Global region-specific influenza early warning by establishing an epidemic intensity threshold system: a modeling study

Background Influenza exhibits marked geographical heterogeneity in epidemic patterns, yet there is no globally standardized approach for determining epidemic thresholds. This study aimed to establish a region-specific epidemic intensity threshold system and evaluate its performance in influenza early warning and peak timing simulation. Methods Weekly influenza surveillance data from 167 countries (2009–2020) were obtained from the WHO Global Influenza Surveillance and Response System (GISRS). Three indicators were calculated for each of the 18 influenza transmission zones: ILI consultation rate (ILI rate), influenza virus positivity rate (IV positivity rate), and influenza incidence. MEM models were fitted according to each zone’s epidemic pattern to establish zone-specific threshold systems. Early warning and peak timing simulation performance were evaluated across target seasons. Model performance was assessed using leave-one-out cross-validation. Results Three epidemic patterns were identified: seasonal single-peak (North America, Europe, and Temperate South America), mixed single- and dual-peak (East Asia), and year-round or irregular (most African and tropical zones). Zone-specific four-level intensity thresholds were derived. The IV positivity rate demonstrated the most robust early warning performance, with alert-to-epidemic-start differences within ±2 weeks in most zones. Exact concordance or ±1 week difference accounted for the majority of observations. Peak timing simulation using influenza incidence achieved errors within ±3 weeks in 11 of 18 evaluable observations, concentrated in Europe and East Asia. Model sensitivity and specificity exceeded 80% in zones with distinct seasonality. The ILI rate performed adequately only in data-rich zones and showed limited utility in tropical and African settings. Conclusion The region-specific threshold system achieved early warning within ±2 weeks for the IV positivity rate in most zones, and peak timing errors within ±3 weeks in 11 of 18 evaluable observations, with sensitivity and specificity exceeding 80% in zones with distinct seasonality. The IV positivity rate is the most reliable indicator for global applications, while ILI-based indicators are best suited to zones with robust surveillance infrastructure and distinct seasonality. These findings support tailoring influenza early warning strategies to local epidemic characteristics.