Precise Prediction for Managing Chronic Disease Readmissions

Khanna, Sankalp ;   Boyle, Justin ;   Good, Norm

Potentially preventable hospital readmissions have a crippling effect on the health of chronic disease patients and on healthcare funding and resource utilization. While several prediction models have been proposed to help identify and manage high risk patients, most offer only moderate predictive power and discriminative ability. We develop and validate several models that utilize cohort population and clinical data and are capable of precisely identifying chronic disease patients with a high risk of rehospitalization within 30 days. Cross validation and receiver operating characteristic curve analysis are used to examine the predictive power of the models. The developed models offer high precision and discrimination and outperform current state of the art models. Delivering between 73% and 79% sensitivity at 93% specificity, the models offer excellent candidate prediction algorithms for the battle against the burden of chronic disease on the public health system.