Spatiotemporal Characteristics of Bacillary Dysentery from 2005 to 2017 in Zhejiang Province, China

Background: This study aimed to analyze the epidemiological and spatiotemporal characteristics of bacillary dysentery in Zhejiang Province and to provide the basis for its monitoring, prevention and control. Methods: This study included cases registered in China Information System for Diseases Control and Prevention from 1 January 2005 to 31 December 2017 in Zhejiang. Descriptive methods were employed to investigate the long trend of this disease: gender distribution, high-risk population, seasonality, and circular distribution was explored to detect the peak period; incidence maps were made to show the incidence trend of disease at county level; spatial autocorrelation was explored and the regions with autocorrelation were detected; and spatiotemporal scan was conducted to map out the high-risk regions of disease and how long they lasted. Statistical significance was assumed at p value of <0.05. Results: A total of 105,577 cases of bacillary dysentery were included, the incidence declining sharply from 45.84/100,000 to 3.44/100,000 with an obvious seasonal peak from July to October. Males were more predisposed to the infection than females. Pre-education children had the highest proportion among all occupation categories. Incidence in all age groups were negatively correlated with the year (p < 0.001), and the incidences were negatively correlated with the age groups in 2005–2008 (p = 0.022, 0.025, 0.044, and 0.047, respectively). Local autocorrelation showed that counties in Hangzhou were high-risk regions of bacillary dysentery. The spatiotemporal scan indicated that all clusters occurred before 2011, and the most likely cluster for disease was found in Hangzhou, Jiaxing and Huzhou. Conclusions: The incidence of bacillary dysentery in Zhejiang from 2005 to 2017 featured spatiotemporal clustering, and remained high in some areas and among the young population. Findings in this study serve as a panorama of bacillary dysentery in Zhejiang and provide useful information for better interventions and public health planning.


Background
Bacillary dysentery, a common infectious disease, is caused by the bacteria of the genus Shigella and is usually transmitted by fecal-oral route or sometimes contact through flies, food and water [1,2]. People with poor hygiene habits or in poor sanitation and with a lack of sanitary water supply are more susceptible to infection [3][4][5][6][7]. People of all ages can get infected with bacillary dysentery, and infection rates are higher among pre-school children and young adults [1]. Its incubation period is generally 1-4 days. Most patients can be cured in 1-2 weeks, but with short immunity, and the

Data Collection
This study covered the whole population in Zhejiang Province, and included all cases registered in China Information System for Diseases Control and Prevention and diagnosed as bacillary dysentery by clinical doctors in Zhejiang Province from 2005 to 2017 in accordance to the unified diagnostic criteria promulgated by Ministry of Health of the People's Republic of China [34].
Demographic data were retrieved from China Information System for Diseases Control and Prevention.

Statistical Software
In this study, ArcGIS software (version 10.1, ESRI Inc.; Redlands, CA, USA) was used for mapping and autocorrelation analysis. The spatiotemporal clusters were detected with SatScan (version 9.4. Martin Kulldorff, National Cancer Institute, Bethesda, MD, USA; Farzad Mostashari, New York City Department of Health and Mental Hygiene, NY, USA). SPSS (version 16.0, IBM Inc., Chicago, IL, USA) was employed for Spearman's rank correlation, Wilcoxon rank sum test. All results were considered statistically significant if p < 0.05 for both sides.

Circular Distribution
Seasonality of bacillary dysentery incidence in Zhejiang Province was characterized by analyzing peak month and peak days. Peak days are the day with the most cases of the disease in the whole year. Circular distribution is generally applicable to the seasonal disease with only one peak period [35]. To calculate circular distribution, 365 days in a year is changed to 360°, and one day is represented by 0.9863°. The monthly median is taken as the middle value of the group and is then converted into degrees. Thus, circular distribution is calculated as the following [36], where fi is the monthly cases of disease, αi is monthly degree, n is the yearly cases of disease, r-value is the index for degree of dispersion, and s is standard deviation of the angle.

