Spatiotemporal and risk analysis of H5 highly pathogenic avian influenza in Vietnam, 2014-2017.

The aim of this study was to describe the spatiotemporal distribution of H5 HPAI outbreak reports for the period 2014-2017 and to identify factors associated with H5 HPAI outbreak reports. Throughout the study period, a total of 139 outbreaks of H5 HPAI in poultry were reported, due to either H5N1 (96 outbreaks) or H5N6 (43 outbreaks) subtype viruses. H5N1 HPAI outbreaks occurred in all areas of Vietnam while H5N6 HPAI outbreaks were only reported in the northern and central provinces. We counted the number of H5N1 and H5N6 outbreak report-positive districts per province over the four-year study period and calculated the provincial-level standardized morbidity ratio for H5N1 and H5N6 outbreak reports as the observed number of positive districts divided by the expected number. A mixed-effects, zero-inflated Poisson regression model was developed to identify risk factors for outbreak reports of each H5N1 and H5N6 subtype virus. Spatially correlated and uncorrelated random effects terms were included in this model to identify areas of the country where outbreak reports occurred after known risk factors had been accounted-for. The presence of an outbreak report in a province in the previous 6-12 months increased the provincial level H5N1 outbreak report risk by a factor of 2.42 (95% Bayesian credible interval [CrI] 1.27-4.60) while 1000 bird increases in the density of chickens decreased provincial level H5N6 outbreak report risk by a factor of 0.65 (95% CrI 0.38 to 0.97). We document distinctly different patterns in the spatial and temporal distribution of H5N1 and H5N6 outbreak reports. Most of the variation in H5N1 report risk was accounted-for by the fixed effects included in the zero-inflated Poisson model. In contrast, the amount of unaccounted-for risk in the H5N6 model was substantially greater than the H5N1 model. For H5N6 we recommend that targeted investigations should be carried out in provinces with relatively large spatially correlated random effect terms to identify likely determinants of disease. Similarly, investigations should be carried out in provinces with relatively low spatially correlated random effect terms to identify protective factors for disease and/or reasons for failure to report.