Forecasting Alert Fatigue in Emergency Department Workflows with Algorithmic Risk Scores

Abstract

The integration of algorithmic clinical decision support systems in emergency departments has fundamentally transformed patient care by providing real-time risk scores and automated warnings. However, the proliferation of these notifications has precipitated a critical unintended consequence known as alert fatigue, wherein clinicians become desensitized to warnings, leading to increased override rates and potential patient harm. This paper investigates the predictability of alert fatigue derived from algorithmic risk scores through a comprehensive mixed evaluation methodology situated within the complex workflow of a high-acuity emergency department. By synthesizing quantitative data from electronic health record access logs with qualitative insights gathered from clinician shadowing and structured interviews, this research delineates the multifaceted nature of alert fatigue. We propose a framework for understanding how continuous exposure to varying tiers of algorithmic risk scores deteriorates clinical responsiveness. The study identifies key workflow disruptions, cognitive burden indicators, and behavioral adaptations that emergency personnel develop in response to high-frequency alerts. The findings reveal that alert fatigue can be reliably predicted not only by the sheer volume of alerts but by the contextual misalignment between algorithmic risk scores and the dynamic clinical reality of the emergency department. By elucidating the underlying mechanisms of cognitive overload, this paper provides critical insights for the redesign of clinical decision support systems, advocating for context-aware, threshold-calibrated alert mechanisms that prioritize clinician cognitive bandwidth and ultimately safeguard patient safety.