Data Collection
This study covered the whole population in Zhejiang Province, and included all cases registered in China Information System for Diseases Control and Prevention and diagnosed as bacillary dysentery by clinical doctors in Zhejiang Province from 2005 to 2017 in accordance to the unified diagnostic criteria promulgated by Ministry of Health of the People's Republic of China [34].
Demographic data were retrieved from China Information System for Diseases Control and Prevention.

Statistical Software
In this study, ArcGIS software (version 10.1, ESRI Inc.; Redlands, CA, USA) was used for mapping and autocorrelation analysis. The spatiotemporal clusters were detected with SatScan (version 9.4. Martin Kulldorff, National Cancer Institute, Bethesda, MD, USA; Farzad Mostashari, New York City Department of Health and Mental Hygiene, NY, USA). SPSS (version 16.0, IBM Inc., Chicago, IL, USA) was employed for Spearman's rank correlation, Wilcoxon rank sum test. All results were considered statistically significant if p < 0.05 for both sides.

Circular Distribution
Seasonality of bacillary dysentery incidence in Zhejiang Province was characterized by analyzing peak month and peak days. Peak days are the day with the most cases of the disease in the whole year. Circular distribution is generally applicable to the seasonal disease with only one peak period [35]. To calculate circular distribution, 365 days in a year is changed to 360 • , and one day is represented by 0.9863 • . The monthly median is taken as the middle value of the group and is then converted into degrees. Thus, circular distribution is calculated as the following [36], where f i is the monthly cases of disease, α i is monthly degree, n is the yearly cases of disease, r-value is the index for degree of dispersion, and s is standard deviation of the angle.

Incidence Maps and Spatial Autocorrelation
The incidence map was made to show the generally distribution of bacillary dysentery at county level. Spatial autocorrelation analysis helps to map the spatial connections in adjacent geographic units, indicating the extent of uneven distribution of values [37]. It can be used to identify various spatial clusters based on spatial weight matrices at a point time [38]. In this study, global and local Moran's I, widely used in studies of infectious disease, were employed to estimate the spatial autocorrelation [39][40][41]. Global Moran's I was used to explore the spatial autocorrelation in the whole province, and local Moran's I to explore the spatial autocorrelation. Global as well as local Moran's I is between −1 and 1. A value approaching 1 indicates a positive correlation, and approaching −1 a negative correlation. Clusters were divided into high-high (HH, high incidence surrounded by high incidence), high-low (HL, high incidence surrounded by low incidence), low-high (LH, low incidence surrounded by high incidence), and low-low (LL, low incidence surrounded by low incidence) [42].

Spatiotemporal Scan
Spatiotemporal scan analysis was performed to identify the most likely clusters (with the largest LLR values), auxiliary clusters (with statistically significant LLRs), and clustering time [43]. In this study, we collected data of the permanent residents, the latitude and longitude of each county and organized the case data by month at the county level. When the high-incidence clusters were calculated, retrospective space-time analysis was performed, Poisson model was chosen as probability model to estimate high-incidence areas, and the maximum spatial cluster size was defined as a circle with a 60-kilometer radius.

