Development to implementation to evaluation: step-by-step approach to embedding a clinical trial of a machine learning algorithm in the electronic health record

Abstract

Abstract Objective To describe the implementation process for the Precision Resuscitation with Crystalloids in Sepsis (PRECISE) trial, a pragmatic, multihospital randomized controlled trial (RCT) of a machine learning (ML) algorithm integrated into the electronic health record (EHR). Materials and Methods Implementing PRECISE involved building 4 key components into the EHR using standard tools within Epic: (1) automated inclusion criteria, (2) real-time subphenotyping. (3) randomization, and (4) medication alternative alert. Results The PRECISE trial was deployed across 6 Emory Healthcare hospitals in June 2024, spanning 6 emergency departments and 17 ICUs with more than 300 ICU beds. Discussion and Conclusion PRECISE operationalizes an ML algorithm within the Epic to identify a subgroup of sepsis patients and prompts the clinician to order the type of fluids thought to confer a mortality benefit in this subgroup. PRECISE demonstrates that precision RCTs of ML algorithms can be executed entirely within the EHR, a framework that represents a model for future learning health systems.