Modelling the Progression of Chronic Diseases Using Hidden Markov Models with Electronic Health Records

Progressive diseases like diabetes, chronic kidney disease (CKD), and cardiovascular diseases (CVD) have progressively worsening over time. It is vital to capture and forecast the hidden transitions between disease stages for early intervention. This paper reports a model of disease progression based on Hidden Markov Models (HMMs) learned from longitudinal Electronic Health Records (EHRs). By modeling patient data routinely collected in clinical practice, we predict the probability of unobserved states of disease and future progression trajectories. Our findings show that HMMs can usefully detect latent disease states and transitions, yielding important information on individual disease trajectories and contributing to personalized treatment planning.