Emergency Department Revisit Quality Review Screening: Exploring Human Decision-Making and Artificial Intelligence Support

An exploratory study investigated Emergency Department (ED) return visit quality assurance processes, specifically focusing on revisits within 1-14 days. Current practices often limit reviews to very short timeframes (e.g., 48-72 hours) to manage workload, potentially missing quality improvement opportunities. This research explored the use of clinicians and a large language model (GPT-4) to assess diagnosis pairs for their potential to warrant further quality assessment.

Why it matters

This research is strategically important as it addresses the efficiency and effectiveness of quality assurance processes in critical service delivery environments. Improving the identification of quality issues, potentially through AI augmentation, can lead to enhanced patient safety, better resource utilization, and sustained trust in institutional services. It points to a future where AI tools could significantly scale quality oversight without proportionally increasing human burden.

Key insights

• Existing ED quality assurance protocols for return visits are often constrained by narrow timeframes (e.g., 48-72 hours), which may restrict the identification of all relevant quality improvement areas.
• The study retrospectively analyzed randomly selected ED visits within a multihospital system that experienced a return visit within 1-14 days.
• Both clinicians and a large language model (GPT-4) were engaged as raters to evaluate characteristics of diagnosis pairs, with a focus on identifying those warranting additional assessment, given only the primary diagnosis of each visit.
• The methodology involved analyzing rater responses to inform subsequent steps, indicating an exploration of both human and AI capabilities in quality screening.
• The study's design implies an interest in expanding the scope of quality reviews beyond current limitations by potentially leveraging technological support.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.10421