key: cord-0879586-uep69lcg
authors: Alkhamis, Moh A.; Youha, Sarah Al; Khajah, Mohammad M.; Haider, Nour Ben; Alhardan, Sumayah; Nabeel, Ahmad; Mazeedi, Sulaiman Al; Al-Sabah, Salman K.
title: Spatiotemporal Dynamics of COVID-19 epidemic in the State of Kuwait
date: 2020-06-30
journal: Int J Infect Dis
DOI: 10.1016/j.ijid.2020.06.078
sha: 4b18134e4ea3974720944ec1a1694939e5c1c5e4
doc_id: 879586
cord_uid: uep69lcg

Abstract Objectives The prompt understanding of the temporal and spatial patterns of the epidemic on a national level is a critical step for the timely allocation of surveillance resources. Therefore, we explored the temporal and spatiotemporal dynamics of the COVID-19 epidemic in Kuwait using daily confirmed case data collected between the 23rd of February and the 7th of May, 2020. Methods We quantified the epidemic progression using the time-dependent reproductive number (R(t)), while we used the spatiotemporal scan statistic model to identify local clustering events. We accounted for the variability in the transmission dynamics within and between two socioeconomic classes, including citizens-residents and migrant workers. Results Overall, the epidemic size in Kuwait continues to grow (R(t)s ≥ 2), indicating significant ongoing spread. Significant spreading and clustering events were detected among migrant workers due to their densely populated areas and poor living conditions. However, the government's aggressive intervention measures substantially lowered epidemic growth in migrant worker areas. Yet, at a later stage of the study period, we inferred active spreading and clustering events among both socioeconomic classes. Conclusions Our analyses unveiled deeper insights into the epidemiology of COVID-19 in Kuwait and provided an important platform for rapid guidance of decisions related to intervention activities.

Since the end of 2019, the world is experiencing an unprecedented pandemic caused by the novel coronavirus (SARS-CoV-2), which causes the disease COVID-19 and has been declared by the world health organization as a public health emergency. Within a short period, COVID-19 spread rapidly to more than 190 countries, causing severe morbidities and mortalities (JHCRC, 2020) . This rapid spread has placed unparalleled implications on healthcare systems and the economy across the globe (McKee and Stuckler, 2020) . To date, there are no effective vaccinations or pharmaceutical therapies that can halt the spread of this emerging viral pandemic. Currently, strict intervention measures of rapid detection, control, and prevention implemented worldwide are the only effective strategies for minimizing the spread of COVID-19 (Lai et al., 2020 ). Yet, the necessity of such drastic actions has had a huge societal and economic impact, making these policies unsustainable to maintain as a long-term strategy.

The first cases of COVID-19 in the State of Kuwait were diagnosed in travelers, who were placed under immediate institutional quarantine, on the 23 rd February. Since then, over 12 thousand cases and 100 deaths have been reported (up to 15th of May 2020; JHCRC, 2020). Like J o u r n a l P r e -p r o o f other countries in the Arabian Gulf, a large proportion of Kuwait's population is made up of lowwage manual laborers (69.2%), who tend to come from poorer countries and have a low level of education. The average salary for a migrant worker in Kuwait has been estimated to be between $250 and 300 per month, compared to $7820.04 per month for the average Kuwaiti household (CSBoK, 2018) . This socio-economic disparity is reflected in the housing arrangement for manual laborers, where most reside in unsanitary crowded group homes to share rent expenses. In fact, a survey by Nadoum et al. has reported that 14% of its migrant laborer respondents in Kuwait had more than 14 roommates in a 2-bedroom, 2-bathroom apartment (Nadoum, 2014) . This has resulted in unique COVID-19 transmission dynamics within distinct communities in Kuwait, making public health intervention measures more difficult to track and implement effectively. Several strategies have already been tried by the government, including targeted area lockdowns, suspending all non-essential services, school, and border closures, as well as partial and full curfews. Despite this, the current daily numbers of detected cases among migrant workers are approximately 10 times the number of cases detected among Kuwaiti citizens and non-Kuwaiti residents of equivalent socioeconomic status (Portal KE-G, 2020) .

