Longitudinal Analysis of Readmission Risk Using Machine Learning

Unnecessary hospital readmissions are a major problem impacting millions of patients and costing billions of dollars per year. Unfortunately, accurate assessment of readmission risk remains an open problem. In this study, several methods and tools for readmission prediction were developed using UNC hospital data available from April 1, 2014 to November 1, 2014. This study investigated the change in readmission risk for patients over time to explore at which times high-risk patients can be most effectively identified. Toward this goal, multiple Machine Learning models of hospital readmission using patient history prior to admission and comparing them with baseline model which uses data during hospitalization were developed. The results of this study find that patients history did not produce better predictive performance than the baseline model that considered just hospitalization data. However, the dataset considered is small and results may not generalize to large data sets over longer period of time.Master of Science in Information Scienc