A latent Markov model for long-COVID symptoms: application to the Sacco hospital data

To analyze the chronic trends of long COVID-19, we propose a hidden Markov model that includes an additional state to represent the probability of recovery. Using time-to-event data with right censoring, we construct a longitudinal dataset that tracks key long COVID-19 symptoms in patients from the start of the pandemic over a two-year period. The model explicitly defines assumptions about the initial and transition probabilities between latent patient subgroups over time. Its flexible formulation, based on multinomial logit models, enables the identification of patient subgroups with specific risk factors. Through decoding, we conduct model-based probabilistic clustering to reveal differences in how patients recover from illness