Human–AI collaboration for prehospital trauma triage: Designing the On Scene Injury Severity Prediction (OSISP) model as a clinical decision support system

Abstract

Objective: This study aims to advance the On Scene Injury Severity Prediction (OSISP), an Artificial Intelligence (AI)-based model, as a Clinical Decision Support System (CDSS) that supports Emergency Medical Service (EMS) personnel during on-scene assessment of adult trauma patients. The objectives are to explore the integration of OSISP with the prehospital trauma workflow and to refine the User Interface (UI) that communicates the predictions. Methods: Workflow integration was studied in a workshop by analysis of a customer journey map created by personnel with experience of working in the EMS setting (n = 8). Literature reviews were conducted to identify key factors enabling efficient human–AI collaboration and implementation options. Identified UI components derived from workshop and literature review findings were then evaluated and selected to refine the OSISP UI. Results: The workshop derived that OSISP is a service to be used on portable IT platforms as a second opinion, support for prioritization, and support during patient assessment. The literature reviews identified key content, characteristics, and goals of communicating predictions to users. The refined UI consisted of eight information components (prediction, entered predictors, missing predictors, and model details), and four functions (notification, exploration mode, and filtering of top three entered and missing predictors), to communicate the OSISP prediction. Conclusions: The refined OSISP UI has potential to integrate well into the clinical workflow during patient assessment, as well as enhance human–AI collaboration through customizable information when communicating predictions. However, usability testing of the OSISP UI is needed to ensure clinical utility.