Clusters of multiple long-term conditions in three UK datasets: a latent class analysis

Summary

Introduction: Latent class analysis (LCA) can be used to identify subgroups within populations based on unobserved variables. LCA can be used to explore whether certain long-term conditions (LTC) occur together more frequently than others in patients with multiple-long term conditions. In this manuscript we present findings from applying LCA in three large-scale UK databanks.

Methods: We applied LCA to three different UK databanks: Secure Anonymised Information Linkage databank [SAIL], UK Biobank, and Understanding Society: the UK Household Longitudinal Study [UKHLS] and four different age groups: 18-36, 37-54, 55-73, and 74+ years. The optimal number of classes in each LCA was determined using maximum likelihood. Sample size adjusted Bayesian Information Criterion (aBIC) was used to assess model fit and elbow plots and model entropy were used to assess the best number of latent classes in each model.

Results: Between three to six clusters were identified in the different datasets and age groups. Although different in detail, similar types of clusters were identified between datasets and age groups which combine disorders around similar systems incl. Cardiometabolic clusters, Pulmonary clusters, Mental health clusters, Painful conditions clusters, and cancer clusters.