Markov State Modelling of Disease Courses and Mortality Risks of Patients with Community-Acquired Pneumonia

Abstract

Community-acquired pneumonia (CAP) is one of the most frequent infectious diseases worldwide, with high lethality. Risk evaluation is well established at hospital admission, and re-evaluation is advised for patients at higher risk. However, severe disease courses may develop from all levels of severity. We propose a stochastic continuous-time Markov model describing daily development of time courses of CAP severity. Disease states were defined based on the Sequential Organ Failure Assessment (SOFA) score. Model calibration was based on longitudinal data from 2838 patients with a primary diagnosis of CAP from four clinical studies (PROGRESS, MAXSEP, SISPCT, VISEP). We categorized CAP severity into five disease states and estimated transition probabilities for CAP progression between these states and corresponding sojourn times. Good agreement between model predictions and clinical data was observed. Time courses of mortality were correctly predicted for up to 28 days, including validation with patient data not used for model calibration. We conclude that CAP disease course follows a Markov process, suggesting the necessity of daily monitoring and re-evaluation of patient’s risk. Our model can be used for regular updates of risk assessments of patients and could improve the design of clinical trials by estimating transition rates for different risk groups.