Machine learning is a powerful method for creating clinical decision support (CDS) tools, but it requires training labels which reflect the desired alert behavior. In the Phase I work for this project, investigators have developed an encoding software called HindSight that examines discharged patients' electronic health records (EHR), identifies clinicians' sepsis treatment decisions and patient outcomes, and passes these labeled examples to an online algorithm for retraining InSight, a machine-learning-based CDS tool for real-time sepsis prediction. Although HindSight has been shown to be successful in improving the performance of InSight in retrospective work, it has yet to be validated in prospective settings; therefore, in this project, the clinical utility of HindSight will be assessed through a multicenter randomized controlled trial (RCT).
Using Clinical Treatment Data in a Machine Learning Approach for Sepsis Detection