Uncovering Comorbidity Clusters Linked to Mood Disorders at Community Health Centers Using Unsupervised Machine Learning

Le résumé fourni par la source

Objective: To investigate the performance of machine learning algorithms in identifying comorbidity patterns associated with mood disorders among multimorbidity patients receiving care at Community Health Centers (CHCs). Methodology: This study utilizes patient electronic health records from OCHIN-affiliated CHCs between 2012 and 2023. Multimorbidity was operationalized using ICD-9 and ICD-10 codes aligned with the Charlson Comorbidity Index. Mood disorders were identified using ICD codes (ICD-9: 296.x; ICD-10: F30-F39). Patient race (White; Black or African American; Asian; American Indian or Alaska Native; Native Hawaiian or Pacific Islander; Multiracial; other; unknown), ethnicity (Hispanic; non-Hispanic; unknown), sex (female; male; unknown), and residential rurality (rural; urban/unknown) were included as covariates. To explore comorbidity patterns, we employed unsupervised machine learning using the CLARA algorithm for medoid-based clustering, informed by hierarchical clustering results. The number of clusters were chosen based on average silhouette width and within-cluster sum-of-squares elbow criteria. Results: Seven distinct clusters of patient characteristics, mental, and physical comorbidities emerged from the sample of 3,305,440 CHC patients. These clusters can be generally characterized as: 1) all male, mixed race and often Hispanic, no mood disorders, low to moderate cardiopulmonary disorder prevalence; 2) mixed sex, mostly non-Hispanic Black/African American, no mood disorders, moderate prevalence of chronic pulmonary disorder and diabetes mellitus; 3) predominately female, unknown race and often Hispanic, some mood disorders, low prevalence of comorbidities; 4) predominately female, mostly White and non-Hispanic, largely rural, all with a mood disorder, high prevalence of chronic pulmonary disease and moderate prevalence of cardiovascular and liver diseases; 5) all female, all non-Hispanic White, no mood disorders, largely rural, moderate prevalence of cardiopulmonary diseases and malignancies; 6) all female, all White, mostly Hispanic, mostly urban, low prevalence of comorbidities; 7) mixed sex, all Black/African American, mostly non-Hispanic, largely urban, high prevalence of cardiopulmonary and liver diseases. Conclusions: Although two clusters captured patients with mood disorders, they differed markedly in demographic composition and geographic context, suggesting that these aspects are important stratifying factors in behavioral health comorbidity patterns. Cluster 4 was largely White rural females, while cluster 7 was Black urban patients of mixed genders. Both clusters with mood disorders also displayed high comorbid chronic pulmonary disease, cerebrovascular disease, myocardial infarction, congestive heart failure, peripheral vascular disease, and rheumatic disease. Implications: These results suggest patterns of comorbidity clusters with mood disordered patients, where demographics and urbanicity are key cluster distinguishers. Future work will incorporate geoinformatics and environmental indicators to further elucidate the structural factors shaping behavioral health risk beyond traditional rural–urban classifications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.