Clinical Decision Support for Septic Shock in the Emergency Department: A Cluster Randomized Trial

Area of Science:

• Pediatric Emergency Medicine
• Clinical Informatics
• Artificial Intelligence in Healthcare

Background:

• Septic shock diagnosis delays in children contribute to preventable mortality.
• Limited evidence exists for effective early recognition strategies in pediatric sepsis.
• Machine learning-based clinical decision support (CDS) was hypothesized to improve early recognition and treatment.

Purpose of the Study:

• To evaluate the effectiveness of machine learning-based CDS in increasing the proportion of children receiving timely septic shock treatment.
• To assess the impact of CDS on key clinical outcomes, including time to antibiotics and development of hypotensive septic shock.
• To determine the feasibility and provider acceptance of implementing predictive CDS in pediatric emergency departments.

Main Methods:

• A prospective, stepped-wedge, cluster randomized trial was conducted across 4 pediatric emergency departments.
• The CDS utilized machine learning models to identify high-risk pediatric patients for sepsis based on electronic health record data.
• Effectiveness was measured by the proportion of patients receiving antibiotics and fluid bolus within 1 hour of sepsis suspicion.

Main Results:

• No significant difference was observed in the primary outcome (antibiotic and bolus within 1 hour) between the intervention and control groups (39.0% vs. 38.9%).
• The CDS did not significantly impact secondary outcomes, including the incidence of hypotensive septic shock or antibiotic timeliness.
• Providers found the CDS valuable and unobtrusive, indicating good adoption and maintenance post-trial.

Conclusions:

• Implementing predictive CDS for pediatric sepsis is feasible and acceptable to healthcare providers.
• The studied CDS, despite its acceptance, did not improve the rate of early treatment for suspected sepsis or prevent progression to hypotensive shock.
• Further research may be needed to refine predictive models or implementation strategies for improving pediatric septic shock outcomes.