Prediction of readmissions in hospitalized children and adolescents by machine learning

Abstract

Pediatric hospital readmission involves greater burdens for the patient and their family network, and for the health system. Machine learning can be a good strategy to expand knowledge in this area and to assist in the identification of patients at readmission risk. The objective of the study was to develop a predictive model to identify children and adolescents at high risk of potentially avoidable 30-day readmission using a machine learning approach. Retrospective cohort study with patients under 18 years old admitted to a tertiary university hospital. We collected demographic, clinical, and nutritional data from electronic databases. We apply machine learning techniques to build the predictive models. The 30-day hospital readmissions rate was 9.50%. The accuracy for CART model with bagging was 0.79, the sensitivity, and specificity were 76.30% and 64.40%, respectively. Machine learning approaches can predict avoidable 30-day pediatric hospital readmission into tertiary assistance.