Execution of intervention matters more than strategy: A lesson from the spatiotemporal assessment of COVID-19 clusters in Nepal

The novel coronavirus disease 2019 (COVID-19) has been the biggest public health problem of the present world. As the number of people suffering from the pandemic is rising, it is likely to claim more life and worsen global health and economy. Nepal, one of the developing countries in South Asia has been strongly influenced by the pandemic and struggling to contain it with multiple interventions, however, spatiotemporal dynamics of the epidemic and its linkage with various intervention strategies has not been studied yet. Here, we employed the prospective spatial-temporal analysis with SaTScan assessing dynamics of the COVID-19 cases from 23 January to 31 August 2020 at district level in Nepal. The results revealed that COVID-19 dynamics in the early stage of transmission was slower and confined in certain districts. However, from the third week of April, transmission spread rapidly across districts of Province No. 2 and Sudoorpaschim Province, primarily introduced by Nepalese citizens returning from India. Despite nationwide lockdown, nine statistically significant active and emerging clusters were detected between 23 January and 21 July 2020, whereas ten emerging clusters were observed for extended period to 31 August. The population density and population inflow from India crossing the sealed border had significant effects on the elevated risk of the epidemic. The capital city Kathmandu has become the highest-risk active cluster since August when travel restriction has been suspended. Movement restriction appears to be the most effective non-pharmaceutical intervention against the COVID-19 for resource-scarce countries with limited health care facilities. Our findings could be valuable to the health authorities within Nepal and beyond to better allocate resources and improve interventions on the pandemic for containing it efficiently.


40
The coronavirus disease 2019 (COVID-19) outbreak has been considered a Public Health recovery aggregated at district. The COVID-19 were tested using the RT-PCR in various lab 131 distributed across the country. We extracted reported positive case and joined them with district 132 shapefile collected from Department of Survey, Government of Nepal. In addition, we obtained 133 gridded population dataset in 100-meter spatial resolution for the year of 2020 from the worldpop 134 geoportal (https://www.worldpop.org/). We summarized it for each district using the zonal 135 statistics tool of ARC GIS which was used later as a base population to assess underlying risk to 136 COVID-19 in the district.

Data Analysis 140
We used geospatial analytics to characterize spatiotemporal dynamics of COVID-19 in Nepal. We 141 divided study period in two parts based on the national level lockdown employed by the federal 142 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint government of Nepal. The country was shut down from 23 March 2020, which was lifted five 143 months later in 21 July 2020. We visualized the temporal dynamics using epidemic curve and 144 several restrictions enforced in different spatiotemporal scales. The spatial distribution of 145 cumulative cases of reported COVID-19 and incidence rate before and after the cutoff date of 21 To quantify spatiotemporal dynamics of the epidemics, we used the SatScan approach (Kulldorff,148 1997) using the SaTScan version 9.6 (Kulldorff, 2018). The SaTScan statistics has been used 149 widely to identify significant spatial/ temporal and spatiotemporal disease clusters including where, is the population in a geographic area, C and P are the total number of reported cases and 163 the total estimated population, respectively. 164

165
The SaTScan also identifies secondary clusters in addition to the most likely cluster for spatial and 166 spatiotemporal scan, and orders them according to their likelihood ratio test. Equation (2) was used 167 for calculating maximum likelihood ratio that identified scanning windows with elevated risk 168 (Kulldorff, 2001). 169 CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint cases within the window Z; N is the total number of observed cases for the entire study areas 173 across all time periods; and µ(T) is the total number of expected cases in the study area across all 174 time periods. 175 Due to the assumption of uniform relative risk (RR) across a cluster, multiple geographic units can 176 belong to significant space-time clusters. To avoid that assumption, relative risk for each spatial 177 unit that belongs to a cluster is computed. The relative risk is for each location belonging to a 178 cluster is calculated as (Liu et al., 2018;Hohl et al., 2020): 179 where, 'c' is the total number reported cases, 'e' is the total number of expected cases, and 'C' is 181 the total number of observed cases in the entire study area. 182 In this study, we chose prospective space time analysis to detect emerging or active space-time 183 clusters that are still occurring at the end of the study period based on the discrete Poisson model 184 (Kulldorff, 1997(Kulldorff, , 2001. We chose discrete Poisson probability model to account heterogeneous 185 distribution of COVID-19 transmission across the space and time (Kim and Castro 2020; Masrur 186 et al., 2020). The space-time scan statistic employs moving cylinders for potential space-time 187 clusters of COVID-19 cases. We performed this analysis on daily reported cases of COVID-19 188 aggregated on 77 districts. As we were interested to locate elevated risk zones to the COVID-19, 189 high rate was chosen for further analysis. We set the upper bounds to have a maximum spatial and 190 temporal scanning window size of 10% of the population at-risk to avoid extremely large clusters; 191 and 50% of the study period, respectively. We utilized Monte Carlo testing with 9999 replications 192 to assess the statistical significance of space-time clusters with default P of 0.05. 193 To understand the space-time propagation of the transmission we computed difference of relative 194 risk between two-study periods and also detected emerging clusters using with shorter temporal 195 scan through biweekly cumulative prospective scanning approach accounting two weeks 196 incubation period (Desjardins et al., 2020) for locating the risk and newly emerged high-risk areas 197 along the timeline. 198 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

