BACKGROUND:
Multimorbidity (MM), defined as the coexistence of two or more chronic diseases, is an increasing global health concern with a particularly high burden in sub-Saharan Africa (SSA). Understanding patterns of disease co-occurrence is essential for effective care, yet MM clustering is complex and context specific. While clustering methods have advanced MM research, their application and validation in SSA remain limited. This study aimed to identify and compare multimorbidity patterns using multiple clustering methods and ensemble approaches in rural South Africa.

METHODS:
We analysed data from the wave 4 HAALSI cohort of adults aged ≥40 years in rural South Africa, including fourteen chronic diseases. Five single clustering algorithms were applied: Latent Class Analysis (LCA), Partitioning Around Medoids (PAM), K-modes, Hierarchical Cluster Analysis (HCA), and Multiple Correspondence Analysis with K-means (MCA+K-means). Outputs were combined using two ensemble approaches—the Adaptive Clustering Ensemble (ACE) and the Cluster-based Similarity Partitioning Algorithm (CSPA)—to derive a consensus solution. Model performance was evaluated using the silhouette index (SI), Davies–Bouldin index (DBI), predictive strength (PS), Adjusted Rand Index (ARI), and Calinski–Harabasz index (CHI).

RESULTS:
The optimal number of clusters varied across single clustering methods, reflecting their differing assumptions. LCA and PAM demonstrated superior performance, whereas K-modes performed poorly. Ensemble clustering produced an interpretable eight-cluster solution, revealing distinct MM profiles. Common patterns included HIV/hypertension, HIV/dyslipidemia, anemia/depression/hypertension, and depression/dyslipidemia, highlighting interactions between infectious, cardiometabolic, and mental health conditions. Sociodemographic differences were evident, with women and older adults predominating in cardiometabolic–mental health clusters and younger individuals in HIV-related clusters.

CONCLUSIONS:
This study provides a data-driven taxonomy of multimorbidity patterns in a rural SSA population and demonstrates the value of ensemble-based approaches for improving cluster interpretability. The coexistence of infectious, metabolic, and mental health conditions challenges single-disease care models and underscores the need for integrated, person-centred healthcare strategies in rural SSA.