Evaluation and Comparison of Statistical Methods for Early Detection of Respiratory Symptom Outbreaks: A Real‐World Observational Study From Iran

Background and Aims This study aimed to evaluate and compare the performance of different outbreak detection models for respiratory symptoms visits, identifying their strengths and weaknesses when applied to visit data. Methods In this observational study, data from outpatient centers in the four Iranian provinces of Gilan, Qom, Khuzestan, and Sistan and Baluchestan were analyzed. The study covered a 6‐month period from September 23, 2019, to March 19, 2020. Four outbreak detection models—Early Aberration Reporting System (EARS), Cumulative Sum (CUSUM), Farrington Model (FM), and Flexible Farrington Model (FFM)—were evaluated and compared against a gold standard for outbreak detection (defined as deviations exceeding two or three times the standard deviation). Sensitivity, specificity, positive and negative predictive values, accuracy, false alarms, missed warnings, timeliness, and Yuden index were reported. Results An average of 5857 respiratory symptom visits were recorded across the four provinces during the study period. Among the models, the FFM demonstrated superior performance. It achieved a weekly sensitivity of 75% and a daily sensitivity of 71.4%, with specificity reaching 100% weekly and 98.9% daily. The positive predictive value was 100% weekly and 96.8% daily, while the negative predictive value was 90% weekly and 87.8% daily. The model's accuracy was 92.3% on a weekly scale and 89.9% on a daily scale. FFM produced no false alarms on a weekly basis and only 3.2% false positives daily, with minimal missed warnings (10% weekly, 12.2% daily). Conclusion The FFM outperformed CUSUM, EARS, and the standard FM in detecting respiratory symptoms outbreaks, demonstrating high sensitivity, specificity, accuracy, and low false and missed alarm rates, making it a reliable tool for public health surveillance.