Predicting Hospital Readmission and Assessing the Cost-Effectiveness of Post-Discharge Intervention Programs

Hospital readmissions are common, costly, and impose a difficult challenge for hospitals to address in the face of increasing costs and budgetary constraints. In this thesis, we develop readmission prediction models for seven clinical programs at the University Health Network using only data available shortly after patient discharge to provide a 30-day unplanned readmission risk prediction. We conduct sensitivity analysis to assess the model performance under varying operational complexity and show improved performance can be achieved using simpler models with a careful selection of predictors. Additionally, we investigate the use of readmission risk predictions in design of cost-effective intervention programs. Specifically, we answer the question of whether a hypothetical intervention program implemented over a period of time in the past with some cost and efficacy level would have been cost-effective or not. Our results demonstrate that cost-effective intervention programs are possible through the use of more targeted intervention strategies using predicted readmission risks.M.A.S