Time-Series Evaluation of an ED-Based Syndromic Alarm System Before and During the COVID-19 Pandemic

OBJECTIVE: To evaluate the performance and limitations of an emergency department (ED)-based syndromic surveillance (SyS) model for detecting respiratory infection outbreaks in Korea, across distinct age groups and epidemiological contexts before and during the COVID-19 pandemic. METHODS: We conducted a retrospective time-series analysis using nationwide data from Level 1 and 2 EDs between January 2017 and December 2022. Syndromic visits were defined as those with fever (≥38.0 °C) or respiratory symptoms. Age-stratified autoregressive integrated moving average (ARIMA) models were trained on data from January 2017-December 2018 and January 2020-December 2021 to forecast syndromic ED visits in January-December 2019 and January-December 2022, respectively. Alarms were triggered when observed visit counts exceeded both the model's 95th percentile prediction interval and historical day-of-week thresholds. Alarm performance was assessed against ED discharge diagnoses of respiratory infectious diseases. RESULTS: The system performed well under stable pre-pandemic conditions, particularly among children aged 0-4 years (3-day alarm sensitivity: 1.000; specificity: 0.964), and moderately among adults aged 65 years and older. In contrast, model performance deteriorated in 2022 under pandemic conditions, especially among adults aged 20-64 years, with alarm sensitivity dropping below 0.300. The decline was driven by persistently elevated syndromic activity during the Omicron wave, which overwhelmed the static thresholds of the ARIMA models. CONCLUSION: ED-based SyS can offer timely and specific early warning for seasonal respiratory outbreaks, particularly in pediatric populations. However, its utility is limited during sustained pandemic waves. Future surveillance systems must incorporate adaptive models, dynamic thresholds, and multiple data streams to remain effective under evolving epidemiological baselines.