In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm

Area of Science:

• Computational epidemiology
• Health data science
• Biostatistics

Background:

• Multimorbidity clustering is crucial for understanding disease patterns.
• Evaluating the performance of different clustering algorithms is essential for accurate analysis.

Purpose of the Study:

• To assess the reproducibility and validity of four clustering algorithms: latent class analysis (LCA), hierarchical cluster analysis (HCA), multiple correspondence analysis followed by k-means (MCA-kmeans), and k-means (kmeans).
• To compare these algorithms for their effectiveness in identifying multimorbidity clusters.

Main Methods:

• Simulated datasets with 26 diseases in predefined clusters were used to compare algorithms via adjusted Rand Index (aRI).
• Real-world data from 50 UK general practices involving male patients (65-84 years) with 49 long-term conditions were analyzed.
• Cluster stability was assessed using 400 bootstrap samples, and within-cluster morbidity profiles were compared using Pearson correlation.

Main Results:

• In simulated data, LCA and MCA-kmeans showed the highest agreement with known clusters (largest aRI).
• In patient data, all algorithms identified a common primary cluster (20-25% of patients).
• LCA and MCA-kmeans identified a similar second cluster (7% of patients), with LCA yielding the most similar partitioning (aRI 0.54).

Conclusions:

• Latent class analysis (LCA) demonstrated superior performance in terms of reproducibility and validity for multimorbidity clustering compared to HCA, MCA-kmeans, and kmeans.
• LCA and MCA-kmeans emerged as the most promising algorithms for identifying meaningful patient subgroups in multimorbidity research.