and inclusion criteria

This is a retrospective cohort study, with data collection beginning in 2013 (Wave 2). The research uses data from 2015 (Wave 3) as the baseline. It collects follow-up data from samples at the end of 2018 (Wave 4) or 2020 (Wave 5) to compare the impact of multimorbidity on the incidence of depression. This study selected samples aged ≥45 with a chronic disease count of ≥2, excluding samples with missing outcome indicators and those that failed ID matching over the years. To clarify the timing of depression onset after being diagnosed with multimorbidity, the study selected samples that were not diagnosed with multimorbidity or depression in 2013 but were diagnosed with multimorbidity and not diagnosed with depression in 2015 as the exposed group. The control group consisted of samples that were neither diagnosed with multimorbidity nor experiencing depression in both 2013 and 2015. Their data were observed during two subsequent follow-ups, specifically in 2018 (Wave 4) and 2020 (Wave 5). In order to exclude the impact of multimorbidity on depression in the control group, researchers required participants to be free of multimorbidity diagnoses in both 2018 and 2020. The data-cleaning process is illustrated in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Data cleaning flowchart.

This flowchart illustrates the steps taken to select and clean data for individuals from the CHARLS database across different years. Exposure Group (left): Participants diagnosed with multimorbidity and depression. Control Group (right): Participants neither diagnosed with multimorbidity nor depression. Matched IDs: Indicates successful matches across analysis waves. Remaining IDs: Final participant counts, indicating their diagnosis status.

https://doi.org/10.1371/journal.pone.0340673.g001

2.3 Follow-up cut-off time

The follow-up is cut off when a sample is diagnosed with depression, which may occur in either 2018 (Wave 4) or 2020 (Wave 5); for samples not diagnosed with depression, the follow-up continues until it stops in 2020 (Wave 5).

2.4 Relevant definitions

2.5 Ethics statement

The data used in this study comes from the China Health and Retirement Longitudinal Study (CHARLS). The CHARLS study received ethics approval from the Biomedical Ethics Committee of Peking University (Approval No: IRB00001052–11015). All participants provided informed consent prior to data collection. The informed consent process included a detailed explanation of the study’s purpose, data usage, and potential risks. All participants provided written consent. Additionally, the relevant ethics review board required ensuring that participants fully understood the study content and agreed to the use of their data in an anonymized manner. This study did not involve minors; therefore, parental or guardian consent was not required. Since this study utilized an existing publicly available dataset, the ethics committee did not require additional ethical review for this research.

3.1 Statistical description

Continuous measures were summarized as median (IQR) for all samples. Categorical measures were reported as frequency and percentage for crude and matched samples and only for the weighted samples.

3.2 Control for confounding factors

To minimize the effects of confounding factors, the analysis used doubly robust estimation combining overlap weighting and outcome regression to compare outcomes in the exposure and control groups. Overlap weighting is a propensity score method that attempts to mimic essential features of randomized clinical trials. Weights are assigned to each sample proportional to the probability of that sample belonging to the opposite treatment group. This results in including all available samples and the exact balance for the mean of all covariates included in the model. In simulations, Overlap weighting has also shown improved precision relative to other weighting options.

Initially, we included 13 potential variables based on prior literature (Gender [10], age [11], smoking status [12], drinking status [13,14], educational level [15], insurance type [16], marital status [17], household registration type (Hukou type) [18], social activity participation [19], as well as body mass index (BMI), sleep duration, Activities of Daily Living (ADL), and Instrumental Activities of Daily Living (IADL) [20]) identified five confounding factors using Directed Acyclic Graphs (DAGs) (S1 Fig). These five confounding factors (Age, BMI, Sleep Duration, Companion Status, and Hukou Type) were used for overlap weighting.

Next, the new weight values were incorporated into the Cox regression model along with the confounding factors from overlap weighting and other potential confounding variables (Gender, Insurance Type, Education Level, Smoking Status, Drinking Status, Social Activity Status, ADL and IADL) for confounding adjustment, to assess the impact of multimorbidity on the incidence of depression. Additionally, age (less than 60 years, 60 years or older), gender, and the number of chronic diseases (2, 3, ≥ 4) were set as different groups for subgroup analysis to explore the impact of multimorbidity status on the incidence of depression in different subgroups. Furthermore, the incidence rates of depression diagnosed after 3 and 5 years of multimorbidity were also calculated.

