Multimorbidity Clusters and Depressive Symptom Networks in Old Age: Evidence From China Authors: Xin Guan, Li-gang Wang, Zhen-jing Liu, Yan-kai Zhang, Xin-mei Hao, Xin-yue Wang, Lei Zhang, Ping Sun Published: 2026 ABSTRACT Objective To identify patterns of chronic disease multimorbidity in older Chinese adults, examine variables associated with the link between multimorbidity and depression, and characterize depressive symptom network features. Methods We analyzed cross-sectional data from the China Health and Retirement Longitudinal Study including 2,117 adults aged ≥60 years. Fourteen physician-diagnosed chronic conditions were subjected to exploratory factor analysis to identify latent multimorbidity clusters. Logistic regression was used to examine correlates of depression. For each cluster, Mixed Graphical Models estimated partial correlations among depressive symptoms assessed by the 10-item Center for Epidemiologic Studies Depression Scale and selected correlates, with centrality and bridge indices computed to characterize network structure. Results Tetrachoric factor analysis identified two stable clusters—a metabolic–vascular cluster (hypertension, dyslipidemia, diabetes, stroke) and a visceral–neurodegenerative cluster (liver disease, kidney disease, arthritis, memory-related disorders). Ten variables were significantly associated with depression (area under the curve=0.734, 95% confidence interval 0.708–0.793), with cognitive impairment showing the strongest association. Network analyses indicated that effort-related and motivational symptoms were central across clusters, whereas cognitive impairment consistently exhibited the highest bridge strength linking multimorbidity-related factors and depressive symptoms. Conclusion Multimorbidity in older adults forms clinically meaningful clusters associated with distinct depressive symptom network characteristics. Cognitive impairment emerged as the most prominent bridge-related correlate. Findings are associative and hypothesisgenerating. Longitudinal and interventional studies are warranted. INTRODUCTION According to the Global Burden of Disease 2021 study, an estimated 332.41 million individuals worldwide are affected by depression, which constitutes the largest share of global disability-adjusted life years, accounting for 36.24% of the total burden [1]. Its burden is particularly pronounced among older eISSN 1976-3026 OPEN ACCESS adults, who are increasingly affected by multiple chronic health conditions [2]. In China, rapid population aging has coincided with an epidemiological transition toward chronic noncommunicable diseases [3]. Data from the China Health and Retirement Longitudinal Study (CHARLS) indicate that more than two-thirds of adults aged 60 years or older live with at least one chronic condition, and nearly half experience two or more [4]. The coexistence of depression and chronic illness represents a major clinical and public health challenge, as physical conditions are often accompanied by depressive symptoms, while depression may in turn complicate disease management and functional outcomes [5]. This bidirectional relationship has been associated with accelerated functional decline, increased health care utilization, and elevated mortality risk. Previous studies have consistently reported that a greater number of chronic diseases is associated with a higher risk of depression [6]. However, most existing research has relied on simple disease counts, implicitly assuming that all conditions contribute equally to mental health outcomes. Such an approach overlooks the substantial heterogeneity of multimorbidity. In reality, chronic diseases often co-occur in non-random patterns due to shared biological mechanisms, behavioral risk factors, and social determinants. For example, cardiovascular and metabolic disorders frequently cluster together [7], whereas neurodegenerative conditions are commonly accompanied by affective and cognitive symptoms [8]. Identifying these latent patterns of multimorbidity may offer more clinically meaningful insights than aggregate disease counts. Exploratory factor analysis (EFA) provides a data-driven approach to detecting underlying groupings of chronic conditions and has been increasingly applied to characterize multimorbidity structures in population-based studies [9]. Despite growing recognition of multimorbidity as a risk factor for depression, the mechanisms linking chronic illness to depressive symptoms remain incompletely understood. Proposed pathways include systemic inflammation, dysregulation of the hypothalamic–pituitary–adrenal axis, functional impairment, and social isolation [10-14]. These processes are likely to interact in complex ways, and their relative importance may differ across disease combinations and individual contexts. Network analysis has emerged as a promising framework for examining depression at the symptom level. In contrast to traditional latent variable models that treat depressive symptoms as interchangeable indicators of a single disorder, network approaches conceptualize depression as a system of mutually interacting symptoms [15]. Within such networks, certain symptoms may occupy more central positions, while others may act as bridges linking affective, cognitive, and somatic domains. Mixed Graphical Models (MGMs) allow the estimation of partial correlations among symptoms and the quantification of centrality and bridge strength, providing a nuanced view of symptom interrelations [16]. Applying network analysis to the context of multimorbidity offers an opportunity to explore how different disease patterns may be associated with distinct configurations of depressive symptoms. The 10-item Center for Epidemiologic Studies Depression Scale (CESD-10) is a widely used and validated instrument for assessing depressive symptoms in Chinese populations [17]. Analysis at the item level enables the identification of specific symptoms—such as sleep disturbance, hopelessness, or anhedonia— that may play differential roles across multimorbidity contexts. Integrating CESD-10 symptom networks with information on relevant demographic, clinical, and functional variables may help clarify shared and unique pathways linking chronic disease patterns to depression. To date, few studies have combined multimorbidity pattern identification with symptom-level network analysis in a large, nationally representative Chinese sample. Existing work has typically focused on single diseases, relied on simplistic disease counts, or examined depression only at the scale-score level. Consequently, several important questions remain unanswered— Do chronic diseases cluster into distinct patterns among older Chinese adults? Which variables are most strongly associated with depression within these patterns? And how do depressive symptom networks differ across multimorbidity clusters? Addressing these questions may contribute to a more precise understanding of depression in the context of complex chronic illness. Given the limited prior evidence, the present study was designed as a cross-sectional, exploratory investigation. Rather than testing prespecified causal hypotheses, we aimed to generate data-driven insights into multimorbidity structures and their associations with depressive symptoms. Specifically, the objectives of this study were to: 1) identify latent clusters of chronic diseases in older Chinese adults, 2) examine variables associated with depression in the context of multimorbidity, and 3) explore associations between cluster membership and depressive symptom network characteristics. To our knowledge, few population-based studies have simultaneously integrated multimorbidity pattern identification, regression-based selection of multimorbidity-related correlates, and symptom-level depression network analysis in older adults. Findings should therefore be interpreted as associative and hypothesis-generating rather than confirmatory. METHODS The overall analytical workflow of the study is summarized in Figure 1. Study design and data source This study was a cross-sectional secondary analysis of data from the 2018 wave of the CHARLS, a nationally representative survey of Chinese adults aged 60 years and older. CHARLS employed a multistage stratified probability sampling design covering 28 provinces, 150 counties, and 450 rural villages and urban communities across China. The 2018 survey collected comprehensive information on depressive symptoms, physiciandiagnosed chronic diseases, functional status, and sociodemographic characteristics using standardized questionnaires and protocols. Data collection was conducted by trained interviewers under centralized quality control procedures. Participants and missing data handling Depressive status was classified based on the CESD-10 total score (Section 2.3 for details). Participants with missing values on any analytic variable were excluded using listwise deletion. Missingness was below 10% for all variables. Depressive symptoms Depressive symptoms were assessed using the CESD-10, a brief and widely validated instrument for Chinese older adults [18]. The CESD-10 includes 10 items rated on a 4-point Likert scale from 0 (“rarely or none of the time”) to 3 (“most or all of the time”), reflecting depressive experiences during the past week. Items assess being bothered by things, difficulty concentrating, feeling depressed, perceiving that everything was an effort, feeling hopeful about the future (reverse coded), feeling fearful, restless sleep, feeling happy (reverse coded), feeling lonely, and feeling unable to get going. Total scores range from 0 to 30, with higher scores indicating greater depressive symptom severity. A cutoff score of ≥10 was used to define probable depression in accordance with prior validation studies in Chinese populations [17]. In addition to total scores, item-level responses were retained for symptom network analyses. Chronic diseases A total of 14 chronic conditions diagnosed by physicians were assessed, including hypertension, diabetes, and arthritis (Supplementary Table 1). All conditions were coded as binary variables (0=no, 1=yes) based on participants’ self-reported medical diagnoses. These variables were then subjected to EFA to identify latent multimorbidity patterns. Candidate bridging variables A range of potential bridging variables were considered based on prior literature [19-23], including physical disabilities, sensory impairments, sleep duration, physical activity levels, social participation, alcohol use, use of assistive devices, perceived social support, employment limitations, multimorbidity burden, and cognitive function. For analytic purposes, these variables were coded numerically for statistical modeling in the statistical models; the detailed correspondence between variable names and codes is provided in Table 1. Cognitive function was assessed using the Mini-Mental State Examination (MMSE). MMSE scores were interpreted using education-adjusted cut-offs validated in Chinese populations: ≤17 for individuals with no formal education, ≤20 for those with primary education (≤6 years), and ≤24 for those with middle school education or higher (>6 years) [24]. Participants scoring below the corresponding threshold were classified as having cognitive impairment. Statistical analysis Data preprocessing All statistical analyses were conducted using standardized preprocessing procedures. Ordinal variables were treated as ordered factors, continuous variables