Epidemiological Trends
Exclusion of those with unclear addresses (less than 1%) left a total of 105,577 patients diagnosed as bacillary dysentery in Zhejiang Province from 2005 to 2017, with an obvious decline in the incidence from 45.84/100,000 in 2005 to 3.44/100,000 in 2017 ( Figure 2). The annual average rate of decline was 6.84 per 100,000 population, but it decreased by 11.50 per 100,000 population in 2009. More than half of the infections occurred in July, August, September and October. Seasonality was obvious with a peak in summer and autumn from July to October          Gender distribution showed that the number of cases in males was higher than in females every year ( Figure 2). The incidences in all age groups declined from 2005, and the incidences and the number of cases in 0-5-year group were much larger than other age groups. The Spearman's rank correlation showed that the incidences in every age group were negatively correlated to year (p < 0.001), while the incidences were negatively correlated to age groups only during 2005-2008 (p = 0.022, 0.025, 0.044 and 0.047, respectively) ( Table 1). In terms of occupation, the proportions of pre-education children, farmers and students were consistently higher than others every year, and the differences between occupations were statistically significant (p < 0.001) ( Figure 5). Gender distribution showed that the number of cases in males was higher than in females every year ( Figure 2). The incidences in all age groups declined from 2005, and the incidences and the number of cases in 0-5-year group were much larger than other age groups. The Spearman's rank correlation showed that the incidences in every age group were negatively correlated to year (p < 0.001), while the incidences were negatively correlated to age groups only during 2005-2008 (p = 0.022, 0.025, 0.044 and 0.047, respectively) ( Table 1). In terms of occupation, the proportions of pre-education children, farmers and students were consistently higher than others every year, and the differences between occupations were statistically significant (p < 0.001) ( Figure 5).

Incidence Maps
The incidences of bacillary dysentery were mapped at county level in Zhejiang Province from 2005 to 2017 (Figure 6), and the maps indicated that counties in and around Hangzhou had higher incidences than other places. Counties in Hangzhou being excluded, the counties with high incidences were found to be scattered. The number of counties with an incidence over 10.00/100,000 declined from 78 in 2005 to 10 in 2017.

Incidence Maps
The incidences of bacillary dysentery were mapped at county level in Zhejiang Province from 2005 to 2017 (Figure 6), and the maps indicated that counties in and around Hangzhou had higher incidences than other places. Counties in Hangzhou being excluded, the counties with high incidences were found to be scattered. The number of counties with an incidence over 10.00/100,000 declined from 78 in 2005 to 10 in 2017.   (Table 2). Local autocorrelation detected 128 high-high, 10 low-high, 2 high-low and 1 low-low clusters (Figure 7). High-high clusters were observed in Hangzhou every year from 2005 to 2017.   (Table 2). Local autocorrelation detected 128 high-high, 10 low-high, 2 high-low and 1 low-low clusters (Figure 7). High-high clusters were observed in Hangzhou every year from 2005 to 2017.

Spatiotemporal Cluster Analysis
Spatiotemporal cluster analysis showed 12 high-incidence clusters of bacillary dysentery in Zhejiang Province from 2005 to 2017. High-incidence clusters were detected in 64 counties, with all clusters occurring before 2011. The most likely clusters were found from May 2005 to November 2008 (LLR = 30394.63, p < 0.001) in 18 counties (mainly in Hangzhou, Jiaxing and Huzhou). Other clusters (the third most likely, the fifth most likely, the eighth most likely, and the eleventh most likely) were observed in the coastal regions (Ningbo, Taizhou, Wenzhou and Zhoushan). In all clusters, the longest duration was 43 months and the shortest only one month (Figure 8).

Spatiotemporal Cluster Analysis
Spatiotemporal cluster analysis showed 12 high-incidence clusters of bacillary dysentery in Zhejiang Province from 2005 to 2017. High-incidence clusters were detected in 64 counties, with all clusters occurring before 2011. The most likely clusters were found from May 2005 to November 2008 (LLR = 30394.63, p < 0.001) in 18 counties (mainly in Hangzhou, Jiaxing and Huzhou). Other clusters (the third most likely, the fifth most likely, the eighth most likely, and the eleventh most likely) were observed in the coastal regions (Ningbo, Taizhou, Wenzhou and Zhoushan). In all clusters, the longest duration was 43 months and the shortest only one month (Figure 8).

