C22-10 Diagnostic Performance of Emergency Medical Dispatch Systems: A Systematic Review and Meta-Analysis

Rationale Emergency medical dispatch accuracy was identified as a top-5 research priority in prehospital emergency medicine. Despite technological advances, overtriage rates exceed 60% for stroke and 90% for out-of-hospital cardiac arrest, with substantial variation in reported diagnostic parameters. A 2018 systematic review concluded that evidence quality was insufficient for meaningful comparison, highlighting the need for updated evidence synthesis incorporating recent post-pandemic data. Methods We conducted a systematic review and meta-analysis following PRISMA-DTA guidelines, searching PubMed and Annals of Emergency Dispatch & Response from March 2017 to August 2024. Inclusion criteria required emergency medical communication center studies reporting complete diagnostic performance metrics (sensitivity, specificity, positive and negative predictive values). Studies were grouped by clinical condition for bivariate random-effects meta-analysis. Quality assessment utilized QUADAS-2 and GRADE criteria. Primary outcomes were diagnostic accuracy parameters of emergency medical dispatch decisions compared to on-site prehospital assessment as the reference standard. Results Fifty-three studies from 17 countries, including 52 million patients, were analyzed (versus 18 studies in 2018 review). Across all conditions, overall performance ranged from sensitivity 8.7-99.9% and specificity 3.9-99.5%, with mean accuracy 26.5-99.1%. Meta-analysis was feasible for three conditions. For out-of-hospital cardiac arrest detection (8 studies, prevalence 0.2-1.7%), pooled sensitivity was 74.2% (95% CI 67.8-79.8) and specificity 96.8% (95% CI 94.2-98.3), with area under summary ROC curve 0.93 (95% CI 0.91-0.95) and heterogeneity I²=78-89%. Stroke detection meta-analysis (5 studies) showed pooled sensitivity 58.1% (95% CI 52.4-63.6) and specificity 98.9% (95% CI 98.6-99.1), with AUC 0.88 (95% CI 0.85-0.91) and heterogeneity I²=45-32%. COVID-19 screening (4 studies) showed sensitivity 73.8% (95% CI 69.2-77.9) and specificity 64.1% (95% CI 58.7-69.2), with AUC 0.76 (95% CI 0.72-0.80) and heterogeneity I²=12-67% (Figure). Target condition prevalence ranged from 0.2% to 61.7%. Evidence quality was moderate-to-high in 45% of studies. In randomized trial evidence, machine learning integration increased out-of-hospital cardiac arrest detection sensitivity from 73.3% to 84.5% (p < 0.001). Nurse-staffed centers showed 11.4% higher stroke detection sensitivity than non-clinical dispatcher centers (p < 0.01). Conclusions Despite recognized heterogeneity across diverse populations and methodologies, meta-analysis demonstrates acceptable-to-excellent diagnostic performance for specific emergency conditions. Out-of-hospital cardiac arrest detection achieves discriminatory performance comparable to established medical diagnostic tests. Stroke detection shows good overall performance despite inherent symptom complexity, while COVID-19 screening faces challenges in symptom-based differentiation. Machine learning augmentation and nurse-led dispatch models demonstrate significant performance improvements. Standardized reporting frameworks addressing prevalence effects and reference standard timing are essential to advance this field.