were standardized using z-scores, and binary variables were coded as 0/1. No data imputation was performed—participants with missing values in analytic variables were excluded via listwise deletion. Sampling weights were not applied because the primary objective of this study was etiologic and exploratory inference rather than estimation of population-level prevalence. Identification of disease clusters EFA based on tetrachoric correlations was conducted to identify latent clusters of chronic diseases. Sampling adequacy was evaluated using the Kaiser–Meyer–Olkin (KMO) statistic, and factorability was assessed with Bartlett’s test of sphericity. The optimal number of factors was determined through parallel analysis. Factors were extracted using weighted least squares estimation and rotated with varimax rotation. Diseases with factor loadings ≥0.30 were retained in the corresponding cluster. Only conditions with sufficient variability (prevalence ≥3%) were included in the analysis to ensure model stability. EFA was selected because the primary aim was to identify latent dimensions underlying patterns of multimorbidity rather than to classify individuals into mutually exclusive subgroups [9]. This variable-centered approach is well suited for modeling shared covariance structures among binary medical conditions using tetrachoric correlations. Selection of bridging variables To identify variables most strongly associated with depression and to ensure parsimonious network models, a two-stage regression-based screening procedure was adopted, consistent with approaches used in prior network studies. First, univariate logistic regression analyses were conducted for each of the 17 candidate variables. Variables with p<0.10 were retained for further evaluation. Second, a multivariable logistic regression model was constructed using backward selection while adjusting for demographic covariates. Adjusted odds ratios with 95% confidence intervals (CIs) were reported. Model diagnostics included assessment of multicollinearity using variance inflation factors (<5 considered acceptable), evaluation of calibration via the Hosmer–Lemeshow test, and assessment of discrimination using the area under the receiver operating characteristic curve. Variables identified as significant in the final model were subsequently retained bridging variables in the symptom-level network analyses. This pragmatic screening strategy was adopted to reduce dimensionality and enhance network stability, following precedents in recent multimorbidity–depression network research. Symptom network construction For each identified disease cluster, separate symptom-bridge variable networks were estimated using MGMs, which allow the simultaneous modeling of binary, ordinal, and continuous variables. Network estimation employed the Extended Bayesian Information Criterion with graphical lasso regularization procedure with a tuning parameter of γ=0.5. Nonparanormal transformations were applied to mitigate potential non-normality. Each network comprised nodes representing the CESD-10 symptoms and the selected bridging variables. Edges reflected regularized partial correlations between variables. Psychological symptoms (CESD-10 items) and physical or sociobehavioral variables were conceptually grouped to facilitate interpretation of bridge connections. Ethical approval The CHARLS was approved by the Biomedical Ethics Review Committee of Peking University (IRB00001052-11015). Written informed consent was obtained from all participants prior to data collection [25]. The present study is a secondary analysis of publicly avail-able, de-identified data. The study protocol was reviewed by the Ethics Committee of Qingdao Mental Health Center and was determined to be exempt from ethical review (Exemption approval number: QDWZXWZLL2026003). All procedures were conducted in accordance with relevant ethical guidelines and regulations. RESULTS Sociodemographic and clinical characteristics A total of 2,117 participants were included in the final analysis. The median age was 66 years (Q₁–Q₃: 63–71), and 65.19% were men. Based on CESD-10 criteria, 471 participants (22.25%) were classified as having depression. Compared with nondepressed participants, those with depression were more likely to be female, have lower educational attainment, reside in rural areas, and be widowed. They also reported significantly poorer self-rated health. In addition, depressed individuals showed higher prevalence of brain damage, visual and hearing impairments, poorer childhood health, cognitive impairment, and multimorbidity. Shorter sleep duration, lower levels of light physical activity, and reduced social participation were also more common among the depressed group. Moreover, they were more likely to report lack of available helpers and limitations in work ability. Overall, depressive symptoms were associated with a constellation of sociodemographic disadvantage, functional impairment, and greater chronic disease burden. Detailed characteristics are presented in Supplementary Table 2. Disease clustering EFA based on tetrachoric correlations identified two interpretable clusters of chronic diseases. Sampling adequacy was acceptable (KMO=0.66), and Bartlett’s test of sphericity confirmed sufficient inter-item correlations for factor extraction (p<0.001). Parallel analysis supported a two-factor solution (Supplementary Figure 1). In the rotated factor structure, hypertension, dyslipidemia, and diabetes (disease1–3), together with stroke (disease8), loaded predominantly on Factor 1, forming a metabolic–vascular cluster. Factor 2 was characterized by higher loadings for liver disease (disease6), kidney disease (disease9), stomach disease (disease10), arthritis (disease12), and memory-related disorders (disease13), representing a visceral–neurodegenerative cluster. The separation between the two latent dimensions was visually evident in the factor loading heatmap (Supplementary Figure 2). Sensitivity analyses using oblique rotation yielded a highly similar loading pattern, supporting the robustness of the identified structure. Overall, the EFA results suggest the presence of two coherent multimorbidity patterns among older Chinese adults. Selection of bridging variables Univariate logistic regression analyses identified several candidate variables associated with depression (Supplementary Table 3). After adjustment for demographic covariates, ten variables remained independently associated with depression in the final multivariable model (Supplementary Table 4). Model performance was evaluated using 1,000 bootstrap replications, demonstrating acceptable discrimination and calibration, with an area under the receiver operating characteristic curve of 0.734 (95% CI 0.708–0.793) (Supplementary Figure 3). After adjustment for demographic covariates, ten variables remained independently associated with depression, including cognitive impairment, poor childhood health, inability to work due to health problems, visual impairment, hearing impairment, low social participation, reduced physical activity, short sleep duration, lack of available helpers, and multimorbidity burden. Bootstrapped difference tests further indicated that the bridge expected influence of cognitive impairment was significantly greater than that of most other variables (p<0.05), supporting its prominent role as a key bridge-related correlate. Symptom network analysis Cluster 1: metabolic–vascular network The MGMs for Cluster 1 revealed a densely interconnected network linking CESD-10 depressive symptoms with multiple variables. Node strength centrality ranged from 0.19 to 1.37. Among depressive symptoms, CESD6 (“felt fearful”), CESD10 (“could not get going”), CESD4 (“everything was an effort”), and CESD1 (“bothered by things”) demonstrated the highest centrality values (strength >1.0), indicating prominent roles within the symptom structure. Among variables, intensive physical activity showed relatively strong connectivity with depressive symptoms. Bridge analysis identified cognitive impairment (variable16) and inability to work due to health problems (variable15) as the strongest bridge nodes, with standardized bridge expected influence values exceeding 2.0. Cognitive impairment also exhibited comparatively high betweenness and closeness indices, highlighting its function as a key connector between depressive symptoms and physical health–related variables. Network structure, centrality estimates, and bridge indices for Cluster 1 are illustrated in Figures 2A, 3A, and 4A. Cluster 2: visceral–neurodegenerative network The network estimated for Cluster 2 showed a somewhat more dispersed configuration compared with Cluster 1. Centrality strength values ranged from 0.13 to 1.37. The most central depressive symptoms in this cluster were CESD10 (“could not get going”), CESD3 (“felt depressed”), and CESD4 (“everything was an effort”), each with strength values above 1.0. Consistent with findings from Cluster 1, cognitive impairment again emerged as the most influential bridge node, with bridge expected influence exceeding 2.5. Inability to work due to health problems also functioned as a bridge node (bridge strength >2.0), although its betweenness centrality was relatively lower, suggesting a more localized connecting role. In contrast, cognitive impairment maintained high values across multiple centrality metrics, indicating a more global bridging function within this network. Network estimates for Cluster 2 are presented in Figures 2B, 3B, and 4B. Cross-cluster patterns and stability Across both clusters, cognitive impairment consistently ranked first in bridge strength, demonstrating notable crossnetwork consistency. The correlation stability (CS) coefficients for bridge strength exceeded 0.50 in both networks, indicating acceptable stability of the centrality estimates. DISCUSSION Using data from the CHARLS, this study identified biologically coherent multimorbidity clusters and examined how these chronic disease patterns were differentially associated with depressive symptom network structure in older adults. Rather than indicating causal relationships, the findings delineate structured associations between disease constellations and symptom-level depressive phenotypes. By integrating EFA, regression modeling, and network analysis, we provide a multidimensional perspective that extends beyond simple disease counts toward a systems-oriented characterization of multimorbidity-related depression. The present study extends previous research in three ways. First, rather than relying on simple disease counts, we employed EFA to empirically derive latent multimorbidity clusters, revealing two distinct patterns that reflect shared pathophysiological mechanisms. Second, moving beyond the conventional case–control comparison, we systematically screened a broad range of candidate bridging variables using a two-stage regression approach, identifying ten robust correlates of depression in the context of multimorbidity. Third, we advanced beyond scale-level analyses by constructing symptom-level network models for each cluster separately, which allowed us to compare centrality and bridge strength of depressive symptoms and their correlates across different multimorbidity patterns. This multistep analytic pipeline provides a more granular understanding of how specific symptom clusters and bridging factors are differentially associated with distinct chronic disease constellations. Two