This study was performed under the regulation of the IRB Committee of The Catholic University of Korea (No. MIRB-20200609-004). This research adhered to the tenets of the Declaration of Helsinki.

RDT data with address of subjects from Gyodong-myeon and Samsan-myeon in Ganghwa-gun, Cheorwon-gun, and Goseong-gun was obtained from preceding studies, and the number of patients in each region is summarized in the Table 1 [11–13]. IgG/IgM RDT mounted with recombinant fragment of major surface antigen (SAG1A), GST-linker-SAG1A, were applied to the sera mentioned above [15]. Briefly, 10 μl of serum was applied to the RDT sample hole and eluted with RDT buffer, i.e., 0.1% casein and 1% Tween 20 in 0.1 M Tris-HCl buffer (pH 8.0) a few seconds later. Reacting bands were read by 3 investigators after 15–20 min, and its density was determined arbitrarily as 0, +, ++, and +++. Definite reactivity was determined as ++ (mid), while weaker and stronger reactivity compared to it was determined as + (weak) and +++ (strong).

Spatial scan statistics continually change the size and location of predefined shaped windows and calculate the relative risk of within and outside the window for the region of interest, identifying clusters with relatively high risk compared to other regions [16]. SaTScan, proposed by Kulldorff (citation) based on the scan statistic, consider windows of each region by gradually increasing its size and allowing the overlap between them. For every window considered, it calculates the likelihood ratio and obtains the maximum likelihood ratio to test the hypothesis that risk of current region of interest is not different from that of outside region. Since the closed form of the distribution of the statistic is not available, SaTScan employs the Monte Carlo (MC) method to conduct the test. It consists of 3 steps as follows. First, generate multiple datasets under the null hypothesis that the cluster does not exist (relative risk (RR)=1, the risk is equal in all areas). Second, calculate the test statistic from each of these random repetitive datasets and ranks those in order. Last, set the test statistic as the upper alpha % of sequence of ordered test statistics in the previous step, and use the value to conduct the test. Null hypothesis is rejected under a significant level of alpha % if the test statistic calculated from the real data is bigger than the value set by MC simulation. It is also possible to calculate the P-value by comparing the test statistic from the real data and the sequence of test statistics from MC simulation [17–20].

Shown in Table 1, a valuable cluster analysis of IgM-RDT cannot be obtained as the incidence of positive patients in IgM-RDT is significantly low, thus the analysis was only performed on IgG-RDT. IgG-RDT measurement results were recorded in 4 discrete levels (at 0, +, ++, and +++) and the proportion of positive patients is above 19.0%. Based on the observations, we choose to employ Bernoulli SaTScan model by dividing IgG-RDT results into positive and negative. For the all tests throughout the paper, we used 5% of significance level. The base point of each Ri shall be the latitude and longitude coordinates of the village hall (or community center) if there is no Ri office. If there are several halls, the coordinates of the midpoint were used. The window size is set to 25%, 15% per annum for the data in Gyodong-myeon and Samsan-myeon in Ganghwa-gun, Cheorwon-gun, and Goseong-gun. In most cases, the upper cluster region and the relative risk were the same when the maximum allowed window was 15% and 25%, while the P-value was lower when it was 15%. In Table 2 to Table 4, relative risk indicates how much the risk of the cluster and surrounding areas varies. Relative risk is an indicator of how many times positive patients will occur in the cluster than a non-cluster. Log likelihood indicates how many times the number of expected patients than observed. The greater the P-value, the less significant the cluster is. Conversely, the smaller the P-value, the more certain the area is a cluster. The R 3.6.3 program was used to visualize clusters on the map.

In Gyodong-myeon with the 25% maximum allowed window, Jiseok-ri and Insa-ri were the most likely cluster and at a significant level of 5%, with loglikelihood ratio=4.47094, relative risk=2.066, and P<0.05), indicating the risk in Jiseok-ri and Insa-ri was 2.066 times higher than other area (Table 2). Dongsan-ri was the secondary cluster (P>0.05). With the 15% maximum allowed window, Insa-ri and Jiseok-ri were the most likely-cluster and at a significant level of 5%, with loglikelihood ratio=4.556293, relative risk=2.066, and P<0.05), indicating the risk in Insa-ri and Jiseok-ri was 2.066 times higher than other area. Dongsan-ri was the secondary cluster (P>0.05). Jiseok-ri and Insa-ri were the cluster in Gyodong-myeon in 2019 (Fig. 1A).

In Samsan-myeun with the 25% maximum allowed window, Seokmo-ri was significant as the most likely cluster with loglikelihood ratio=9.559452, RR=2.211 and P-value was lower than 0.001, indicating the risk in Seokmo-ri was 2.21 times higher than other area (Table 2). The second cluster was Seokpo-ri (P>0.05). When the window was 15%, Seokpo-ri was the first cluster with P-value >0.05, not significant cluster. In the results of the SaTScan analysis in Samsan-myeon, Seokmo-ri was the most likely cluster at the maximum window 25% (Fig. 1B). We noticed that the results of detected cluster could be changed by a size of window used in analysis. In this case, Seokmo-ri was bigger than 15% of whole Samsan-myeon area, which causes a huge difference between 2 analyses using different window size. It was worth to mention that we suggest to make a decision on a window size based on the expert’s perspective and the scientific question that we want to know the answer, not only based on the numbers from statistical analyses.

With the 25% maximum allowed window, Yangji-ri and Igil-ri were the most likely cluster and at a significant level of 5%, with loglikelihood ratio=4.909, relative risk=1.886, and P<0.05, indicating the risk in Yangji-ri and Igil-ri was 1.886 times higher than other area (Table 3). Saengchang-ri and Dochang-ri were the secondary cluster, Daema-ri was the 3rd (P>0.05). When the window was 15%, Yangji-ri and Igil-ri were a most likely cluster with loglikelihood ratio=4.909, relative risk=1.886, and P<0.05. Saengchang-ri and Dochang-ri were the secondary cluster, Daema-ri was the 3rd (P>0.05). Yangji-ri and Igil-ri were the most likely cluster in Cheorwon-gun at both window sizes (Fig. 2).

According to analysis on the Goseong-gun cluster focused on Hyeonnae-myeon, Madal-ri and Baebong-ri were the most likely cluster at a significant level of 5%, with loglikelihood ratio=3.622, relative risk=1.793, and P=0.05, indicating the risk in Madal-ri and Baebong-ri was 1.793 times higher than other area (Table 4), at 25% window size. Jukjeong-ri was the secondary cluster, Chodo-ri was the 3rd (P>0.05). When the window was 15%, Madal-ri and Baebong-ri were the most likely cluster with loglikelihood ratio=3.622, relative risk= 1.793, and P-value=0.042(< 0.05), indicating the risk in Madal-ri and Baebong-ri was 1.793 times higher than other area. Jukjeong-ri was the secondary cluster (P>0.05). Madal-ri and Baebong-ri were the most likely cluster in Goseong-gun at both window size (Fig. 3).