This study aims to identify predictors for patients likely to be readmitted to a hospital within 28 days of discharge and to develop and validate a prediction model for identifying patients at a high risk of readmission. Numerous attempts have been made to build similar predictive models. However, the majority of existing models suffer from at least one of the following shortcomings: the model is not based on Australian Health Data; the model uses insurance claim data, which would not be available in a real-time clinical setting; the model does not consider socio-demographic determinants of health, which have been demonstrated to be predictive of readmission risk; or the model is limited to a particular medical condition and is thus limited in scope. To address these shortcomings, we built several models to predict all-cause 28-day readmission risk and included Socio-economic Indexes for Areas (SEIFA) data as proxies for socio-demographic determinants of health. Additionally, instead of using insurance claims data, which could require several weeks to process, we built our models using data that is readily available during the inpatient stay or at the time of discharge. The set of default prediction models that were examined include logistic regression, elastic net, random forest and adaptive boosting (Ada Boost). This study examined A not for profit tertiary healthcare organisation from fiscal year 2012-2013 through fiscal year 2017-2018. The out-of-sample results show that all of the models performed similarly and adequately to predict readmission risk.
Wickramasinghe, Nilmini; Degano,, Day Manuet; and McConchie,, Steven, "Real-time Prediction of the Risk of Hospital Readmissions" (2019). BLED 2019 Proceedings. 55.