latent clusters were identified—a metabolic–vascular cluster and a visceral–neurodegenerative cluster. The metabolic– vascular pattern, comprising hypertension, dyslipidemia, diabetes, and stroke, is consistent with prior evidence linking cardiometabolic burden to late-life depressive symptoms [26]. In this cluster, motivational and energy-related symptoms— including psychomotor slowing and fatigue—occupied central network positions. This configuration is compatible with literature suggesting that vascular and metabolic dysregulation commonly co-occur with alterations in frontostriatal and limbic systems involved in behavioral activation and reward processing [27,28]. The visceral–neurodegenerative cluster, including liver disease, kidney disease, and memory-related disorders, exhibited a comparatively diffuse symptom architecture, reflecting the multisystemic nature of organ dysfunction and neurocognitive decline. Despite this heterogeneity, motivational deficits again emerged as structurally central, suggesting that diminished drive and goal-directed behavior may represent a common feature across multimorbidity-related depressive presentations. Across analytic approaches, cognitive impairment demonstrated a consistent bridging position between multimorbidity clusters and depressive symptom networks. This observation aligns with extensive evidence describing reciprocal associations among chronic disease burden, cognitive decline, and depression [29,30]. Biologically, metabolic and visceral conditions are frequently accompanied by microvascular alterations, inflammatory processes, and reduced neuroplasticity, which are associated with disruptions in frontoparietal and limbic circuits [31]. Psychosocially, cognitive decline is often accompanied by reduced autonomy, diminished social engagement, and increased perceived burden [32,33]. Within the present network framework, cognition occupied a structurally strategic position, indicating a stable associative link between somatic disease patterns and depressive symptom connectivity. Functional and behavioral variables—including work disability, physical inactivity, and sensory impairment—were also associated with depressive symptom structure. These factors likely reflect intermediate pathways through which chronic disease burden coexists with emotional vulnerability. Their presence underscores the multidimensional interplay among biological, cognitive, and psychosocial domains in later-life depression [34]. Methodologically, this study integrates latent disease clustering with symptom-level network analysis, offering a refined perspective on heterogeneity within multimorbidity. Rather than treating chronic conditions as additive exposures, this approach illustrates how different disease constellations correspond to distinct depressive symptom configurations. Convergent findings across analytic strategies enhance internal coherence and support the robustness of cognition-centered associations. Clinically, the results suggest that cognitive assessment may be particularly informative in older adults with clustered cardiometabolic or visceral conditions. In addition, the centrality of motivational symptoms highlights the potential relevance of behavioral activation and function-oriented interventions within multimorbidity management. Overall, these findings support a multidimensional assessment framework that considers cognitive, behavioral, and physical domains when evaluating depression in aging populations. Limitations Several limitations warrant consideration. First, the crosssectional design precludes causal inference and limits conclusions regarding temporal ordering among multimorbidity, cognitive impairment, and depressive symptoms. Second, chronic conditions were primarily based on self-report, introducing potential misclassification bias. Third, sampling weights were not applied because the objective was exploratory etiologic inference rather than estimation of nationally representative prevalence. Consequently, findings should not be interpreted as population-level prevalence estimates and may have limited generalizability to the broader Chinese older population. Fourth, although network analysis provides insight into structural associations, estimated edges represent partial correlations rather than directional or mechanistic pathways. Fifth, although the CS coefficients exceeded the recommended threshold, network estimates may still be influenced by sampling variability, particularly for weaker edges. Replication in larger and independent samples is warranted. Sixth, unmeasured confounders may partially account for observed associations. Seventh, findings derived from a nationally representative Chinese cohort may not be directly generalizable to other cultural or healthcare contexts. Conclusion This exploratory study suggests that chronic diseases in older Chinese adults cluster into metabolic-vascular and visceralneurodegenerative patterns that are associated with distinct depressive symptom network characteristics. Across clusters, cognitive impairment consistently emerged as the most prominent bridge-related correlate linking multimorbidity and depressive symptoms. By integrating latent disease clustering, regression-based selection of bridging variables, and symptom-level network analysis, this study provides a multidimensional perspective that extends beyond simple disease counts. These findings underscore the importance of symptom-level approaches for understanding depression in the context of multimorbidity. Longitudinal and interventional studies are needed to clarify temporal relationships and to evaluate whether cognition-focused strategies may inform depression management in older adults with multimorbidity.