3.3 Missing variables and data handling

Data exhibited missing values, particularly for Hukou Type and Education Level, where missing data was approximately 20%, while other data had relatively fewer missing values (for details on missing data, see Table 1). Multiple imputation [21] methods were used to handle the missing data for indicators and other variables. Five imputed datasets were constructed based on the data in Table 1, and the most appropriate imputation method was selected based on the data type. For continuous data, the Predictive Mean Matching (PMM) method was employed for imputation, while a regression imputation model was used for categorical variables to ensure data integrity. The Rubin formula was applied to obtain model estimates and compare the corrected standard errors of the imputations [22]. The outlier was removed for existing outliers (BMI > 100), and the mode was used for imputation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of overlap weighting results.

https://doi.org/10.1371/journal.pone.0340673.t001

3.4 Sensitivity analysis

Propensity Score Matching (PSM) methods were utilized for sensitivity analysis, employing nearest neighbor matching with a caliper value set at 0.2 further to validate the effect size from the Cox regression. Additionally, to assess the robustness of the established association between multimorbidity and the incidence of depression against potential unmeasured confounding factors, the E-value was calculated using the method proposed by VanderWeele and Ding [23].

A significance level of 0.05 for two-sided comparisons was considered statistically significant. Hazard ratios (HRs) with 95% confidence intervals (CIs) are reported. Due to the nature of the study and the potential for type I errors resulting from multiple comparisons, all findings should be interpreted as exploratory. All analyses were conducted using RStudio version 4.3.3.

4.1 Covariate overlap weighting results

Based on the presence of multimorbidity, overlap weighting was applied to control for confounding factors in Fig 2 and other potential confounders in Fig 3 to ensure comparability between groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Results of Overlap Weighting for Confounding Factors.

This figure displays the absolute standardized mean differences for Confounding Factors before and after overlap weighting. Red dots represent the unweighted sample, while blue dots indicate the weighted sample.

https://doi.org/10.1371/journal.pone.0340673.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Results of Overlap Weighting for Potential Confounding Factors.

This figure displays the absolute standardized mean differences for Potential Confounding Factors before and after overlap weighting. Red dots represent the unweighted sample, while blue dots indicate the weighted sample. Note: The variable “Activity status” remained “intact” in both groups and was therefore excluded from the overlap weighting procedure.

https://doi.org/10.1371/journal.pone.0340673.g003

4.2 General results

This study included a total of 3,495 samples (1,560 females, accounting for 44.64%; median age: 62 years [IQR, 56–68 years]; median BMI: 23.70 [IQR, 21.64–25.87]; median sleep duration: 7 hours [IQR, 6–8]; 87.64% had a spouse/partner present; for Hukou type, 78.16% were urban, 21.08% were rural, and 0.76% were others).

After applying overlap weighting on the five variables—age, BMI, sleep duration, cohabitation with a spouse/partner, and Hukou type—the maximum balance between groups was achieved at baseline (see Table 1, Figs 2 and 3), with more significant changes in the Absolute Mean Differences values of the confounding factors between groups, in the sample after overlap weighting, 44.56% were female; median age: 62 years [IQR, 54–68 years]; median BMI: 24.01 [IQR, 21.42–26.14]; median sleep duration: 7 hours [IQR, 6–8]; 87.69% had a spouse/partner present; for Hukou type, 78.21% were urban, 21.09% were rural, and 0.70% were others.

During the follow-up period, the proportion of samples diagnosed with depression was 32.50%, with the exposed group at 42.65% and the control group at 30.88%. In the exposed group, the depression incidence rate three years after being diagnosed with multimorbidity (2018, Wave 4) was 26.09%, and five years later (2020, Wave 5) was 16.56%.

4.3 Multivariate cox analysis

A Cox proportional hazards regression was conducted on all variables after overlap weighting to explore the impact of multimorbidity on the incidence of depression. The results showed that the hazard ratio (HR) was 1.431, with a 95% confidence interval (CI) of 1.202–1.703 and P < 0.05. Individuals with multimorbidity had approximately a 43.10% increased risk of developing depression compared to those without multimorbidity.

4.4 Sensitivity analysis

Cox proportional hazards regression was performed on the original (unweighted) samples using propensity score matching to validate the findings. The results indicated an HR of 1.415, 95% CI: 1.128–1.775, P < 0.05, further supporting the results obtained from overlap weighting. Additionally, to assess the robustness of the identified relationship between multimorbidity and outcomes concerning potential unmeasured confounders, VanderWeele and Ding’s method was used to calculate the E-value, yielding an E-value of 1.88. To render the observed risk ratio (HR = 1.431) no longer significant, unmeasured confounders must be associated with depression and multimorbidity at least 1.88 times the risk. The robustness of the HR obtained from the Cox regression after overlap weighting is moderate.

