[Study on infectious disease outbreak detection methods based on simulation experiments and case validation]

Objective: By constructing infectious disease scenarios with different outbreak characteristics, this study systematically compared the performance of six outbreak detection methods and validated them using real epidemic data, providing scientific evidence for optimizing surveillance and early warning models and enhancing epidemic response capacity. Methods: Time series with baseline and outbreak periods were simulated by adjusting three parameter types: sequence length (long/short), baseline level (low/medium/high), and linear trend (small/medium/large), yielding 12 outbreak scenarios. Outbreak detection was performed using six methods: cumulative sum control chart (CUSUM), exponentially weighted moving average (EWMA), moving percentile method (MPM), Farrington flexible algorithm, Bayesian outbreak detection algorithm (BODA), and autoregressive integrated moving average (ARIMA). The performance of each method under different scenarios was compared. Furthermore, the methods were validated using COVID-19 data from Shanghai between November 2021 and March 2022. Results: In simulation experiments, CUSUM performed best in most scenarios, particularly in long sequences with high baseline levels (sensitivity 0.88, specificity 0.93), followed by BODA and the Farrington flexible algorithm. EWMA showed high sensitivity (0.97) in short sequences with low baseline levels. MPM and ARIMA exhibited relatively limited ability to detect sudden outbreak signals. Validation with real epidemic data yielded results consistent with the short-sequence, low-baseline simulation scenario: CUSUM achieved the highest sensitivity (1.00) and F1 score (0.84), while BODA showed a high specificity (0.97), effectively controlling false alarms. Conclusions: CUSUM and BODA are effective primary methods for infectious disease outbreaks. The Farrington flexible algorithm provides supplementary value in seasonal or long-term surveillance; EWMA is suitable for short-term monitoring, whereas MPM and ARIMA demonstrated relatively limited detection performance in this study. These findings provide empirical evidence for optimizing infectious disease surveillance and early warning systems and have important implications for enhancing early warning and emergency response capacity in public health events in China.