Part of the challenge of controlling the continuous rapid spread of COVID-19 is its complex spatial and temporal epidemiology due to rapid changes in the human population dynamics and its demographic and environmental drivers. This likely accounts for the notable differences in epidemic magnitude and transmission dynamics across the world. As each country has unique demographics, healthcare infrastructure, cultural, and political factors that shape the behavior of the circulating viruses, analytical tools that don't account for these heterogeneities can be ineffectual. Consequently, the use of temporal and spatial analytical tools for rapid risk-based surveillance activities during emerging pandemics can offer valuable near-real-time insights into the severity of epidemic spread and effectiveness of intervention measures within and between communities in a population. Rapid risk-based surveillance is also critical for efficient deployment of resources and early identification of high-risk groups for disease transmission, such as those with a low-socioeconomic status (Khalatbari-Soltani et al., 2020) .

In this study, we explored the temporal and spatial dynamics of COVID-19 in Kuwait, combining two analytical methods. Specifically, we quantified temporal and spatiotemporal patterns of epidemic spread at a country-level and within communities. Our objectives were to infer daily epidemic progression and identify spatiotemporal patterns of clustering events, and to utilize these metrics to assess the effectiveness of intervention measures. Moreover, we evaluated our selected analytical methods in the context of COVID-19 surveillance in Kuwait, to address some of their current utilities and limitations.

See 'Supplementary Text 1' and ' Supplementary Figure 1' for a brief description about the study setting. We obtained our data from the Ministry of Health's official passive and active surveillance records. Passive surveillance records included self-reported cases across multiple government hospitals, while active surveillance data were collected through targeted testing of arriving travelers and heavily infected areas. Our final data comprised 5988 confirmed cases detected by real-time PCR tests between 23 th of February and 7 th of May 2020 (Supplementary Table 1 ). Data included the date of detection, nationality, sex, age, and geographical location of the confirmed cases (i.e., latitude and longitude of current home address). We identified 12 resident nationalities as equivalent to the socioeconomic status of the average Kuwaiti citizen, while the remaining 31 nationalities were identified as migrant workers (Supplementary Table 1 ). We used ArcGIS version 10.5 (https://www.arcgis.com) to generate all the maps presented in this study. Also, we used a kernel density function to spatially smooth the geographical locations of the reported cases with a spatial resolution of 5m 2 .

We quantified epidemic progression of COVID-19 using the effective time-dependent reproductive numbers (R(t)) model between 23 rd of February and 7 th of May in Kuwait. We compared transmission dynamics between citizens and residents on one side and migrant workers on the other. The (R(t)) model is based on a likelihood procedure utilized for the SARS epidemic in 2002, as described elsewhere (Wallinga and Teunis, 2004) . Briefly, the method estimates R(t)s for each point of time since the beginning of the epidemic by averaging over all possible disease transmission networks compatible with the observed cases. We compute R(t)s for each day, starting from the onset day of the epidemic using the R package 'R0' (Obadia et al., 2012) . We then inferred our daily R(t)s for the epidemic as the sum of the probabilities that a given observed case was the source of infection for the subsequent case based on the elapsed time of the study period. We interpreted daily significant spreading events when the 95% CI of the inferred R(t) for a given day do not include R(t) = 1. Additionally, we used our selected model to predict daily observed cases for Kuwait, citizens, residents and migrant workers.

We used the multivariable permutation scan statistic (MPSS) model (Kulldorff, 1997 , 2001 , Kulldorff et al., 2005 , Kulldorff et al., 2007 implemented in SatScan TM to detect retrospective and prospective spatiotemporal clustering events. We performed the retrospective analysis to identify all past and current (i.e., active) significant clustering events throughout the study period (Kulldorff, 1997) . In contrast, the prospective analysis of the scan statistics was used to identify clustering events that are still active (i.e., emerging clusters) until the last day of the study period (Kulldorff, 2001) . The scan statistic method detects local clustering events using multiple hypothetical cylinders as scanning windows, in which the base and the height of each cylinder represent the spatial and temporal dimensions of the potential cluster, respectively. Each cylinder is centered at the geographical location (i.e., longitude and latitude) where a case has been detected. The permutation model of scan statistic test uses case data only within each candidate cylinder (c) and computes the ratio of the observed number cases (Oc) to expected number cases (Ec) under the null hypothesis that observed cases are randomly distributed in space and time. Ec were calculated as the sum of all observed cases, multiplied by the size of the scanning window, divided by the size of the whole study area (Kulldorff, 1997) . We used the observed-to-expected ratio (Oc/Ec) to estimate the likelihood that a candidate cylinder represents an actual significant clustering event of COVID-19 cases (Kulldorff et al., 2005) . Also, we used the multivariable extension of the scan statistic (i.e., MPSS) to simultaneously estimate Oc/Ecs and detect clustering of COVID-19 in multiple partitioned datasets (Kulldorff et al., 2007) . Since the proportion of Ocs in migrant workers was substantially higher than citizens and residents combined, we divided our data into two independent sets, to calculate adjusted numbers of Ecs for each location and time frame combination. The MPSS can infer adjusted spatiotemporal clusters and distinguish whether the significant clustering event was caused by migrant workers alone, citizens and residents alone, or both (i.e., a confounded cluster). The candidate cluster with the highest likelihood ratio test estimate were ranked as the primary most likely significant cluster, while the remaining none overlapping clusters were ranked as secondary clusters. We reported all secondary clusters with a p-value < 0.05 level.

