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

• DOI: 10.1016/j.jclinepi.2022.10.011
• Publication Date: 2022-10-11T01:24:52Z
• Authors: Linda Nichols; Tom Taverner; Francesca Crowe; Sylvia Richardson; Christopher Yau; Steven Kiddle; Paul Kirk; Jessica Barrett; Krishnarajah Nira...; Simon Griffin; Duncan Edwards; Tom Marshall
• Keywords: K-Means Clustering; Hierarchical clustering

Background and ObjectivesTo investigate the reproducibility validity of latent class analysis (LCA) hierarchical cluster (HCA), multiple correspondence followed by k-means (MCA-kmeans) (kmeans) for multimorbidity clustering.MethodsWe first investigated clustering algorithms in simulated datasets with 26 diseases varying prevalence predetermined clusters, comparing derived clusters to known using adjusted Rand Index (aRI). We then them medical records male patients, aged 65 84 years from 50 UK general practices, 49 long-term health conditions. compared within morbidity profiles Pearson correlation coefficient assessed stability 400 bootstrap samples.ResultsIn datasets, closest agreement (largest aRI) was LCA MCA-kmeans algorithms. In dataset, all four identified one 20–25% dataset about 82% same patients across both found a second 7% dataset. Other were only algorithm. gave most similar partitioning (aRI 0.54).ConclusionLCA achieved higher aRI than other