Spatiotemporal Distribution of Varicella in South Korea

Varicella is a highly contagious disease caused by the varicella-zoster virus (VZV). Given its tendency to cluster geographically, spatial analyses may provide better understanding of the pattern of varicella transmission. We investigated the spatial characteristics of varicella in Korea and the risk factors for varicella at a national level. Using national surveillance and demographic data, we examined the spatial distribution of incidence rates and their spatial autocorrelation and calculated Moran’s index. Spatial regression analysis was used to identify sociodemographic predictors of varicella incidence at the district level.


Conclusion
There was a temporal uptrend in the incidence of varicella in Korea from 2006 to 2017 in the setting of positive spatial associations. The varicella incidence according to geographic region varied with population density, childhood percentage, suggesting the importance of the community-level surveillance and monitoring strategies.

Background
Varicella is a highly contagious disease caused by the varicella-zoster virus (VZV) [1]. In South Korea, varicella vaccination was introduced to the National Immunization Program (NIP) in 2005. Nevertheless, the incidence of varicella increased from 22.5 to 154.8 cases per 100,000 persons from 2006 to 2017 [2].
As varicella tends to cluster geographically in susceptible populations residing in proximity, spatial analyses may provide better understanding of pattern of varicella transmission. This study, examined the Page 3/16 spatial characteristics of varicella in Korea and the risk factors for varicella at a national level.

Methods
Setting and data collection South Korea covers 100,032 km 2 and had a population of approximately 52 million in 2017. It consists of 17 provinces (si-do) divided into 250 districts (si-gun-gu). Varicella was made nationally noti able in July 2005. We used the National Noti able Disease Surveillance System database to collect the number of reported varicella cases at the district level from January 2006 through December 2017. Sociodemographic data on the population density, childhood percentage, percentage of children under 12 years of age among total population, number of hospitals per 1,000 persons, and vaccine coverage rate for each district were retrieved from the Korean National Statistics O ce. Direct standardization was used to determine the varicella incidence rate in each district.

Statistical analysis
An epidemic curve of monthly varicella cases from January 2006 through December 2017 was drawn to reveal seasonal peaks, and the annual incidence was plotted to identify the annual trend during this periods.
To examine the spatial distribution of incidence rates and their spatial autocorrelation, we visualized the district incidence rates using a 10-color scale and calculated Moran's index. To nd local clusters including 'hot spots' (high values next to high, HH), and 'cold spots' (low values next to low, LL), local indicators of spatial association (LISA) analysis was performed, and Monte Carlo simulation was used to evaluate the p-value in conducting LISA analysis. Spatial regression analysis was performed to nd sociodemographic predictors of varicella incidence at the district level. The spatial lag and spatial error model is an extension of the traditional ordinary least square (OLS) regression model that includes the spatial dependency of variables or errors in the model. We used GeoDa software (version 1.12, The University of Chicago, IL, USA) to conduct the spatial analyses and QGIS software (version 3.2.1, QGIS Development Team) to visualize maps of incidence rates and local clusters.

Temporal trend
The annual incidence of varicella increased over the 12-year study period, with a surge in 2017 (26,032 more cases than the previous year) (Fig. 1A). During this period, the incidence of varicella increased from

Spatial Pattern
The cases of varicella in the 250 districts were summarized according to surveillance years and categorized into 17 provinces (si-do) ( Table 1, Fig. 2). Figure   The spatial clustering of varicella was examined using global autocorrelation analysis (Table 2). A positive spatial autocorrelation was found for the varicella incidence over the entire surveillance period.
Moran's index ranged from 0.1400 to 0.3210 and was signi cant in all cases. Local spatial clusters were seen in the color maps (Fig. 4). During 2006-2014, the High-High (HH) clusters were mostly con ned to Gangwon-do in the northeast and neighboring Yongin-si, Yeoju-si, Ichon-gun, and Yangpyeong-gun in Gyeonggi-do. The neighboring districts also contained 'hot spot' clusters during the last surveillance period of 2015-2017. The Low-Low (LL) clusters were mostly in southern Korea during the early surveillance period. Subsequently, the clusters gradually scattered and faded.

