key: cord-321340-hwds5rja
authors: Sun, H.; Dickens, B. L.; Cook, A. R.; Clapham, H. E.
title: Importations of COVID-19 into African countries and risk of onward spread
date: 2020-05-24
journal: nan
DOI: 10.1101/2020.05.22.20110304
sha: 
doc_id: 321340
cord_uid: hwds5rja

Background The emergence of a novel coronavirus (SARS-CoV-2) in Wuhan, China, at the end of 2019 has caused widespread transmission around the world. As new epicentres in Europe and America have arisen, of particular concern is the increased number of imported coronavirus disease 2019 (COVID-19) cases in Africa, where the impact of the pandemic could be more severe. We aim to estimate the number of COVID-19 cases imported from 12 major epicentres in Europe and America to each African country, as well as the probability of reaching 10,000 infections in total by the end of March, April, and May following viral introduction. Methods We used the reported number of cases imported from the 12 major epicentres in Europe and America to Singapore, as well as flight data, to estimate the number of imported cases in each African country. Under the assumption that Singapore has detected all the imported cases, the estimates for Africa were thus conservative. We then propagated the uncertainty in the imported case count estimates to simulate the onward spread of the virus, until 10,000 infections are reached or the end of May, whichever is earlier. Specifically, 1,000 simulations were run separately under two scenarios, where the reproduction number under the stay-at-home order was assumed to be 1.5 and 1.0 respectively. Findings We estimated Morocco, Algeria, South Africa, Egypt, Tunisia, and Nigeria as having the largest number of COVID-19 cases imported from the 12 major epicentres. Based on our 1,000 simulation runs, Morocco and Algeria's estimated probability of reaching 10,000 infections by end of March was close to 100% under both scenarios. In particular, we identified countries with less than 100 cases in total reported by end of April whilst the estimated probability of reaching 10,000 infections by then was higher than 50% even under the more optimistic scenario. Conclusion Our study highlights particular countries that are likely to reach (or have reached) 10,000 infections far earlier than the reported data suggest, calling for the prioritization of resources to mitigate the further spread of the epidemic.

Abstract 10 11

Background 12

The emergence of a novel coronavirus in Wuhan, China, at the end of 2019 has caused 13 widespread transmission around the world. As new epicentres in Europe and America have arisen, of 14 particular concern is the increased number of imported coronavirus disease 2019 (COVID-19) cases 15

in Africa, where the impact of the pandemic could be more severe. We aim to estimate the number 16 of COVID-19 cases imported from 12 major epicentres in Europe and America to each African 17 country, as well as the probability of reaching 10,000 infections in total by the end of March, April, 18 and May following viral introduction. 19 20

We used the reported number of cases imported from the 12 major epicentres in Europe and 22 America to Singapore, as well as flight data, to estimate the number of imported cases in each 23

African country. Under the assumption that Singapore has detected all the imported cases, the 24 estimates for Africa were thus conservative. We then propagated the uncertainty in the imported 25 case count estimates to simulate the onward spread of the virus, until 10,000 infections are reached 26 or the end of May, whichever is earlier. Specifically, 1,000 simulations were run separately under 27 two scenarios, where the reproduction number under the stay-at-home order was assumed to be 28 1.5 and 1.0 respectively. 29 30 Findings 31

We estimated Morocco, Algeria, South Africa, Egypt, Tunisia, and Nigeria as having the largest 32 number of COVID-19 cases imported from the 12 major epicentres. Based on our 1,000 simulation 33 runs, Morocco and Algeria's estimated probability of reaching 10,000 infections by end of March was 34 close to 100% under both scenarios. In particular, we identified countries with less than 100 cases in 35 total reported by end of April whilst the estimated probability of reaching 10,000 infections by then 36 was higher than 50% even under the more optimistic scenario. 37 38 Conclusion 39

Our study highlights particular countries that are likely to reach (or have reached) 10,000 infections 40 far earlier than the reported data suggest, calling for the prioritization of resources to mitigate the 41 further spread of the epidemic.

