Many health systems are searching for ways to better predict hospital readmissions. Current approaches typically use data from the inpatient admission or health insurance claims. However, these models have performed poorly and do not account for other forms of available data that could enhance prediction.
We conducted a retrospective evaluation of patients discharged from Penn Medicine with conditions associated with high readmission rates. We obtained available data from the electronic health record and zip code level data to design enhanced risk prediction models. We will compare traditional regression modeling techniques to machine learning algorithms.
This study is currently being conducted.
Pennsylvania Department of Health