Background
Hand, foot, and mouth disease (HFMD) is a common communicable disease usually affecting children, particularly those aged 5 years and younger [
1]. It is most frequently caused by Coxsackievirus A16 and Enterovirus 71 [
1,
2], and is often characterized by a distinct clinical presentation of fever, or vesicular exanthema on their hands, feet, mouths, or buttocks [
3,
4].
Over the last decades, many large-scale outbreaks of HFMD were reported in East and Southeast Asia and have caused major public health concerns worldwide, especially in the affected countries [
5-
8]. In China, several outbreaks have also been reported, such as Linyi in Shandong (2007) [
9], Fuyang in Anhui (2008) [
10], Shanghai (2009) [
11], Nanchang in Jiangxi (2010) [
12], etc. Since its beginning in Fuyang, large-scale epidemic have resulted in the deaths of many children, HFMD was classified as a class C notifiable infectious disease by the Ministry of Health in China on 2 May 2008 [
13]. To date, HFMD is still one of the leading causes of child death and a serious public health issue in China.
Currently, there are still no available effective vaccines or antiviral treatments specifically for HFMD. Thus, it is quite important to identify the possible risk factors for HFMD. Many epidemiological studies have been conducted to explore individual-level risk factors (e.g., demographic factors, socio-economic determinants and behavioral factors) [
14-
20]. Although these individual factors may play a role in HFMD incidence, group-level factors are more important for public health intervention. Considering the spatial distribution characteristics of HFMD, several researches (i.e., ecological studies) have been conducted at group-level (e.g., county, community, or city) based on spatial analysis methods (e.g., spatial regression model and spatial paneled model), which was helpful to understand the emerging trend of HFMD in certain area and explore the ecological causes of disease incidence [
21], and can guide us to take the correct and timely public health interventions to prevent the outbreak. Among these, some studies found that the incidence of HFMD was associated with some climatic indicators (e.g., average temperature, relative humidity and annual sunshine hours) [
22-
26]. Therefore, the effects of climatic indicators on HFMD should be paid more attention on, especially in the context of climate change. In addition, the occurrence of HFMD presents significant seasonality, i.e., temporal characteristic [
23,
27-
31]. Therefore, spatio-temporal model should be preferred, which can not only help to recognize the spatial and temporal trend of disease, but also make prediction and guide us to formulate and implement appropriate regional public health intervention strategies to prevent and control this disease.
However, few studies from spatio-temporal interactions perspective, have attempted to explore this relationship and to assess how climate affect HFMD incidence. On the other hand, Bayesian methods can take into account possible correlations and covariates’ effects and fully utilize the related information of disease and prior knowledge, it has been extensively used in many studies and has been recognized as a powerful means to provide more robust estimates [
32-
35].
Thus, the goal of this study is to investigate the relationships between HFMD incidence and climatic indicators based on Bayesian approaches. Using the data of new cases of HFMD reported during 2008–2012 in Shandong Province, four Bayesian models were constructed and compared, the best fitting one was further adopted to estimate the effects of six key climatic indicators, with the aim of increasing our knowledge of the true underlying geographic distribution of HFMD rates.
Discussion
Shandong Province had been one of the most serious HFMD epidemic areas in China, with annual average incidence rate 104.40 per 100,000 during 2008 to 2012. Most HFMD cases (95.04%) were aged less than 6 years old, with an average male-to-female sex ratio 1.69 (see Table
1). Single seasonal peaks except double peaks in 2009, could be found between April and August during the study period (see Figure
2), which were different from other districts of China. For example, in Jiangsu Province, double peaks (the highest occurrence between April and June and the second occurring in November) appeared [
45]; in Hong Kong, warmer seasonal peak (May-July) and winter peak (October-December) were detected [
17]; in Guangdong Province, HFMD incidence peaked in April-May and September-October [
19]. These difference might be partly due to some risk factors, such as climatic, geographic and social factors [
15,
19]. In addition, spatial autocorrelation test indicated that HFMD had positive spatial autocorrelation at county level in Shandong Province (Table
2). This also confirmed that the occurrence of HFMD might be closely related to certain ecological factors in the specific area.
Considering the goodness-of-fit by the DIC, the model including the spatial and temporal interactive effects was best fitted (DIC =6938.09, Table
3), which demonstrated a strong spatio-temporal heterogeneity in HFMD risk at the county scale, with clusters of high risk areas, as previously reported [
27]. And the HFMD incidence rate was correlated with annual average temperature, annual average pressure, annual average relative humidity, annual average wind speed, and annual sunshine hours (Table
4), which were partly compatible with previous ecological analyses from the literature [
20,
22,
46], inferring that HFMD continued to be a disease related to climate. In this study, much more precise estimates of
RR were obtained from Spatio-temporal interactive model, confirming the hypothesis that the risk of the disease was related to what was occurring in the neighbourhood of each spatial unit. Figure
3 clearly displayed the spatio-temporal characteristic of
RR, the epidemic of HFMD extended continuously over time, and the spatial pattern changed from concentrating on minority counties to expanding to more areas (especially in 2009 and 2010). Also, many high risk counties were located in areas for which HFMD clusters have been detected [
27]. Certainly, these climatic factors are hard to change, while adaptive and objective prevention and control mechanism can be established according to the active detected possible impacts, e,g., advanced prevention for high risk areas.
Bayesian spatio-temporal interactive model was a valuable tool for the spatial and temporal interactive assessment of disease patterns that could help to identify county differences, and explore possible risk factors simultaneously. This method could solve most of the problems faced by traditional statistical methods, such as the spatial autocorrelation and the potential dependence between the covariates [
47]. In previous studies, Scan statistics methods had been used to determine the spatial or spatio-temporal distribution of HFMD [
23,
27-
30]. These approaches might absorb the surrounding regions and generate false-positives areas due to a lack of specificity, that is, some non-cluster areas might be determined to be clustered since they encompassed many neighbourhoods and tended to detect larger clusters than expected [
48,
49]. However, the scan statistics could be complementary to Bayesian method. That was to say, the scan statistic detected general regions in which the risk was significantly high and the Bayesian posterior distribution further helped to identify the neighbourhoods contributing strongly to the scan statistic circle [
50]. Thus, results for cluster analysis should be interpreted with knowledge of the spatial rate distribution, such as spatial Bayesian rates in particular [
51]. Furthermore, the analysis unit (i.e., county) was also considered adequately given the disease and risk factors information that was available and the spatial level at which policies were taken. However, one must note that ecologic bias was inevitable in any ecological study [
52]. In addition, some results might be biased due to the artificial grouping of observations and variables at the county level. Despite these limitations, this approach was very useful to explore spatially aggregated data and to highlight the most risky areas to conduct more accurate analysis.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.
The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
YL substantially contributed to the study design, analysis and interpretation of data, and drafted the manuscript. XW contributed to the data collection and data sorting, and critically revised the manuscript. CP participated in the data collection, and critically revised the manuscript. ZY contributed to the interpretation of results and critically revised the manuscript. HL participated in the statistical analyses and critically revised the manuscript. FX conceived of the study, and participated in its design and helped to revise the manuscript. All authors read and approved the final manuscript.