Improving safety and quality of emergency care using machine learning-based clinical decision support at triage

Using a dataset of over 6 million ED encounters across 21 KPNC medical centers, we will use machine-learning methods to develop triage models that predict critical illness, hospital admission, and fast-track eligibility among pediatric and adult patients. We will incorporate unstructured data, identify clinically relevant probability thresholds for each outcome, and validate the models on a retrospective dataset to assess predictive accuracy and then prospectively in the EHR. We will pilot test at 2 EDs to assess feasibility, incorporating stakeholder engagement. Lastly, we plan a pragmatic, step-wedge trial to study the impact of the tool on safety and accuracy of triage predictions, disparities in triage predictions, ED length of stay, and nursing acceptance and adoption of the tool.

0