Discussion
In this study, we filled some gaps in knowledge of bacillary dysentery in the aspects of its longterm trend, high-risk population, peak period and comparative spatiotemporal clustering, which provided solid evidence for targeted strategy in the prevention and control of bacterial dysentery in Zhejiang. The average incidence of bacillary dysentery declined by a large margin since 2005 to a low

Discussion
In this study, we filled some gaps in knowledge of bacillary dysentery in the aspects of its long-term trend, high-risk population, peak period and comparative spatiotemporal clustering, which provided solid evidence for targeted strategy in the prevention and control of bacterial dysentery in Zhejiang. The average incidence of bacillary dysentery declined by a large margin since 2005 to a low level (4.25 per 100,000 population) in 2014, which was lower than the average in China as a whole and other Chinese provinces [21,44,45] and lower than that reported in the United States (6.5 per 100,000 population) [46]. Such findings suggest that prevention and control have been effective. Zhejiang Province began to manage water supply and built sanitary toilets on a large scale in 2008, and these measures could have significantly reduced the incidences of infectious diseases transmitted by fecal-oral route including bacillary dysentery [47][48][49]. Some studies showed that the incidences of the infections featured seasonality with an obvious temporal peak appearing from July to October (in summer and autumn) each year, which was consistent with our results [50][51][52]. Zhejiang, a coastal province located in southeast China, enjoys a subtropical monsoon climate with high temperatures and abundant precipitation in summer, accounting for the high incidences of bacillary dysentery. Preference for cold or uncooked food in Zhejiang could be another reason for the high incidences in summer [53,54]. Declined cases from July to October meant less obvious seasonality from 2009, and the risk factors for disease also decreased at the same time, suggesting that measures including water management and hygiene improvement adopted by the government from 2008 might have made a huge contribution.
The incidences of bacillary dysentery declined by a large margin in both men and women and the incidences in the former were always statistically significantly higher than those in the latter. Men were more likely to be infected because they performed more outdoor activities, drank raw water and had poorer hygiene habits after work [55][56][57]. Consistent with other studies [21,58,59], among the pre-education children, the largest group of bacillary dysentery, 0-5-year-old group still had the highest incidences, although all incidences declined greatly, warranting strengthened monitoring and interventions for them, e.g. teaching good hygiene practices. The number of counties with incidences higher than 10.00/100,000 decreased, suggesting that the public health measures were effective in recent years. The incidence maps showed that counties in Hangzhou always had the highest incidences across the province from 2005, an observation consistent with the findings of spatiotemporal cluster scan analysis. As the provincial capital, Hangzhou enjoyed rapid economic development, dense population and many immigrants, which may have led to the high incidences of disease [60,61].
Moran's I was chosen to estimate spatial autocorrelation because it is not easily affected by skewed distribution [62]. At the same time, some studies have shown that Moran's I focuses more on the co-correlation between spatial objects, and Moran's I is more widely used [63]. The high-high clusters reflected a high impact of the high-incidence areas on surrounding area. Global autocorrelation and local autocorrelation indicated that the incidence of bacillary dysentery was not randomly distributed at county level in Zhejiang Province, and that there was a spatial correlation every year. Local autocorrelation found constant high-high clusters in Hangzhou with small fluctuations with time, implying more attention should be paid to monitoring and preventing the disease in these areas.
Spatiotemporal scan reflected clustering. In this study, clusters occurred almost every year from 2005 to 2008. A project aimed at reconstructing the water supply and lavatories was launched in Zhejiang Province in the 1980s, which was further strengthened in 2008 [49]. Given that sanitary interventions effectively reduce the intestinal infectious diseases, the project might have resulted in a sharp decline of bacillary dysentery after 2008. Social and economic development might account for the observation that all high-incidence clusters of bacillary dysentery occurred before 2011.

Conclusions
The incidence of bacillary dysentery declined in Zhejiang Province from 2005 to 2017, with high incidences in some areas and peaks in summer and autumn. The findings provide evidence of effective early interventions such as protection and health education of high-risk populations, especially males and children <5 years old; surveillance of the high-risk areas, especially in peak period every year; the management of water resources; and construction of sanitary toilets.