3 Background 46 47

In late December 2019, a novel coronavirus (SARS-CoV-2) was identified among patients presenting 48 with viral pneumonia in Wuhan city, China 1 . Since then the number of coronavirus disease 2019 49 cases and deaths increased rapidly 2,3 , and the city was locked down by the Chinese 50 government on 23 rd January 2020. By late February, there had only been limited importations from 51 and to places outside China 4 . However, new epicentres in Europe and America emerged shortly 52 thereafter, causing a second wave of importations that further accelerated the spread of the 53 pandemic 4 . Most countries have since then imposed travel restrictions to prevent further 54 importation of COVID-19 cases 5 . By 30 th April 2020, over three million cases and 200,000 deaths had 55 been confirmed worldwide 4 . 56

A particular area of focus has been on countries in Africa, with worries about missed imported cases 57 and what the impact will be of widespread transmission given the other heavy health burdens in 58 these countries. The first confirmed case in Africa was reported in Egypt on 14 th February 2020, and 59 two weeks later, the virus was found in sub-Saharan Africa with a reported case in Nigeria 4 . By the 60 end of April, over 37,000 cases had been reported in the whole of Africa, with substantial variation in 61 the reported cumulative incidence across different countries 4 . This inter-country heterogeneity can 62 be due to a wide range of factors, such as the number of imported infections, the capacity to 63 conduct tests for COVID-19, surveillance efforts, as well as travel and movement restrictions which 64 vary widely from country to country depending on the local context 5 . The reported data alone thus 65 do not provide a clear depiction of the outbreak situation especially in countries with very limited 66 surveillance capacities, and additional studies are needed to narrow the knowledge gap between the 67 reported data and the real disease burdens. 68

Previous work has estimated the risk of importation from China at the early stage of the pandemic 6 , 69 assessed each African country's capacity to respond to outbreaks 6 , systematically collated 70 information on the importation events reported by the sub-Saharan countries 7 , and projected the 71 spread of the epidemic seeded by the early cases represented in the World Health Organization 72 Situation Reports 8 . It is still unclear how many infections may have been introduced to Africa from 73 the new epicentres in Europe and America, although the reported case data do suggest that the size 74 of this second wave of importations has been much larger than the first wave of importations from 75

China 7 . In this study, we aim to estimate the number of COVID-19 cases imported from the major 76 epicentres in Europe and America, and the magnitude of onward spread in each African country. 77

This method is insensitive to the different testing and reporting systems that are in place in different 78 countries. 79 . CC-BY-NC-ND 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)

The copyright holder for this preprint this version posted May 24, 2020 . . https://doi.org/10.1101 Methods 80 81 Data 82

We collated data on the daily number of imported cases in Singapore reported by 31 st March from 84 the following 12 epicentres: Austria, Belgium, France, Germany, Italy, Netherlands, Portugal, Spain, 85 Switzerland, Turkey, United Kingdom, and United States, which accounted for over 90% of 86

Singapore's reported number of imported cases from countries outside of Asia 9 . These data will be 87 used later to estimate the number of imported cases in Africa. In addition, we obtained the total 88 number of cases (imported and autochthonous combined) reported by each African country by end 89 of March and April from the World Health Organization's situation reports 4 . 90

For each country, we collated the date on which each of the following policies came into force: (1) 92 banning non-citizens and non-residents from entry (the start date could vary depending on the 93 epicentre country from which a visitor arrived); (2) mandatory (self-) quarantine for travellers 94 arriving from each of the 12 epicentre countries mentioned earlier; (3) Stay-at-home order for all 95 non-essential workers (hereinafter referred to as "stay-at-home order"). We reviewed the following 96 sources: (1) country-level internal and international restrictions collated by the International SOS 5 , 97

(2) Oxford COVID-19 Government Response Tracker 10 , (3) international travel restrictions collated by 98 the International Air Transport Association 11 , as well as (4) Wikipedia, where a separate page was 99 available for each country containing information regarding the government response. For each 100

Wikipedia page, we manually reviewed the online reports listed in the references to exclude data 101 with unconfirmed or unreliable sources. If stay-at-home order came into force in different states of 102 the same country at different times, only the earliest date was recorded. 103

