Background
Measles is a highly contagious disease and a leading cause of death among children worldwide [
1]. More than 95% of measles deaths have been reported in low-middle income countries and regions with resource-poor health facilities [
2]. China, a middle-income country, is one of the member nations in the WHO Western Pacific Region, accounting for approximately 70% of the reported measles cases in this region since 2003 [
3]. According to the schedule of expanded program on immunization (EPI) in China, routine measles vaccination covers two doses of measles-containing vaccine (i.e. MCV1, MCV2). Children should be vaccinated with MCV1 at the age of 8 months, and MCV2 should be given to children between the ages of 18 and 24 months. In addition, the current schedule of supplementary immunization activities (SIA) for measles is during September 11–20 per year. Despite the routine two-dose measles vaccination coverage across most regions of China being over 90%, large-scale outbreaks of measles still occurred in Guangxi, Beijing, Zhejiang and Shangdong provinces [
4‐
7].
Prior to 2003, Guangxi was one of the provinces in China with the highest measles incidence [
8]. Since 2004, the incidence has dropped below the national average [
8]. However, large-scale measles outbreaks re-emerged in Guangxi during 2013–2014 [
9]. Therefore, measles again became a major public health concern in Guangxi. To effectively improve the timeliness of routine measles vaccination and to implement supplementary immunization activities for measles in high-risk regions, a better understanding of the spatial, temporal and spatio-temporal patterns of measles is needed. Unfortunately, these patterns remain largely unknown in Guangxi.
Scan statistics, which are used in spatio-temporal analyses, have been widely applied to explore the patterns of disease and detect high-risk diseases clusters, particularly for diseases having limited resources for prevention [
10]. However, most previous studies examined regular (i.e. circular or cylindrical shape) diseases clusters, which are unrealistic, since the true shapes of clusters are usually less regular [
11‐
16]. Few studies have explored irregularly shaped diseases clusters, which might more realistically reflect the actual, albeit complicated, patterns of disease [
17,
18]. Moreover, most studies have utilized the relative risk (RR) derived from scan statistics to identify high-risk regions [
11,
13,
14,
16]. Few studies have measured the population attributable risk percent (PAR%) to evaluate the regional burden of disease [
7]. Additionally, there is a paucity of studies on the spatio-temporal patterns of measles, particularly in Guangxi, a region with large-scale measles outbreaks.
Therefore, a spatio-temporal analysis was conducted to identify spatial, temporal and spatio-temporal irregular clusters of measles incidence at the county level in Guangxi between 2004 and 2014, integrating flexibly shaped scan statistics and PAR%. Results of this study may not only help understand the spatio-temporal shift trend of measles in Guangxi, but also facilitate evaluating the geographical inequities in the burden of measles. Moreover, findings may provide essential evidence for implementing supplementary immunization activities for measles and strengthening the surveillance of timeliness and completeness of routine measles vaccination in overburdened regions, as well as other developing countries with a similar context.
Discussion
The present study is the first attempt to detect irregular clusters of space, time, and space-time, combining flexibly shaped scan statistics and population attributable risk percent. The findings revealed that a temporal uptrend of measles incidence existed in Guangxi between 2010 and 2014, spatio-temporal clusters of measles had shifted from Central to West and Southwest Guangxi, and regions with high RR and PAR% were mainly clustered in West, Southwest and North Central Guangxi. These findings not only deepen our understanding of the spatio-temporal patterns of measles in Guangxi, but also assist in identifying regions with high risk and heavy burden of measles, where intensifying surveillance of timeliness and completeness of routine measles vaccination and implementing supplementary immunization activities for measles should be prioritized.
Yellow card alerting system, an innovative vaccination intervention way, works as a core tool for policy intervention to classify counties alerted by a yellow card into EPI blacklist [
27]. Since the yellow card alerting system, routine two-dose measles vaccination program and supplementary immunization activities for measles were implemented in Guangxi, the measles incidence has remained at a low level [
27]. However, large-scale outbreaks of measles re-emerged in Guangxi during 2013–2014, possibly attributed to the accumulation effect of a fairly large number of susceptible children in the last decade, which ultimately weakened the herd immunity against measles [
28]. The seasonal peaks of measles predominantly occurred between April and June, in addition to a minor peak in December 2007. The minor seasonal peak could be due to the abnormally high temperatures in winter during that year, which created ideal conditions for the spread of measles [
29].