Spatial Regression Analysis
We assumed that sociodemographic factors in uence the epidemics of varicella at the district level, such as population density, childhood percentage, number of hospitals per 1,000 persons, and vaccine coverage rate ( Table 3). The vaccine coverage rate of each province exceeded 96%, and its geographical distribution is shown in Fig. 5. Using these variables as predictors, with annual varicella incidence as the dependent variable, we tted a spatial regression model. The spatial error dependence was signi cant and explained 36.6% of the variation (Table 4), whereas the spatial lag dependence did not. Population density and number of hospitals per 1,000 persons, which is a proxy for local health infrastructure, had negative coe cients, and the former was signi cant. Proportion of children had a signi cant positive coe cient, whereas the vaccine coverage rate, which was categorized into four ordinal values based on the quartiles of its distribution to avoid multicollinearity, had a non-signi cant positive coe cient.

Discussion
In this study, we found a temporal uptrend in the incidence of varicella in Korea from 2006 to 2017 in the setting of positive spatial associations con ned to the northeast (Gangwon-do) that gradually spread and faded over time, which led to an overall increase in varicella incidence across the country. The upward trend in varicella in Korea despite the adoption of universal one-dose vaccination is consistent with previous studies, which have suggested insu cient effectiveness of the vaccine. In a population-based study, the effectiveness of the varicella vaccine was 13% (95% CI: − 17.3 to 35.6), and the immunity waned rapidly 3 years after vaccination [3]. Furthermore, a population-based study of the effectiveness of onedose varicella vaccination on disease severity suggested that one-dose varicella vaccination resulted in milder symptoms, resulting in failure to isolate patients, leading in turn to outbreaks among those in close contact, such as children in kindergarten or elementary school [4].
This is the rst study to investigate the spatial epidemic characteristics of varicella on a nation-wide scale.
Nevertheless, the occurrence of local 'hot-spots' in remote areas, such as Gangwon-do, may be similar to the results of studies on a province scale or of other respiratory diseases, such as mumps and measles. A spatiotemporal analysis of varicella in Valencia, Spain from 2008 to 2012 identi ed spatiotemporal clusters where the population was economically disadvantaged or perhaps less educated and less aware of vaccination schedules [5]. In spatio-temporal analysis of measles [6] and mumps [7] in China, high-risk clusters were mainly distributed in the urban-rural transition zones or semi-urban areas because, with parents migrating to urban areas for employment opportunities, children were left in impoverished and remote area from vaccination clinics and become susceptible to disease.
We also found that the childhood percentage had a positive effect on the incidence of varicella at the district level, whereas population density and number of hospitals per 1,000 persons had negative effects.
The spatial regression revealed that childhood percentage, a high-risk population, in uenced the incidence of varicella. In our study, childhood percentage showed vulnerability to varicella outbreaks. This concurs with a previous study conducted APC analysis of varicella incidence in Korea [8], in which the peak incidence was 4-6 years of age. In a spatial analysis of mumps in Korea, childhood percentage was a signi cant risk factor for mumps because children are more susceptible than other age groups [9].
The number of hospitals per 1,000 persons in a district, which we used to indicate the health infrastructure, also had affected the incidence of varicella, albeit not signi cantly (p-value = 0.0693). The fewer healthcare providers in a district, the higher the varicella incidence. This may be associated with the low economic status of a district, in accordance with the results of the spatial analysis mentioned above.
The lack of a relationship between the vaccine coverage rate and incidence of varicella may be explained by the high level of coverage, ranging from 92.3-100%. Given the high vaccination coverage in Koreans before the introduction of a universal one-dose varicella vaccination in 2005, the vaccine may not have affected the incidence of varicella greatly.
Our study had several limitations. First, given the data was obtained from passive collected surveillance system, there may be underreported cases especially those with mild breakthrough infections. Second, varicella cases were obtained from aggregate data at the district level, and not from individuals because of the inaccessibility of personal information. Factors such as vaccination coverage, disease severity, and socioeconomic status at the individual level, which may drive a varicella epidemic were not included in the spatial regression model and have yet to be examined in detail. Finally, there might be multicollinearity among the predictors of varicella incidence. Despite these limitations, this study is the rst to describe the spatial epidemiological characteristics of varicella using spatial analysis at the district level in Korea, and it identi ed high-risk clusters and risk factors.