To infer the R(t)s, we used a serial interval with a log-normal distribution and a mean = 4.7 and a standard deviation = 2.9 days (Nishiura et al., 2020) as parameter estimates to calculate the generation time from the observed epidemic curve. Also, we used 100,00 simulations to obtain the 95% confidence interval (CI) for each daily inferred R(t)s. While, for both retrospective and prospective MPSS models, we set the base and the height of the scanning windows to vary up to a maximum size equivalent to the inclusion of 50% of the reported COVID-19 cases. Also, we set the models to scan for areas with a high infection rate, using a scanning window with a maximum spatial extension of 2km 2 (average neighborhood size), and minimum temporal duration of 3 days (average time between case observation and confirmation dates). The time aggregation was set to one day due to the rapid spread of the epidemic and the short duration of study; in other words, the models work with a temporal resolution of one day. We used 999 Monte Carlo simulations to obtain the distribution of the likelihood ratio test and its corresponding p-value for each candidate cluster under the null hypothesis, described above.

In Kuwait, approximately 78.8% of the cases were detected in migrant workers, mostly of Indian nationality (40.1% of migrant workers; Supplementary Table 1) . Imported primary cases comprised approximately 11% of the total number of cases. Overall, the mean age of detected cases was 45 years old, while the proportion of males was approximately 76% (Supplementary Table 1 ). The observed epidemic curve of COVID-19 demonstrated sporadic occurrences of cases till the 30th of March, followed by a marked increase in the number of cases from the 1st of April until the end of the study period ( Figure 1 ). Figure 2 illustrates the temporal patterns of the inferred R(t)s, observed and predicted daily cases throughout the study period in Kuwait, citizens and residents combined, and migrant workers. On the level of Kuwait, the first inferred significant spreading events (R(t)s > 1) were between the 26 th of March and 4 th of April, followed by a substantial decline until the 14 th of April. Another small increase in the spreading events was inferred between the 15 th of April and the end of the study period ( Figure 2A ). No significant spreading events were inferred in citizens and resident communities until the 17 th of April. Yet, a notable rise in the daily secondary cases caused by a primary case (R(t)s  2.3) was inferred between 18 th and 25 th of April ( Figure 2C ), similar in terms of the temporal patterns to the national level R(t) curve (Figure 2A ). Our results indicate that significant spreading events among migrant workers occurred between the 26 th of March and the 5 th of April, which were marked by a distinctly high number of daily secondary cases (R(t)s ranging between 2.5 and 3.5; Figure 2E ). A sharp decline in the number of daily significant spreading events was inferred after the 5 th of April. Yet, a small rise in the number of spreading events after the 17 th of April followed another decline ( Figure 2E ). No significant spreading events were inferred at the end of the study period (Figures 2A, C, & E) . However, temporal patterns of daily and predicted cases were generally similar on national, citizens, residents and migrant workers levels ( Figures 2B, D, & F) , and were indicating that the exponential growth phase of the epidemic has been established after the 18 th of April ( Figures 2B, D, & F) .