We obtained the total number of air ticket bookings for each origin-destination route allowing for up 105 to two connections during March 2017 from the Official Airline Guide. This will be used later to 106 estimate the ratio of air passenger volumes between pairs of origin and destination countries, which 107 we assumed to be relatively stable over time. 108 109

Estimating the number of imported cases 111 . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020 . . https://doi.org/10.1101 For each African country , we denote the daily number of air passengers that arrived from an 112 epicentre country by → ( ) ( = , + 1, … , → ), where refers to the start date of the 113 COVID-19 epidemic in the epicentre country , and → refers to the last day that non-citizens and 114 non-residents travelling from country were allowed to enter country . Each day the probability 115 that an air passenger travelling from country to country was an imported case is denoted by ( ) , 116 which we assume to be dependent on both the origin country and time , but independent from 117 the destination country . Hence, the total number of COVID-19 cases imported from an epicentre 118 country to an African country by the time the travel ban came into force (denoted by → 119 below) can be approximated using a Poisson distribution (Refer to the supporting information for 120 the derivation details): 121

. 122

We used the imported COVID-19 case data reported by Singapore as well as flight data to provide a 123 conservative estimate for → , under the assumption that Singapore, being one of the countries 124 with the highest surveillance capacity 12 , has detected all the imported cases. Owing to the delay 125 from infection to hospital admission, we considered all cases imported from country to Singapore 126 that were reported by date ( → + 9) (hereinafter denoted as , ) based on Linton et al.'s 127 estimated mean incubation period and time from illness onset to hospital admission 13 . We assumed 128 that the ratio between the daily number of air travellers from epicentre to country and to 129 Here, , refers to the proportionality constant to be estimated using the reported value of , 137 and flight data, and was assigned a uniform prior with support (0, 1). We performed Markov Chain 138

Monte Carlo to sample from the posterior distribution of , using the JAGS software 14 , with 20,000 139 iterations burn-in and 150,000 iterations thinned for a posterior sample of size 5,000. The posterior 140 sample for all the model parameters was then used to estimate the uncertainty distribution of the 141 total number of COVID-19 cases imported from the 12 major epicentres to each country. 142 . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10. 1101 In March 2020, a spike in the number of cases imported from United Kingdom and United States was 143 observed in Singapore, which was partly due to the increase in the number of returning Singaporean 144 students studying overseas 15 . This change in flight patterns, however, may not be applicable to all 145 African countries. Therefore, to be even more conservative, we also derived the imported case count 146 estimates excluding United Kingdom and United States from the 12 epicentre countries previously 147 considered. The resulting estimates were subsequently used in the simulations of the onward spread 148 of SARS-CoV-2 to get our estimates of case numbers over time. stay-at-home order, was assumed to follow a negative binomial distribution with mean 2 and 165 dispersion parameter 0.58 8 . Once the stay-at-home order came into force, we created two scenarios 166 for the percentage reduction of the reproduction number: (1) 25% reduction, and (2) 50% reduction. 167

To be conservative, we assumed that the stay-at-home order, once implemented, can be sustained 168 up to the end date of our simulations. We ran the simulation algorithm following Churcher et al. 18 , 169 and derived the estimated probability of reaching 10,000 infections by the end of March, April, and 170

May respectively for each country. (Refer to the supporting information for the implementation 171 details) 172

. CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. We estimated Morocco, Algeria, South Africa, Egypt, Tunisia, and Nigeria as having the largest 175 number of COVID-19 cases imported from the 12 new epicentres in Europe and America (Table 1  176 and Figure 1 ). All of these countries had their lower bound estimate of the imported case count 177 exceeding 100 (Table 1) . By contrast, nine countries (e.g. Lesotho, Eswatini, and South Sudan) were 178 found to have a very low risk of importation, with the upper bound estimate of the imported case 179 count below 10 (Table 1 ). In a more conservative scenario where United Kingdom and United States 180 were excluded from the list of epicentre countries, the estimated number of imported cases did not 181 change drastically for most countries, albeit with some exceptions such as Kenya, whose estimate 182 decreased from 97 (95% CI: 75-120) to 27 (95% CI: 16-41) ( Table 1) . 183

