Japanese encephalitis (JE) is a mosquito-borne disease, which primarily occurs in rural and suburban areas of Southeast Asia and the Western Pacific region [1]–[3]. Japanese encephalitis virus (JEV) is transmitted in an enzootic cycle among mosquitoes and vertebrate amplifying hosts, primarily in domestic pigs and ardeid birds naturally [4]. Vectors, primarily Culex tritaeniorhynchus, are abundant in rural areas where their larvae breed in rice paddies and pools of stagnant water. Humans are a dead-end host and get infected when bitten by infected mosquitoes [5]. A recent literature review estimated that annually a total of 68,000 JE cases are reported worldwide [6]. JE cases present a wide spectrum of clinical manifestations, which vary from non-specific febrile flu-like illness to severe clinical manifestations including behavioral abnormality, alteration in sensorium and respiratory paralysis [7].

Historically, large human outbreaks of JE were observed in China in the 1960s (up to 15,000 JE cases were reported) and 1970s (up to 17,000 JE cases were notified) [8]. The number of JE cases has declined substantially from 20.92/100,000 in 1971 to 0.23/100,000 in 2008 since the beginning of a nationwide immunization programme in the early 1970's [9], [10]. However, JE still remains a significant public health problem in mainland China, with approximately 50% of global cases annually [6]. In recent years, evidence suggests that JEV has expanded its geographic limits within China. For example, JEV has been identified in mosquitoes and pigs in Tibet, where was previously believed to be free of JE because of its altitude [11]. Because of the high mortality and protracted severe sequelae, JE still causes a severe health burden [7], [12]–[14].

In the past decades, spatiotemporal analysis techniques have been widely used in infectious disease surveillance and outbreak investigation [15]–[17]. It is used to visualize epidemiological data, detect and evaluate hotspots or clusters and improve surveillance and efficient vector control programmes. Studies using spatiotemporal analysis have been widely used in the field of disease mapping such as Sleeping Sickness [18], dengue fever [19], [20], and malaria [21], [22], however its application in JE has been minimal [23]. Few studies have explored the spatiotemporal patterns of JE cases in China [24], [25].

In the absence of specific treatment for the disease and ineffective and unskilled vector control and management of the amplifying hosts in resource-poor regions [26], interventions need to target vaccination to areas most in need. To inform the efficient targeting of vaccination programmes, it is important to characterize the spatiotemporal pattern of JE cases. Our study was designed to partly address these gaps in knowledge and aimed to describe the nationwide JE epidemic status throughout China, to explore the presence of spatial and seasonal patterns of JE cases, to identify the spatiotemporal clusters of JE cases at the county level and hence to provide evidence-based suggestions for policy-makers and service providers for disease control and prevention.

The results of this study indicate that there was a significant spatiotemporal heterogeneity of JE throughout mainland China during 2002–2010, in that JE reported cases were clustered during the four periods under analysis. Both LISA and spatial scan statistics analysis identified similar clusters, mainly concentrated in Southwestern China and the geographical range of JE transmission increased over the study period. This concentration maybe associated with the distribution of rice-planting areas, the extent of pig rearing and the proportion of the rural population [13], [33], [34]. Rice-planting areas are man-made vectors breeding sites that are preferred by JE vectors [35]. The proportion of the rural population is usually associated with levels of poverty and the availability of funds for JE control and prevention [36]. Non-epidemic areas were mainly distributed in the North and Western regions, where the disease was probably not autochthonous, although researchers have isolated JEV from mosquito samples and detected JEV antibody in local population and pigs [37].

Spatial scan statistics, considering both spatial and temporal features, are commonly used in disease surveillance for geographical cluster detection and evaluation, for which they have been shown to have good statistical power [38]. This allows health officials to investigate disease outbreaks whether due to environmental factors, differences in behavioural factors, the transmission of an infectious agent, and the genetic makeup of the population or not [39]. The methods used in the study were previously used in Nepal in a study which showed that the distribution of JE cases had shifted, with clusters found in the central hill areas [23]. Local indicators of spatial association identify patterns in geographic units which deviate from the average of neighboring values in the surrounding locations [40]. LISA, a spatial cluster analysis, has a better chance of detecting true cluster areas with low false-positive rates especially performing well on outlier detection [41]. Elliptic version of the spatial scan statistic rather than commonly used circular version was selected in our study because it has proven to be more appropriate at detecting whether regular shapes or irregular shapes for most of situations [31], [40]–[42]. Our study identified circular clusters (e.g. primary clusters in 2002, 2003–2005, and 2006) and irregular shapes, indicating elliptic scan window should be preferentially used when the shape of the cluster is not known. The results of LISA and spatial scan statistics were consistent which indicates that these methods are reliable and could have wider applications in the fields of disease surveillance and management in China, in particular to the surveillance and monitoring of other vector-borne diseases [19]–[22].

Overall, our results show that JE cases were widespread but relatively concentrated in rural areas in China, especially in Southwestern China. Our findings indicate that preventative strategies for JE, including boosting existing surveillance, financial support from central and provincial governments, infrastructure redevelopment and in-house workforce training should be particularly focused to counties in Southwestern China identified in this study.

In addition to contributing to the scientific advancement of JE epidemiology in China, the results from this study also provide important evidence to health authorities, policy-makers and public health practitioners and other service providers to improve JE control. For example, prevention and control measures including service guideline establishment and refinement, disease and vector/host surveillance, immunization program implementation, local vector control, health education and promotion campaigns, community engagement and environmental management, should focus on the high risk areas identified in our study. Targeting of prevention strategies at high-risk clusters is likely to increase the program's effectiveness. Individuals in highest risk areas should be informed of the risk and the possibilities for risk reduction.

The results of this study have to be interpreted in light of the studies' limitations: firstly, the data are from a passive surveillance system, which means that some cases of JE might go underreported because of their sub-clinical symptoms [12], [13], [43]. Secondly, we used an elliptic scan window in the spatial scan statistics. Although the elliptic scan algorithm searches more zones than a circular scan window, it requires more computational time. In addition, the elliptic scan window has been reported to perform poorly compared to the circular window at detecting for large clusters, as it may select long and narrow string of noncontiguous areas [39].

This study described the spatiotemporal patterns of JE in mainland China and identified spatial and temporal high risk clusters at the county level over the last 9 years, with important public health implications for targeting JE control in the country. Further studies are needed to explore the role of the physical environment (e.g. landscape and climate) and social environment (e.g. human movement, farming activities, housing conditions and personal health behavior environment), population immunity, mosquito control measures in driving the spatiotemporal distribution identified in this study.