A cost sensitive approach to predicting 30-day hospital readmission in COPD patients

Abstract

Chronic Obstructive Pulmonary Disease is a painful chronic disease responsible for many unplanned hospital readmissions. Recent Federal legislation has begun to financially penalize hospitals which have excess patient readmissions. Predictive analytics offers a method to statistically predict which patients are at greatest risk for hospital readmission. Many readmission models currently exist, but few incorporate cost. Our research proposes several methods to directly incorporate cost into patient readmission prediction. Additionally, a method for evaluating the cost of existing models is proposed. Results show traditional evaluation methods such as AUC to have little relation to actual financial penalties (correlation = -0.21) and that dynamic cost evaluation to result in the largest cost savings.