Model-Based Clustering of Nested Lifetime Data: Profiling COVID-19 Heart Failure Patients

This study presents a novel multilevel cluster weighted model for lifetime data with random covariates and frailties. In the framework of model-based clustering, we extend cluster-weighted model to handle time-to-event responses and data with hierarchical structures. The proposed methodology allows to identify latent clusters of observations characterised by random covariates with different distributions, whose time-to-event dynamic and heterogeneity at the grouping level also differ. The objective function that defines the model is maximized using a stochastic EM algorithm tailored to right-censored lifetime data. The development of this method is motivated by a study on the survival of COVID-19 heart failure patients, admitted to multiple hospitals in Lombardy region. By applying the proposed method, we identify latent clusters of patients that differ in terms of clinical characteristics and, for each cluster, we investigate the survival pattern, the association of particular respiratory diseases with their death hazard and the supplementary effect of the facility.