Dynamic Mortality Prediction in Critically Ill Children during Inter-hospital Transports to PICUs Using Explainable AI

Abstract Critically ill children who require inter-hospital transfers to Paediatric Intensive Care Units (PICUs) are sicker than other admissions and have higher mortality rates. Accurate identification of severely ill patients at risk of high mortality is crucial throughout the transport episode, with the potential to inform interventions and post-transport medical resource allocation. Current practice primarily relies on early clinical assessments within the initial hours of transport. Real-time mortality risk during transport is lacking due to the absence of data-driven assessment tools. Addressing this gap, our research introduces PROMPT (Patient-centred Real-time Outcome monitoring and Mortality PredicTion), an explainable end-to-end machine learning pipeline designed to forecast 30-day mortality in transported critically ill children. PROMPT integrates continuous time-series vital signs and medical records with episode-specific transport data to provide real-time mortality prediction. Our study collected data during inter-hospital transports of critically ill children between January 2016 and May 2021. The results demonstrated that with PROMPT, both the random forest model and logistic regression models achieved the best performance with AUROC 0.83 (95% CI: 0.79–0.86) and 0.81 (95% CI: 0.76-0.85), respectively. Employing SHapley Additive exPlanations (SHAP) for its interpretation, the model not only explains individual risk factors for outcome, but also visualises dynamic risk assessment throughout the transport episode. In conclusion, the proposed model has demonstrated proof-of-principle in predicting mortality risk in transported children and providing individual-level model interpretability during inter-hospital transports.