Duck production systems and highly pathogenic avian influenza H5N8 in France, 2016-2017

Abstract

In winter 2016-2017, Highly Pathogenic Avian Influenza (HPAI) H5N8 virus spread across Europe, causing unprecedented epizootics. France was massively affected, resulting in the culling of over 6 million poultry. Boosted regression tree (BRT) models were used to quantify the association between spatial risk factors and HPAI H5N8 infection in poultry holdings and to generate predictive maps for HPAI infection. Three datasets were combined to build the model: a dataset of the reported outbreaks in poultry, a dataset of the poultry holdings where the virus has not been reported and a set of relevant spatial risk factors, including poultry production and trade, and water bird habitat. Results identified key associations between the 'foie gras' production systems and HPAI H5N8 risk of occurrence and indicate that strengthening surveillance of fattening duck production systems and making the transportation of fattening ducks more secure would be key priority options for HPAI prevention and control.

Figure 1

Representation of predictive deviance (average and standard error) of regional and local BRT models using spatial cross validation (CV). The bars in dark grey represent a significant difference in predictive deviance values between the Set 1 models and the Set 2 models using spatial CV. Dn_Meat: Density of holdings with ducks raised for meat per commune (/ha), Dn_Fat: Density of fattening duck holdings per commune (/ha), Dn_Fat_Out: Density of fattening duck holdings with outdoor access per commune (/ha), Dn_InMvt: Density of incoming fattening duck movements per commune (/ha), Dn_OutMvt: Density of outgoing fattening duck movements per commune (/ha), Dn_FFUnit: Density of force-feeding units per commune (/ha), Dn_Sl: Density of poultry slaughter houses per commune (/ha), Dn_Hydro: Density of waterways per commune (/ha), Dist_Hydro: Distance between the commune centroids and the closest water bodies (km).

Figure 2

BRT profiles of the four predictor variables for the final local BRT models. The grey dashed lines represent the estimate for each bootstrap and the black dashed line is the mean over all bootstraps. The lines at the top of plots show the distribution of data of the variable on the X-axis. The variable Dn_Pop is log-transformed in the profile. Dn_OutMvt: Density of outgoing fattening duck movements per commune (/ha), Dn_Duck: Density of duck holdings per commune (/ha), Dn_Pop: Density of human population per commune (/ha), Dn_Chick: Density of chicken holdings per commune (/ha).

Figure 3

Predicted mean proportion of HPAI H5N8 infected holdings and probability of having at least one HPAI H5N8 infected poultry holding in the commune for the final local BRT model. The dashed black line represents the local scale.

Figure 4

Geographical distribution of the positive (with one or more HPAI H5N8 outbreaks in poultry) and negative (with no HPAI H5N8 outbreaks in poultry) communes included in the BRT models at the regional and local scales. Communes in grey colour were not considered and were those without poultry.