General overview of the COVID-19 in Nepal
The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint 2020. The rising trend continued in the later part of July and entire August despite the local level 216 restrictions enforced in different districts. 217 Spatial distribution of cumulative cases and district level incidence rate of COVID-19 reported 218 before and after the cutoff date is presented in the Figures 3a and 3b, respectively. By July 21, the 219 epidemic was more intense in several western districts such as Dailekh, Doti, Achham and Bajura 220 and low lying Tarai districts bordering with India including Rautahat, Kailali, Mahotari and 221 Sarlahi, although it was already spread across the country. Spatial pattern of the incidence rate was 222 slightly different than the patterns of total cumulative cases which determined by the population 223 distribution. Bajura, Doti, Achham, Dailekh were districts with higher incidence. Higher incidence 224 rates were also reported from Palpa, Parbat and Arghaghkanchi districts of Gandaki province. By 225 August 31, the epidemic had become more intense across the country ( Figure 3b). The highest 226 number of cases were reported from Kathmandu followed by Parsa, Sarlahi, Rautahat while the 227 least cases were reported from Mustang, Manang and Humla, respectively. In the same period, the 228 highest incidence was observed in Doti, followed by Bajura and Dailekh, where the incidence was 229 above 30/1000. 230 231 Figure 3. Spatial distribution of cumulative number and rate of incidence/10000 of COVID-19 232 cases from a) 23 January -21 July and b) 23 January -31 August 2020 233 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint level between January 23 and July 21, 2020 in Nepal. Table 1 provides the characteristics of these  236 clusters with varying size, relative risk and onset time and duration. The most likely cluster i.e. 237 cluster 1 and other secondary clusters; 3, 4 and 5 emerged from June 12 while cluster 6, 7 and 8 238 lately emerged almost at the end of study period. The relative risk of these clusters also varied 239 significantly. For example, the RR of cluster 1 (most likely cluster) was 16.95 while those of the 240 cluster 2 and 3 were 9.87 and 9.81, respectively. Cluster 5, 8 and 9 were low risk clusters with RR 241 less than 3.00. 242  Figure 4a shows the locations and spatial patterns of the nine emerging space-time clusters of 245 COVID-19 at the district level in Nepal between January 23 and July 23, 2020. Cluster 1 contains 246 11 districts of Karnali and Sudoorpaschim province. Cluster 2, the first secondary cluster, is the 247 largest cluster with 145 km radius and covers 19 districts of western Nepal. Cluster 3, 5 and 6 were 248 smaller compared to the first two clusters with radius 53, 44 and 27 km and number of districts 249 inside the clusters were 5, 3 and 3, respectively. Cluster 4 and 9 were single district cluster of 250 Saptari and Sindhupalchok, correspondingly. There were 28 out of 77 districts outside these 9-251 emerging clusters having RR=0; at the time of this analysis, they were non emerging COVID-19 252 risk districts. 253 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint From 23 January -21 July 2020; and b) 23 January -31 August 2020. 256

Emerging district level clusters: 23 January-31 August 2020 257
Ten statistically significant emerging space-time clusters were detected between 23 January and 258 31 August 2020. Table 2 summarizes the characteristics of these cluster in terms of size, onset 259 time, duration and relative risk level. Clusters 4, 7, 8 emerged from July 21 and persisted till the 260 end of the study period while cluster 9 and 10 were emerged lately and persisted only for few days. 261 The cluster 1, which is the most likely cluster, emerged on June 30 while clusters 2 and 3 arose on 262 first week of August. Relative risk also varied significantly among these clusters. Cluster 5 (RR 263 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint

Progression of relative risk of COVID-19 in Nepal 278
The changing patterns of relative risk (RR) over two emerging periods have been shown in the 279 Gulmi and Mahottari while moderate rise and fall in RR was observed in 11 and 12 districts 283 symbolized by light red and light green, respectively. Some districts with RR = 0 over the two 284 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint periods indicated no difference in relative risk which were regarded as "non-emerging" COVID-285 19 districts. However, it should be noted that these districts had also experienced the outbreak 286 during the study period. Some of them became emerging clusters (with elevated RR) at some point 287 in time when scanned over a shorter temporal window (Figure 6). CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review) preprint
The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review) preprint
The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint expanded but the unexpectedly high RR was more stabilize. From the beginning of July, the 314 districts with higher relative risk further expanded on the proximity of Dailekh and Kailkot and 315 the vicinity of Palpa and Syangja which continue until the first week of August. By the August 31, 316 which is the last date of study period, the elevated RR was expanded to 56 districts. 317 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

Discussion
The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint people from those areas without testing increased the cases in particular areas of Nepal. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint lockdown on 21 July, people from different districts rushed to the capital city and the number of 372 cases raised abruptly that developed a strong cluster (Cluster 5) with very high relative risk. 373 Epidemics in crowded cities disperse rapidly and have larger total attack rates than less populated 374 cities (Rader et al., 2020). is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint Table 4. District level emerging space-time clusters of COVID-19 from 23 January to 31 August 586 2020 in Nepal (RR: relative risk). All results are statistically significant at P<0.001. 587 588 589 590 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

Figure Captions
The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint 597 598 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 10, 2020. ; https://doi.org/10.1101/2020.11.07.20227520 doi: medRxiv preprint