Overall, the highest spatial point prevalence of COVID-19 cases was observed in Farwaniya (30%), followed by Asima (26%) and Hawalli (18%) governorates ( Figure 3A ). Our retrospective MPSS model inferred 25 significant spatiotemporal clustering events (p-value < 0.05) occurred throughout the study period. The temporal duration of the retrospective clusters ranged between 35 and 5 days, while the spatial extension ranged between 0.02 and 1.93 km 2 (Table 1) . Of the 25 significant clusters, 13 were identified in dormitories and housing areas mostly occupied by migrant workers located in Asima, Farwaniaya, Ahmadi governorates ( Figure 3B ). Moreover, 16 significant retrospective clusters were caused by migrant workers cases alone, while the remaining 9 clusters were caused by both citizens-residents and migrant workers combined (Table 1) . No significant clustering events were caused by citizens-residents alone. Additionally, notable differences were inferred in the Oc/Ec estimates for citizens-residents and migrant workers within the adjusted multivariable clusters (Table 1) . Our prospective MPSS model revealed 13 significant (p-value < 0.05) emerging spatiotemporal clusters with temporal duration ranging between 6 and 13 days and a spatial extension ranging between 0.02 and 2 km2 ( Figure 3C , Table 2 ). Similar to the retrospective analysis, significant clustering events were either caused by migrant workers alone or by both socioeconomic categories, with no significant clustering events caused by citizens-residents alone (Table 2) . Nevertheless, all significant active clustering events were detected in residential areas shared by both migrant workers and citizens-residents ( Figure 3C ). No substantial differences were inferred in the Oc/Ec estimates for citizens-residents and migrant workers within the adjusted clusters (Table 2 ). Finally, 9 clustering events maintained their geographical proximity to the earlier clusters ( Figures 3B and C) , while 4 clusters emerged in new geographical areas located in Jahra and Farwaniya governorates ( Figure 3C ).

Since the first case of COVID-19 was reported in Kuwait, epidemic size remained significantly small with sporadic infections for approximately over a month (Figures 1 and 2) . No significant signs of active community transmission events between the 23 rd of February and the 17 th of March 2020 (Figure 2 ). This might be attributed to the government's aggressive intervention measures of testing and forced institutional quarantine of arriving travelers after observing the rapid spread of the epidemic in neighboring countries. However, during the implementation of such measures, 4 significant clustering events were detected and were composed of citizens-residents and migrant workers cases (Table 1; Figure 3B ) in areas shared by the two communities (figure 3B). This was followed by significant a wave of spreading (R(t)s > 1) and clustering (p-values < 0.001) events initiated after the 17 th of March 2020 (Figure 2A and Table 1 ). The continuous implementation of strict control measures on infected cases appeared to help in lowering the general community transmission events in Kuwait on the 7 th of April 2020 (R(t)s  1; Figure 2A ). In spite of these efforts, 7 significant clustering events were still inferred between 25 th of March and 11 th of April, in which 5 are strictly in migrant workers' areas and 2 are in areas shared between the two communities ( Figure 3B ; Table 1 ). On the 7 th of April, the government imposed a complete zonal quarantine on Aljleeb and Mahboula neighborhoods for 28 days, where the highest density of migrant workers is located ( Figure 3A ). Further, a 28 day targeted house quarantine was enforced on migrant workers living in scattered dormitories in Eastern Asima governorate. Around the same time, the government imposed a partial curfew law on a national level, which temporarily limited significant spreading and clustering events for approximately one week (Figure 2A ; Table 1 ). Epidemic progression then notably regressed to an exponential growth phase ( Figure 2B ), which is reflected by the inferred significant spreading events ( Figure 2A ) and the occurrence of newly emergent clusters ( Figure 3C ; Table 2 ) till the end of the study period.

We found that COVID-19 transmission dynamics differed substantially between migrant workers and citizens-residents' communities ( Figures 2C, D, E, F ; Table 1 ). The immediate institutional quarantine of arriving travelers, relatively low population density (Rocklov and Sjodin, 2020, Team, 2020) , and the high socioeconomic status may have limited the spreading and clustering events within the citizen-resident community during the initial phase of the epidemic (Figures 2C & D; Table 1 ). In contrast, significant spreading and clustering events were inferred among migrant worker communities across Kuwait (Figures 2E & F; Table 1 ). This is expected, since high population density and the low socioeconomic status of migrants workers are likely to be important drivers for COVID-19 transmission and circulation among such a community (Cristina Rapone, 2020 , Liem et al., 2020 , Organization, 2020 , Rocklov and Sjodin, 2020 , Team, 2020 . Despite that, targeted and aggressive government intervention measures toward migrant worker communities were effective in temporarily lowering spreading and clustering events (Figures 2E & F; Table 1 ). The prospective cluster analysis results corroborated this finding, as a substantial decline in the number of active clustering events in the quarantined areas and dormitories was observed ( Figure 3C ). Notably, both migrant worker and citizens-residents' communities appeared to be following similar trends in spreading and clustering of infection, according to both analysis methods utilized in this study (Figure 3 ; Table 2 ). This is not surprising since migrant workers from nonquarantined areas were resuming their daily manual labor jobs and consequently probably had frequent contact with citizens and residents. Another potential reason for the newly established infection within the citizens-residents' community might be due to the frequent violation of social distancing measures prior to the holy month of Ramadan (e.g., family visits and massive grocery shopping activities). That said, the number of daily detected cases among migrant workers remains five times higher than in the citizens-residents community as of the 1 st of April till the end of the study period (Figures 1).