Based on our 1,000 simulations of the onward SARS-CoV-2 spread, both Morocco and Algeria's 184 estimated probability of reaching 10,000 infections by end of March was close to 100% under both 185 scenarios that we considered (Figures 2A, 2D) , whilst the reported total number of cases in each 186 country by end of March was ~500 ( Figure 2G ). Under the assumption that stay-at-home order 187 reduces the reproduction number to 1.5, we found four African countries where the estimated 188 probability of reaching 10,000 infections by end of March was higher than 50% ( Figure 2A ). This 189 number quickly rose to 34 countries reaching this number of infections by the end of April, and 47 190 countries by end May (Figures 2B, 2C) . For the alternative scenario where the reproduction number 191 is reduced to 1.0 by stay-at-home order, the numbers of African countries with a higher-than-50% 192 estimated probability of reaching 10,000 infections by end of March, April, and May were 3, 23, and 193 32 respectively (Figures 2D-2F) . Notably, four countries (Angola, Gambia, Mozambique, and Sao 194 Tome and Principe) were found to have reported less than 100 cases by end of April whilst the 195 estimated probability of reaching 10,000 infections by then was higher than 50% even under the 196 more optimistic scenario ( Figures 2E, 2H ), suggesting that a very substantial number of cases may 197 have been undetected. The percentiles of the uncertainty distribution for the date by which 10,000 198 infections are reached in each country under the two scenarios were shown in Table 2 . 199 . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. In the first wave of importations of infections from Wuhan, China, to other places outside China we 207 estimated that most places at risk were in Asia, Europe and USA 19 . Though there were links between 208

China and African countries, these were fewer than those between China and the rest of Asia, 209

Europe and USA 19 . The shut down in China severely curtailed continuing importations out of China 210 and so these importations rapidly stopped. 211

Lower initial importations into Africa compared to Asia and Europe certainly tallies with what has 212 been seen. There have been very few reported cases in Africa in the first wave of importations, and 213 no reports of onward transmission. There was much discussion at the time whether the lack of 214 reported imported cases in Africa was because imported cases were not being picked up. This may 215 be some of the story, but our analysis would suggest that this was not the whole story, and it was 216 more that the early risk of importation into Africa was lower than other places 19 . However the 217 results we present in this paper estimate that this risk has dramatically increased with the spread of 218 the virus in Europe and the USA. This also tallies with what we have seen, as countries in Africa 219 started to report their first imported cases from Europe and the USA 4 . As of April 30 th 2020, South 220

Africa had reported the highest number of cases at 5350 4 , and we estimated South Africa to have 221 had one of the highest numbers of imported cases from the new epi-centres, although it was also 222 rated highest at risk in Africa of importations from China in previous analysis 6 . Senegal is one of the 223 countries for whom the risk has notably increased from the risk of importation from China as 224 estimated in previous analyses 6, 19 . We only considered importations from the major epicentres in 225 Europe and America, and so the number of importations from all countries will be even higher. 226

Our study provides countries with information on the estimated timing of reaching 10,000 227 infections, which can be used for planning. Under the assumption that stay-at-home order reduces 228 the reproduction number from 2.0 to 1.5, our estimates suggest that a number of African countries 229 will reach (or have reached) 10,000 infections even earlier than the predictions of Pearson et al. 8 230 This could be due to a number of imported infections being undetected and hence not reflected in 231 the situation reports, as well as the delay from infection to reporting, both of which were accounted 232 for in our study. Notably, we estimated two countries in North Africa, namely Algeria and Morocco, 233 as having the highest probabilities of reaching 10,000 infections by the end of March, which may 234 . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10.1101/2020.05.22.20110304 doi: medRxiv preprint have occurred even prior to the lockdown. Countries in Sub-Saharan Africa having the earliest 235 estimated timings of reaching 10,000 infections include Angola, Côte d'Ivoire, Senegal, and South 236

Africa. In countries where stringent social distancing measures have yet to be implemented at the 237 time of writing (e.g. Tanzania), the unfolding of the epidemic was estimated to be substantially 238 faster than previous estimates suggest 8 . On the other hand, we projected that countries such as 239 Seychelles will reach 10,000 infections later than Pearson et al.'s forecasts 8 owing to the stay-at-240 home order. The epidemic was found to be further slowed down in many countries when we 241 assumed the reproduction number to be reduced by 50% due to stay-at-home order. 242

