Improving patient flow through hospitals with machine learning based discharge prediction

Accurate predictions of hospital discharge events could help improve patient flow through hospitals and the efficiency of care delivery. However, the potential of integrating machine learning with diverse electronic health records (EHR) data for this task has not been fully explored. We used EHR data from 01 February 2017 to 31 January 2020 in Oxfordshire, UK to predict hospital discharges in the next 24 hours. We fitted separate extreme gradient boosting models for elective and emergency admissions, trained using the first two years of data and tested using the final year of data. We examined individual-level and hospital-level model performance and evaluated the impact of training data size and recency, prediction time of day, and performance in different subgroups. Our individual patient level models for elective and emergency admissions achieved AUCs of 0.87 and 0.86, AUPRCs of 0.66 and 0.64 and F1 scores of 0.61 and 0.59, respectively, substantially better than a baseline logistic regression model. Aggregating individual probabilities, the total daily number of hospital discharges could also be accurately estimated, with mean absolute errors of 8.9% (elective admissions) and 4.9% (emergency admissions). The most informative predictors included antibiotic prescriptions, other medications, and hospital capacity factors. Performance was generally robust across patient subgroups and different training strategies, but lower in patients with longer lengths of stay and those who eventually died in hospital. Our findings highlight the potential of machine learning in optimising hospital patient flow and facilitating patient care and recovery.