Overall, purely spatial clusters of measles constantly emerged in West, Southwest, North and Central Guangxi, though the endemic areas of measles had considerably shrunk after the implementation of the expanded program on immunization. With parents migrating to urban areas for employment opportunities, numerous children were left behind in rural areas, particularly in impoverished and remote areas, where the travel time to vaccination clinics are long and the timeliness and completeness of measles vaccination was low [
9]. This might be the reason why the clusters were located in West and North Guangxi, regions with high labour exports. Simultaneously, many migrant children in urban or semi-urban areas did not receive measles vaccination and thus became susceptible to measles [
30]. In this light, it was not surprising that clusters of measles occurred in labour-importing regions as well, i.e. Liuzhou in Central and Nanning in Southwest Guangxi.
Regarding the populations vulnerable to non-immunization, previous studies revealed that both left-behind children and migrant children were the disadvantaged groups in China and suffered from inequities in timely vaccination service [
9,
30,
31]. Furthermore, these children have brought great challenges to the elimination of measles in China. Thus, the shift of space-time clusters from Central to West and Southwest Guangxi might be partly because of the dynamic changes in demographic structure among children in the above regions.
Although a similar spatial pattern of measles was identified in terms of RR and PAR%, heterogeneity still existed in Xilin, Pingguo, Liangqing, Qingxiu, Ziyuan and Sanjiang counties. These regions had low or mild levels of RR but a high level of PAR%, probably because of the high-density of children aged ≤24 months living there. In China, two doses of measles vaccine should be routinely delivered to children within 24 months after birth [
25]. Given the advantage of PAR% in evaluating the public health significance of preventive measures, the burden of measles would be reduced sharply if routine two-dose measles vaccination and supplementary immunization activities for measles were delivered on time to the age-appropriate children in regions with high PAR% [
7]. Therefore, combining PAR% with scan statistics would be a novel approach in the analysis of disease clusters.
The shape of the scanning window is critical for scan statistics and affects the ability to detect clusters of disease accurately and sensitively [
32]. So far, numerous studies have applied a circular or cylindrical window to detect regular and compact clusters [
11‐
16]. However, these windows might fail to reflect the true shapes of clusters, due to the fact that clusters, in reality, are irregular, eccentric and complex [
22,
23,
33]. Thus, regularly shaped clusters should be interpreted with caution. To accurately depict the shapes of non-compact clusters and to avoid falsely absorbing neighbouring regions where the actual hazard is not elevated into clusters, we preferred flexibly shaped scan statistics to detect irregular clusters of measles. This is another novelty of the present study.
This study has some limitations. First, the shapes of spatio-temporal clusters were somewhat inflexible. Although the elliptical-cylinder scanning window is more sensitive to clusters that are eccentric, non-compact and noncontiguous in shape than the cylindrical window, it is impossible to completely reflect the true shape of an irregular cluster. Second, the modifiable areal unit problem was inevitable, comprising a zoning effect and a scale effect [
34]. On one hand, the adjustment of administrative district at county-level might lead to a zoning effect, which would slightly affect the spatio-temporal patterns of measles. On the other hand, since data at the town-level were not available, the county-level was deemed as the spatial scale to detect clusters. The aggregated data at county-level might obscure the clusters detected at town-level, resulting in the so called ecological fallacy, which is a sub-problem of scale effect. Third, setting the maximum spatial cluster size of the scanning window is somewhat arbitrary, though we preferred 15% of population at risk recommended by other studies [
22,
23]. Fourth, the accuracy of the flexibly shaped scan statistics was not evaluated in this study [
33,
35,
36]. Fifth, PAR% was underestimated, as we used the proportion of population aged ≤24 months as the percent of exposure to measles, which might not exactly reflect the level of exposure to measles in the whole population. Lastly, factors underlying the spatio-temporal shift trend of measles were not explored quantitatively. Further studies will be conducted to determine the role of socio-economic and demographic factors, geographical barriers and vaccination provider-related determinants in driving the transmission dynamics.
Despite these limitations, our findings have important implications for vaccination policymakers, vaccination practitioners and public health authorities to take essential measures against outbreaks of measles in Guangxi. Given regions with high PAR%, which were mainly clustered in West, Southwest and North Central Guangxi, supplementary immunization activities for measles should be given priority in these regions. Given the shift of high-risk clusters of measles to West and Southwest Guangxi, a prospective real-time measles surveillance system based on prospective scan statistics should be established in these areas to improve the timeliness and completeness of routine measles vaccination among age-appropriate children. In view of the increasing temporal trend of measles incidence, more measles-related health resources should be allocated in Guangxi than in previous years, particularly in high-risk and heavy-burden regions, and more attention should be paid to develop an early-warning system of measles outbreaks.
Acknowledgments
This study is part of Xianyan Tang’s thesis to fulfill the requirements for PhD degree in Epidemiology at Prince of Songkla University and the China Medical Board under the project of “Join Research Capacity Strengthening of the Western Rural Health Network, China”. We would like to thank Yanning Ni at Guizhou University for her assistance in visualizing spatio-temporal clusters in three-dimensions.