The high population density and poor living conditions in migrant workers' housing areas will continue to challenge the success of the currently implemented intervention measures in Kuwait and in other countries with similar social structures (Cristina Rapone, 2020 , Liem et al., 2020 , Organization, 2020 , Rocklov and Sjodin, 2020 , Team, 2020 . Unless major policy changes are implemented to improve the housing situation and wages for migrant workers, their geographical areas will become established hotspots for virus circulation and reemergence. This will result in more depletion of Kuwait's already overburdened healthcare resources (Cristina Rapone, 2020 , Liem et al., 2020 , Organization, 2020 .

Our limitations include the fact that we did not account for the direct effect of population density in our analyses. That said, we attempted the discrete Poisson model of the scan statistic test, which utilizes population density data to estimate relative risks for most likely clusters (Desjardins et al., 2020 , Kulldorff, 1997 . Unlike the permutation model, the Poisson model's results did not detect clusters in the areas with low population densities but with a high prevalence of migrant worker dormitories, particularly in Eastern Asima governorate. While our analytical methods rely on simplistic assumptions, they provided epidemiologically plausible insights into the dynamics of the epidemic in Kuwait. Unlike some models, which require extensive data preparation or computational resources, our selected analytical tools are easy to implement, making them well-suited for near-real-time surveillance and decision making. Additionally, the overparameterization and the heterogeneity of COVID-19 characteristics within countries are the main limitations affecting the reliability of more complex models in predicting rapid epidemic dynamics (Roda et al., 2020) . Nevertheless, accounting for susceptible population characteristics, asymptomatic cases, and negative tests would significantly improve our inferred results, as well as risk-based surveillance efforts (Alkhamis et al., 2012 , Desjardins et al., 2020 , Lipsitch et al., 2020 . The end date of the present study (i.e., 7 th of May) might be considered another limitation, as the epidemic reached its exponential growth phase by the end of the study period. Therefore, the number and magnitude of spreading events and emerging clusters are likely to be higher than what we inferred as the number of cases continues to increase.

Our study makes a substantial contribution to our understanding of the temporal and spatiotemporal epidemic dynamics of COVID-19 in Kuwait, highlighting the distinct spreading and clustering events within and between migrant workers and citizens-residents communities. Despite these aggressive measures, by the end of the study period, we inferred a substantial epidemic growth reflected by the predicted cases, and the number of emerging clusters. Densely populated areas and poor living conditions of migrant workers resulted in the highest number of significant spreading and clustering events within their communities. Nevertheless, targeted intervention measures within migrant worker communities substantially lowered the magnitude and number of spreading and clustering events, respectively. We emphasize the importance of maintaining migrant workers' health and living conditions in limiting the transmission of COVID-19 in Kuwait. Importantly, we demonstrated the utility of our selected analytical methods for rapid decision-making related to the allocation of surveillance and intervention resources. To our knowledge, the approach presented in this study in rapidly analyzing temporal and spatiotemporal dynamics of the COVID-19 epidemic has not been widely utilized in the Middle East.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. A) Spatially smoothed geographical locations of confirmed cases by a kernel density function (case per 5m 2 ). The green square indicates the Aljleeb neighborhood, while the purple square indicates the Mahboula neighborhood. B) Significant clusters inferred by the retrospective scan statistic model. C) Significant clusters inferred by the prospective scan statistic model. The radius of the circles (km 2 ) is proportional to the predicted spatial extent of a given cluster. The clusters are rankordered according to their inferred significance (1 = the primary most likely cluster). Confounded

SaTScanTM user guide for version 9

Serial interval of novel coronavirus (COVID-19) infections

The R0 package: a toolbox to estimate reproduction numbers for epidemic outbreaks

Protecting migrant workers during the COVID-19 pandemic: Recommendations for policy makers and constituents

COVID-19 Updates, State of Kuwait-Live Dashboard

High population densities catalyze the spread of COVID-19

Why is it difficult to accurately predict the COVID-19 epidemic?

Geographic Differences in COVID-19 Cases, Deaths, and Incidence -United States

COVID-19 Dashboard

Different epidemic curves for severe acute respiratory syndrome reveal similar impacts of control measures

This research was funded by the Kuwait Foundation for the Advancement of Science (Project No: 2020/1428).

This study was approved by the State of Kuwait Ministry of Health ethics committee. We anonymously obtained the data from the public health department's surveillance records.