Circadian and Seasonal Patterns of Recorded Emergency Medical Service Workload by Diagnostic Category: A Two-Year Retrospective Analysis with Forecasting from Türkiye.

Abstract

Introduction: Emergency Medical Services (EMS) demand varies across circadian, weekly, and seasonal timescales. This study is among the first such integrated analyses in the Turkish 112 EMS context.

Study objective: The primary objective of this study was to characterize circadian patterns in overall EMS case-record volume and across seven clinical groups, as well as hourly, weekly, and seasonal variation in diagnostic composition across eight categories. Secondary objectives were to estimate adjusted temporal associations and evaluate frozen transfer of aggregate daily forecasts from 2024 to the immediately subsequent 2025 holdout.

Methods: This retrospective study analyzed 60,811 routinely collected EMS case-record rows from Karaman Province, Türkiye over 731 days in 2024-2025. Interaction, cosinor, and negative binomial (NB) models characterized temporal patterns; aggregate daily case-record volume was forecast using weekly Fourier terms with autoregressive integrated moving average (ARIMA) errors trained on 2024 and frozen for 2025. The data represented provincial case-record workload, an operational proxy for population demand rather than a census of all inbound calls or unique patients.

Results: Case-record volume peaked at 21:00 and was lowest at 06:00; adjusted volume at 21:00 was 3.53-times that at 06:00 (95% CI, 3.28-3.81). Hourly diagnostic composition differed modestly (Cramér V = 0.061). Rhythmicity was detected in all seven clinical groups (all joint-test P values < 0.001), but low full-cell R² values indicated substantial unexplained hour-date variation; acrophases ranged from 14:41 to 19:14 and relative amplitudes from 28.6% (Respiratory) to 67.9% (Trauma). Standardized Trauma share was higher in summer than winter (25.7% versus 18.1%), whereas Respiratory share was higher in winter than summer (9.7% versus 3.6%). Adjusted volume was higher on Monday than Saturday and higher in August than November (IRRs 1.11 and 1.23, respectively). In the frozen 2025 holdout, the selected model achieved mean absolute scaled error (MASE) = 0.722 and 95% prediction-interval coverage of 96.2%.

Conclusion: Clinical-group rhythms and seasonal composition differences were observed. The frozen evaluation quantified aggregate daily forecast transfer only from 2024 to the immediately subsequent 2025 within one provincial series. Findings support local monitoring of recorded evening workload and seasonal diagnostic composition, particularly the higher Trauma share in summer and the higher Respiratory share in winter.