Prediction models show differences in highly pathogenic avian influenza outbreaks in Japan and South Korea compared to Europe

Abstract

Avian influenza poses substantial risks to animal welfare and public health. The recent surge in highly pathogenic avian influenza (HPAI) outbreaks has led to extensive poultry culling, highlighting the need for early warning systems. Using data on H5 HPAI virus (HPAIV) occurrence from the World Organization for Animal Health and the Food and Agriculture Organization, we employed a spatial time-series modelling framework to predict occurrences in Japan and South Korea, 2020-2024. This framework decomposes time-series data into endemic and epidemic components and has previously been used to model HPAIV in Europe. We identified 1,310 HPAIV detections from 2020 to 2024, the majority being H5N1 (55.3%) and H5N8 (35.0%). These data consisted of 827 and 483 detections in wild and domestic birds, respectively. The model included seasonality and covariates in both endemic and epidemic components and revealed consistent yearly seasonal patterns. This contrasts with previous modelling of European data where seasonality changed over time. The model predicted 81% of detections as epidemic, primarily due to within-region transmission (53%), whereas only 19% were endemic. This model effectively predicts weekly H5 HPAIV detections, aiding decision-makers in identifying high-risk periods. This study confirms the robustness and usefulness of endemic-epidemic modelling of HPAIV in different regions of the world.

Keywords: Asia; Disease occurrence; Early-warning modelling; Endemic-epidemic modelling; Highly pathogenic avian influenza; Wild and domestic birds.

Fig. 1

(A) Map of reported highly pathogenic avian influenza virus (H5 subtype) detections per 10,000 km² between 2020 and 2024 (21 October 2020 to 22 May 2024) summed over 17 regions in Japan and South Korea. The dataset contained a total of 1,310 detections of varying sizes in both wild and domestic birds. (B) Waterway connections (dark blue) created for the spatial adjacency order of the regions (light blue) included in the endemic-epidemic time-series model in this study (remote islands belong to regions on the mainland and thus are considered part of the region with no need for connecting waterways). The maps were created using the package tmap in R 4.3.3.

Fig. 2

The number of reported highly pathogenic avian influenza virus (H5 subtype) detections summed over 17 regions in Japan and South Korea shown over time, 2020–2024 (21 October 2020 to 22 May 2024).

Fig. 3

Model fit for the 8 regions of Japan illustrates the relative contributions of model components for the final multivariate time-series model, with grey showing the endemic component, blue the within-region epidemic component, and orange the between-region epidemic component. The dots represent the actual counts of reported highly pathogenic avian influenza virus (H5 subtype) detections in domestic and wild birds. While the depicted actual counts from week 11–15 in 2024 are shown, they were not included in the model’s training set and therefore are not part of the model fit.

Fig. 4

Model fit for the 9 regions of South Korea illustrates the relative contributions of model components for the final multivariate time-series model, with grey showing the endemic component, blue the within-region epidemic component, and orange the between-region epidemic component. The dots represent the actual counts of reported highly pathogenic avian influenza virus (H5 subtype) detections in domestic and wild birds. While the depicted actual counts from week 11–15 in 2024 are shown, they were not included in the model’s training set and therefore are not part of the model fit.

Fig. 5

The aggregated model fit across all regions of Japan and South Korea illustrates the relative contributions of model components for both the original baseline model and the final multivariate time-series model. The dots represent the actual counts of reported highly pathogenic avian influenza virus (H5 subtype) detections in domestic and wild birds. While the depicted actual counts from week 11–15 in 2024 are shown, they were not included in the model’s training set and therefore are not part of the model fit.

Fig. 6

The simulation-based long-term forecast for the final model starts from week 43 in 2020 (left-hand dot) and extends through week 21 in 2024. The plots display the weekly number of predicted and observed detections of highly pathogenic avian influenza virus (H5 subtype) aggregated across all regions. The fan charts represent the 1% and 99% quantiles of the simulations (n = 10,000) each week, with the mean depicted as a white line. Actual reported detections are shown with open circles. Data from weeks 11 to 15 in 2024 were not used to train the model.