Personalized Chronic Disease Follow‐Up Appointments: Risk‐Stratified Care Through Big Data

Managing patients with chronic conditions is challenging. It requires timely care adjustments based on the patient's health status. We leverage big data to optimize patient monitoring frequencies and improve treatment. Our research is motivated by the need to improve patient care at the Veterans Affairs (VA) hospitals. We propose an integrated model to better serve patients with chronic kidney disease (CKD). CKD is prevalent, complex, and costly. The demand for kidney care has steadily increased; however, there is a decline in the availability of nephrologists. We propose a finite‐horizon Markov decision process (MDP) model, which utilizes evidence‐based and data‐driven approach to identify the best follow‐up appointment schedule for patients. The MDP model helps attain an optimal dynamic treatment plan to enhance patient's quality of life. It is parameterized by data from 11 US Department of Veterans Affairs hospitals, containing 68,513 CKD patients (mostly males between 60 and 90 years old) geographically dispersed throughout the United States between January 1, 2009 and February 21, 2016. Through various estimates and assumptions, we propose an optimal monitoring policy. We find that CKD severity, comorbidities, age, and distance to nephrologist all play roles in shaping patients’ needs of care. Through the VA clinical data, we have numerically validated our recommendation and shown that it considerably outperforms the current kidney care guidelines adopted by the VA.