Random Forests on Ubiquitous Data for Heart Failure 30-Day Readmissions Prediction

摘要: Heart failure is the most common reason for unplanned hospital readmissions. Typical 30 day readmission prediction models either use data that are not readily available at majority of US hospitals or modeling techniques do provide adequate accuracy. Moreover, tendency ongoing studies to incorporate clinical only present in modern electronic health record systems (EHRs). This problematic as population affected by heart disease, rural poor, also same whose have slowest adoption rates advanced EHR systems. We apply machine learning technique random forests administrative claims predict all-cause readmissions congestive patients a system located central Virginia, USA. form two model variants based on datasets comprised procedure data, diagnosis combination both, and basic demographic data. Our results show significant predictive performance, yield importance rankings candidate variables, address high-need areas.