Management of Hospital Readmission Intervention Program for Diabetes Patients: A Machine Learning Approach

Abstract

This paper presents a simple intervention program to reduce the risk of hospital readmission for people with type 2 diabetes (T2D). To this end, we deployed machine learning models to identify T2D patients who are most likely to be readmitted and assessed them using both a set of numerical metrics and graphical tools. Then, we applied models to subgroups based on three demographic features---race, age, and gender---to fine tune subgroup specific predictive gains and insights on the way of profiling patients who need to be monitored carefully. Overall, we found that the proposed intervention program would be financially feasible with currently available prediction models and would help health administrators determine the scale of the proposed intervention program based on estimated benefits and costs.