Markov-modulated marked Poisson processes for modeling disease dynamics based on medical claims data

Area of Science:

• Biostatistics
• Health Informatics
• Epidemiology

Background:

• Medical claims data contain informative, non-random observations reflecting underlying patient health status.
• Modeling disease dynamics requires methods that account for both the timing and content of healthcare interactions.

Purpose of the Study:

• To introduce Markov-modulated marked Poisson processes (MMMPPs) for modeling patient disease trajectories using claims data.
• To jointly model event times and event-specific information (marks) dependent on latent disease states.

Main Methods:

• Developed a framework using MMMPPs, where a continuous-time Markov chain governs the rate of healthcare interactions.
• Modeled observation processes (event times) and mark processes (event data) as state-dependent.
• Applied the MMMPP model to chronic obstructive pulmonary disease (COPD) claims data, analyzing drug use and consultation intervals.

Main Results:

• MMMPPs effectively model the informative nature of healthcare claims data.
• The model identified distinct patterns of healthcare utilization linked to disease processes in COPD patients.
• Interindividual differences in disease state dynamics and transitions were revealed.

Conclusions:

• MMMPPs provide a robust statistical framework for analyzing complex disease dynamics from longitudinal healthcare claims.
• This methodology enhances understanding of patient health trajectories and healthcare-seeking behavior.
• The approach offers valuable insights for personalized medicine and public health research.