Identification of Chinese plague foci from long-term epidemiological data

Abstract

Carrying out statistical analysis over an extensive dataset of human plague reports in Chinese villages from 1772 to 1964, we identified plague endemic territories in China (i.e., plague foci). Analyses rely on (i) a clustering method that groups time series based on their time-frequency resemblances and (ii) an ecological niche model that helps identify plague suitable territories characterized by value ranges for a set of predefined environmental variables. Results from both statistical tools indicate the existence of two disconnected plague territories corresponding to Northern and Southern China. Altogether, at least four well defined independent foci are identified. Their contours compare favorably with field observations. Potential and limitations of inferring plague foci and dynamics using epidemiological data is discussed.

Fig. 1.

Cluster analysis output. (A) Colors indicate the cluster group to which the local time series belongs. Note the coherent spatial distribution of time series with similar time-frequency patterns. (B) Aggregated time series for each cluster group (same color coding). In this and in the following figures, only plague infected or related regions are represented.

Fig. 2.

Ecological niches associated with each of the six subsets corresponding to NW (A), N (B), NE (C), SW (D), S (E), and SE (F). Intensity of color indicates level of suitability index (darker stands for higher suitability). The central north territory is not considered because it contains no plague report outside the epidemic period of 1894–1951.

Fig. 3.

Correspondence between inferred and sampled plague foci. (A) Human plague locations (color-filled circles; circles of a given color correspond to gridded occurrence time series that have similar time/frequency behavior) superimposed on the suspected dominant plague host territories (obtained by georeferencing the map of rodents tested positively to plague antibodies as reported in ref. 20). (B) Similarly with a georeferenced biome map from the same source. (C) Map cumulating plague niches predicted with northern and southern occurrence points as input subsets using ENM superimposed on the suspected dominant plague host territories. (D) Similarly with a georeferenced biome map from the same source. (A and B) Occurrences tend to extend outside the contours of host territories; this is simply because the classification is based on all data, whereas ENM is based only the valid set of inputs (i.e., nonepidemic occurrences). (C and D) Note good correspondence between niches predicted with climate variables and experimentally determined enzootic foci.