An intelligent intervention strategy for patients against chronic complications based on reinforcement learning

Abstract

Chronic complications refer to defects that initially have a limited impact on chronic patients’ normal lives, but that may accumulate into a major danger if they are not treated in a timely manner. A natural problem is how to take appropriate intervention strategies to prevent chronic patients from developing severe chronic complications. Using the reinforcement learning framework, we develop a personalized preventive intervention model (abbreviated as PPIM) to intelligently administering chronic patients’ oral medications across the lifespans. In the developed model, the reward of a medical decision under a patient’s personal status is up to the medication efficacies and side-effects. During each course, the medical decision with the highest estimated reward will be taken, and then the patient’ status will be updated accordingly based on a state transition function among the chronic complications stages. Theoretical analysis proves that PPIM can effectively prevent and delay the progression of chronic complications. Simulations conducted on a public dataset of 1429 diabetic patients show that compared with the fixed one size fits all intervention strategy, PPIM can reduce the five-year incidence of diabetic nephropathy by 15.42% and can delay the time for diabetic patients to develop diabetic nephropathy by 5.72 years on average.