Projects | In Progress

Improving Sepsis Care with AI-Based Clinical Decision Support

Research Areas

Communication & Decision Making

PAIR Center Research Team

MD, MSHP

Gary Weissman

Nicholas Bishop

BS, MS

Benjamin Schmid

Antonia Angeli Gazola

MD, MSCE

Meeta Prasad Kerlin

MD, MSHP

Alexander T. Moffett

Topics

Overview

Sepsis and acute respiratory distress syndrome (ARDS) are common in intensive care units (ICUs) and often require invasive mechanical ventilation (IMV). Sepsis and ARDS carry significant morbidity and mortality risks and lead to complications that impact patients’ quality of life. Further, managing IMV care for sepsis and ARDS is inherently complex and demanding. While evidence-based guidelines exist, inconsistent practices across clinicians and institutions persist.

Artificial intelligence (AI)-based clinical decision support systems (CDSS) offer a promising approach to enhancing IMV management. Currently, some AI-based CDSS models have been applied to IMV management. However, they have largely focused on one or a handful of ventilator settings. Additionally, the current AI-based CDSS models for IMV often lack rigorous evaluations necessary for validating safety, effectiveness, and usability.

This project aims to address these gaps by developing and evaluating an AI-based CDSS for IMV management of patients with sepsis and ARDS. Using data from Penn Medicine, our team will train and validate an AI-based CDSS to generate treatment recommendations for ventilated patients. Compared to other models, our AI-based CDSS will make recommendations across more than a dozen components of IMV care.

After, our team will conduct a survey featuring a Turing test among critical care clinicians at Penn Medicine and other healthcare networks across the United States. In this survey, clinicians will read a series of vignettes describing cases of patients with sepsis and ARDS. Each vignette will also feature a treatment strategy for the patient; the strategy may be generated by the AI-based CDSS or by a human clinician. For each vignette, clinicians will assess if the treatment is safe for the patient and if the treatment originated from a human or an AI-based CDSS. We aim to assess the safety and appropriateness of the treatment strategies. We also aim to investigate if the AI-based CDSS can generate treatment recommendations comparable to human clinicians.