Partially observable collaborative model for optimizing personalized treatment selection

Precision medicine that enables personalized treatment decision support has become an increasingly important research topic in chronic disease care. The main challenges in designing a treatment algorithm include modeling individual disease progression dynamics and designing adaptive treatment selection strategy. This study aims to develop an adaptive treatment selection framework tailored to an individual patient’s disease progression pattern and treatment response. We propose a Partially Observable Collaborative Model (POCM) to capture the individual variations in a heterogeneous population and optimize treatment outcomes in three stages. The POCM first infers the disease progression models by subgroup patterns using population data in stage one and then fine-tunes the models for individual patients with a small number of treatment trials in stage two. In stage three, we show how the treatment policies based on the Partially Observable Markov Decision Process (POMDP) can be tailored to individual patients by utilizing the disease models learned from the POCM. Using a simulated population of chronic depression patients, we show that the POCM can more accurately estimate the personal disease progression than the traditional method of solving a hidden Markov model. We also compare the POMDP treatment policies with other heuristic policies and demonstrate that the POCM-based policies give the highest net monetary benefits in majority of parameter settings. To conclude, the POCM method is a promising approach to model the chronic disease progression process and recommend a personalized treatment plan for individual patients in a heterogeneous population.