Persistence of highly pathogenic avian influenza H5N1 virus defined by agro-ecological niche

Abstract

The highly pathogenic avian influenza (HPAI) H5N1 virus has spread across Eurasia and into Africa. Its persistence in a number of countries continues to disrupt poultry production, impairs smallholder livelihoods, and raises the risk a genotype adapted to human-to-human transmission may emerge. While previous studies identified domestic duck reservoirs as a primary risk factor associated with HPAI H5N1 persistence in poultry in Southeast Asia, little is known of such factors in countries with different agro-ecological conditions, and no study has investigated the impact of such conditions on HPAI H5N1 epidemiology at the global scale. This study explores the patterns of HPAI H5N1 persistence worldwide, and for China, Indonesia, and India includes individual provinces that have reported HPAI H5N1 presence during the 2004-2008 period. Multivariate analysis of a set of 14 agricultural, environmental, climatic, and socio-economic factors demonstrates in quantitative terms that a combination of six variables discriminates the areas with human cases and persistence: agricultural population density, duck density, duck by chicken density, chicken density, the product of agricultural population density and chicken output/input ratio, and purchasing power per capita. The analysis identifies five agro-ecological clusters, or niches, representing varying degrees of disease persistence. The agro-ecological distances of all study areas to the medoid of the niche with the greatest number of human cases are used to map HPAI H5N1 risk globally. The results indicate that few countries remain where HPAI H5N1 would likely persist should it be introduced.

Figure 1

Akaike’s information criteria of HPAI H5N1 model groups. Standard regression model AIC of HPAI H5N1 seasonal persistence (PrS) as function of logistic regression model AIC of HPAI H5N1 human case presence (Hc) for all countries (top left), and all countries and Chinese (CHN), Indonesian (IDN), and Indian (IND) provinces (top right). Logistic regression model AIC of HPAI H5N1 human cases persistence (HcP) as function of logistic regression model AIC of HPAI H5N1 human case presence (Hc) for all countries (bottom left), and all countries and Chinese (CHN), Indonesian (IDN), and Indian (IND) provinces (bottom right).

Figure 2

Goodness of fit indicators of HPAI H5N1 model groups for all countries (top left and right), and all countries and Chinese (CHN), Indonesian (IDN), and Indian (IND) provinces (bottom left and right). All-countries models (top left and right) are represented by the R ² of the seasonal persistence standard regression model (PrS) as a function of the area under curve (AUC) of the human case presence logistic regression model (Hc). All-countries and province models (bottom left and right) are presented by the AUC of the human case persistence logistic regression model (HcP) as a function of the AUC of the human case presence logistic regression model (Hc). Left plots show all individual models (gray dots) and the median R ² or AUC for each collection of models containing a variable (black dots). The right plots show only R ² or AUC median values for collection of models containing each variable (the dotted lines are the median R ² or AUC estimated over all models). Variable abbreviations are as follows: AgPopDn agricultural population density, AgPopDnOI agricultural population density by output/input chicken, ChDn chicken density, ChMtDn chicken meat density, DuDn duck density, DuChDn duck by chicken density, OICh output/input chicken, Pppc purchasing power per capita, ToPopDn total population density, Tmin, Tmax, Trng temperature min, max, and range, UrPopDn urban population density, UrRt urban to total population ratio.

Figure 3

Cluster plot of the partitioning around medoids presented in the two principal component space (top), and map of the five agro-ecological clusters or niche in countries and Chinese, Indonesian, and Indian provinces (bottom) where H5N1 was reported (areas unsuitable to poultry production are masked in white).

Figure 4

Characterization of each niche by the number of H5N1 human cases per million people (top left), the product of duck by chicken density (DuChDn, log scale [top right]), agricultural population density (AgPopDn, log-scale [bottom left]), and purchasing power per capita (Pppc [bottom right]).

Figure 5

Distance to Henan (China) in the space defined by the variables agricultural population density (log-scale), duck density (log-scale), duck by chicken density (log-scale), chicken density (log-scale), the product of agricultural population density and chicken output/input ratio (log-scale), and purchasing power per capita (areas unsuitable to poultry production are masked in white).