Using AI, Simulations To Detect Missed Stroke Diagnoses in Emergency Departments

Two AHRQ-funded studies advance understanding of diagnostic error in stroke care. A Journal of Stroke and Cerebrovascular Diseases study used natural language processing to identify neurologically related text markers in emergency department (ED) notes—such as “language,” “motor,” and “imaging”—that may indicate missed or delayed stroke diagnoses. Predictive models using these 11 markers performed well across two academic hospitals, suggesting potential for early identification of high-risk patients. The authors noted that validation in ED settings is needed. Meanwhile, a study in Annals of Emergency Medicine used simulation and applied statistics to examine how factors like physician distraction affect diagnostic accuracy. Among 27 physicians evaluating 100 simulated cases, distractions and the absence of a witness to speak for the patient significantly reduced diagnostic confidence, with distractions having twice the impact when no witness was present. Researchers said the approach offers a promising model for studying diagnostic error and improving training, despite the small sample size and the use of simulations and not real-world settings. Identification of neurological text markers associated with risk of stroke https://pubmed.ncbi.nlm.nih.gov/40513788/ Evaluating Acute Stroke Diagnosis Using Simulation Scenarios https://pubmed.ncbi.nlm.nih.gov/40202470/