Many countries in Africa have considerable experience in dealing with other infectious disease 243 outbreaks, most notably Ebola, and will be able to call upon that experience for COVID-19. Countries 244 hit in this third wave of transmission, including those in Africa have some advantage as there have 245 been a variety of responses from around the world from which to assess what to do or not to do. 246

However there will need to be consideration of how effective measures can be adapted to different 247 settings 20 . Issues such as high HIV prevalence in some countries, and a younger demographic may 248 both affect the cases and deaths observed in different ways. This relationship however is yet to be 249 determined and there will need to be rapid research in countries in Africa to determine what the risk 250 of disease is in different populations and how best to respond in light of many other competing 251 health priorities. 252

Many countries in Africa are on high alert for incoming cases from Europe and USA, taking measures 253 such as quarantine of arrivals or shutting down travel from affected countries. This is a sensible 254 response given the vast amount of transmission on-going in these places. However as travel is either 255 maintained or reopened between countries closer by, risk of importations from other countries 256 should continue to be considered. Close attention should therefore be paid to where will be the next 257 epicentre, perhaps within Africa, and how this could translate into imported cases for each country, 258 particularly for those countries that we estimate to have experienced lower numbers of imported 259 cases previously and therefore lower onward transmission. 260

Not accounted for in our study currently is the impact of less stringent interventions on the local 261 SARS-CoV-2 spread, such as the effect of prohibiting large public gatherings, closure of social venues 262 and schools, and restrictions on inter-district travels. It is still unclear as to whether and to what 263 extent these interventions were effective in their local context, and hence in our simulations we only 264 considered stay-at-home order for all non-essential workers as an effective intervention to reduce 265 local transmission. Future modelling work considering the impact of different interventions in 266 different places will be vital for determining how each country can continue to respond. 267

In addition, we have made simplifying assumptions about the change in travel patterns in response 268 to the pandemic in each African country relative to that in Singapore, due to the unavailability of 269 . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. . CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020 . . https://doi.org/10.1101 respectively. Reproduction number in the absence of stay-at-home order in each country was 398 assumed to be 2. Reported total number of cases (G-H) were extracted from the World Health 399

Organization's situation reports. 400

. CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020 . . https://doi.org/10.1101 Table 1: Estimated number of COVID-19 cases (with 95% credible interval) imported from the 12 new epicentres in Europe and America (second column), and after excluding United Kingdom and United States from the list of epicentre countries (third column) to create a more conservative estimate (refer to Methods for more details).

. CC-BY-NC-ND 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.

The copyright holder for this preprint this version posted May 24, 2020. Table 2 : Summary statistics for the estimated date by which 10,000 infections are reached in each African country. Reproduction numbers used for the simulation were 2.0 before, and 1.5 or 1.0 after stay-at-home order came into force in each country. Simulations were performed until 31 st May, or 10,000 infections are reached, whichever is earlier, based on 1,000 model runs. . CC-BY-NC-ND 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)

The copyright holder for this preprint this version posted May 24, 2020. . https://doi.org/10. 1101 

Genomic characterisation and epidemiology of 325 2019 novel coronavirus: implications for virus origins and receptor binding

of Novel Coronavirus-Infected Pneumonia

Clinical features of patients infected with 330 2019 novel coronavirus in Wuhan

5. International SOS. TRAVEL RESTRICTIONS, FLIGHT OPERATIONS AND SCREENING

Covid-19: S'porean students abroad heading home after government 367 advisory but a few plan to stay put

A New Framework and Software to Estimate 371

Time-Varying Reproduction Numbers During Epidemics

Serial interval of novel coronavirus (COVID-19) 375 infections

Measuring the 378 path toward malaria elimination

Estimating number of global importations 381 of COVID-19 from Wuhan, risk of transmission outside mainland China and COVID-19 382 introduction index between countries outside mainland China. medRxiv

Limiting the spread of 385 COVID-19 in Africa: one size mitigation strategies do not fit all countries

Estimated imported case count from 12 epicentres Estimated imported case count from 10 epicentres