Human Factors and Implementation Evaluation of Pediatric AI Sepsis Model in the Pediatric Emergency Department

点击翻译按钮获取中文摘要

Sepsis is a leading cause of morbidity and mortality in children with estimated mortality rate of 16.9% and pediatric hospitalization cost of $7Billion. Artificial Intelligence (AI) has been touted as tool to improve sepsis outcomes and many AI models have been developed for improving outcomes related to pediatric. However, there is little high-quality evidence of improved patient outcomes in clinical studies and there are studies with conflicting outcomes related to AI interventions. Current real-world evaluations only measure predictive performance of the models, but do not provide insight into factors contributing to success/ failure of AI interventions. While human centered evaluations were done previously in laboratory settings and in simulations, we are not aware of any real-world evaluations in acute care settings. The goal of this proposal is to measure implementation outcomes and establish the feasibility of measuring the mechanism of impact via human performance aspects such as trust, situational awareness, and workload. In aim 1 of the project, we will evaluate the implementation of an existing pediatric AI sepsis model in the pediatric emergency department. In aim 2 we will demonstrate the feasibility of measuring the influence of sepsis AI on human performance in a real-world acute care setting. This project will use a combination of interview methods and electronic health record data to demonstrate the feasibility of measuring human performance. Demonstrating feasibility will provide preliminary data for a subsequent hybrid implementation trial of sepsis AI interventions in acute care settings. This line of study will ultimately yield insights to harness AI technology for improving acute and critical illness outcomes.