4.5 Subgroup analysis

5.1 The significant correlation between multiple chronic diseases and depression risk in middle-aged and elderly populations

This study found that individuals with chronic diseases had a significantly increased risk of depression. The following points may explain how multiple chronic diseases contribute to an elevated risk of depression: ① Increased psychological stress and negative emotions. Patients with multiple chronic diseases often experience greater psychological stress and negative emotions [25]. The anxiety, stress, and other emotional challenges brought on by chronic illnesses frequently lead to depressive symptoms. For example, many chronic disease patients feel anxious and stressed due to the long-term management of their illnesses, which in turn affects their emotional state [26]. Furthermore, certain chronic diseases, such as stroke or Alzheimer’s disease, may cause changes in brain function, further exacerbating the risk of depression [27]. Compared to the general population, chronic disease patients typically have complex treatment management needs due to the necessity of long-term treatment, dealing with side effects, and coordinating with multiple healthcare providers, leading to increased psychological burdens and more frequent emotional fluctuations [28–30]. Research has also shown that elderly patients have decreased abilities to recognize and manage emotions, particularly those who struggle to express or identify their feelings, which results in a significantly increased risk of depression [29,30]. ② Long disease course and psychological burden of illness. Chronic disease patients often have a long disease duration accompanied by gradual physical decline, and the long-term health challenges bring continuous psychological burdens that increase the risk of depression. Patients fighting against chronic diseases often experience increased feelings of helplessness, which exacerbates their psychological burden over time. Especially when faced with multiple chronic conditions, the psychological burden becomes even heavier; prolonged illness stress leads to emotional exhaustion, ultimately increasing the risk of depression [31]. For instance, diabetes patients need to manage blood sugar levels and pay attention to diet, exercise, and other lifestyle factors, increasing the burden of self-care. As the disease progresses, these complex lifestyle adjustments and self-management needs may further exacerbate depressive feelings [32]. ③ The association between increased disability, mortality risk, and medical costs. The limitations in physical function, increased mortality risk, and medical cost burdens brought about by multiple chronic diseases create a vicious cycle for the psychological health of chronic disease patients. Studies have shown that with the increase in multiple chronic diseases, the disability rate among patients rises significantly, and the quality of life declines markedly, which typically accompanies an increased risk of depression. Relevant literature also indicates that depressive symptoms not only increase patients’ medical expenses but also lower overall health status, exacerbating their economic and emotional stress [30,33]. Moreover, a meta-analysis supports the positive correlation between multiple chronic diseases and higher mortality risk (HR: 1.44, 95% CI: 1.34–1.55), further backing the notion that multiple chronic diseases may aggravate depressive symptoms [33]. ④ Changes in lifestyle and self-identity. Long-term management of chronic diseases requires patients to make adaptive adjustments in their lifestyles, which can affect their sense of self-identity. For instance, prolonged treatment may lead patients to become distanced from their social circles, and a decline in quality of life may further alter their self-perception, intensifying depressive emotions. For patients with multiple chronic diseases, adverse changes in self-image can increase feelings of loneliness and helplessness, leading to further deterioration of mental health [29].

5.2 The relationship between the number of multiple chronic diseases and depression risk in middle-aged and elderly populations

This study found that as the number of multiple chronic diseases increases, the risk of depression rises significantly, especially among patients with 2 to 3 chronic conditions. This trend may be due to the more significant physiological burden and decline in quality of life brought on by multiple chronic diseases, making individuals more susceptible to depression. However, when the number of chronic diseases reaches 4 or more, the increase in depression risk is no longer significant, possibly due to: ① The formation of adaptive mechanisms: Patients may develop adaptive mechanisms or a phenomenon of “psychological numbness” in the long process of coping with multiple chronic diseases, gradually adapting to their multi-morbid status and alleviating the negative impact of emotional stress [34]. ② Enhanced social support: Patients with more multiple chronic diseases often receive more support from family and society, which can provide emotional comfort and practical help, aiding in the alleviation of depressive symptoms [35]. ③ More comprehensive medical support: Patients with 4 or more chronic diseases usually receive more intensive medical and mental health support, further helping to reduce the risk of depression [33]. ④ The non-significant finding in participants with 4 or more chronic conditions may be partly attributable to the small sample size in this subgroup, which limits statistical power.

Among middle-aged and elderly populations, older adults and women have a higher risk of depression. As age increases, older patients face declines in physical function and increasing caregiving demands while managing multiple chronic diseases, leading to a significant rise in depression risk [36]. Additionally, women have a higher risk of depression in the context of multiple chronic diseases, possibly because they are more likely to internalize stress and take on more caregiving responsibilities, making them more susceptible to social and psychological pressures when health deteriorates [37].

The results of this study suggest that mental health management for chronic disease patients should become a focus in public health and clinical practice, particularly for older adults and female patients. Routine chronic disease management should not only include monitoring physiological indicators but also emphasize early screening for depression and psychological support to mitigate the adverse effects of multiple chronic diseases on mental health. Furthermore, the significant association between the increasing number of chronic diseases and depression risk indicates that clinicians should pay special attention to the early recognition of depressive symptoms, especially in patients diagnosed with 2–3 co-morbid chronic diseases.