key: cord-0711531-9yzb8wj3
authors: Rice, Benjamin L.; Annapragada, Akshaya; Baker, Rachel E.; Bruijning, Marjolein; Dotse-Gborgbortsi, Winfred; Mensah, Keitly; Miller, Ian F.; Motaze, Nkengafac Villyen; Raherinandrasana, Antso; Rajeev, Malavika; Rakotonirina, Julio; Ramiadantsoa, Tanjona; Rasambainarivo, Fidisoa; Yu, Weiyu; Grenfell, Bryan T.; Tatem, Andrew J.; Metcalf, C. Jessica E.
title: High variation expected in the pace and burden of SARS-CoV-2 outbreaks across sub-Saharan Africa
date: 2020-07-24
journal: medRxiv
DOI: 10.1101/2020.07.23.20161208
sha: cd3154bb17689cc6c911db968ac6f8778678a7e7
doc_id: 711531
cord_uid: 9yzb8wj3

A surprising feature of the SARS-CoV-2 pandemic to date is the low burdens reported in sub-Saharan Africa (SSA) countries relative to other global regions. Potential explanations (e.g., warmer environments(1), younger populations(2–4)) have yet to be framed within a comprehensive analysis accounting for factors that may offset the effects of climate and demography. Here, we synthesize factors hypothesized to shape the pace of this pandemic and its burden as it moves across SSA, encompassing demographic, comorbidity, climatic, healthcare and intervention capacity, and human mobility dimensions of risk. We find large scale diversity in probable drivers, such that outcomes are likely to be highly variable among SSA countries. While simulation shows that extensive climatic variation among SSA population centers has little effect on early outbreak trajectories, heterogeneity in connectivity is likely to play a large role in shaping the pace of viral spread. The prolonged, asynchronous outbreaks expected in weakly connected settings may result in extended stress to health systems. In addition, the observed variability in comorbidities and access to care will likely modulate the severity of infection: We show that even small shifts in the infection fatality ratio towards younger ages, which are likely in high risk settings, can eliminate the protective effect of younger populations. We highlight countries with elevated risk of ‘slow pace’, high burden outbreaks. Empirical data on the spatial extent of outbreaks within SSA countries, their patterns in severity over age, and the relationship between epidemic pace and health system disruptions are urgently needed to guide efforts to mitigate the high burden scenarios explored here.

The trajectory of the SARS-CoV-2 pandemic in lower latitude, lower income countries including 22

in Sub-Saharan Africa (SSA) remains uncertain. To date, reported case counts and mortality in 23 SSA have lagged behind other geographic regions: all SSA countries, with the exception of 24 South Africa, reported less than 27,000 total cases as of June 2020 5 (Table S1 ) -totals far less 25 than observed in Asia, Europe, and the Americas 5, 6 . However, recent increases in reported 26 cases in many SSA countries make it unclear whether the relatively few reported cases to date 27 indicate a reduced epidemic potential or rather an initial delay relative to other regions. 28 29 Correlation between surveillance capacity and case counts 7 obscure early trends in SSA 30 ( Figure S1 ). Experience from locations in which the pandemic has progressed more rapidly 31 provides a basis of knowledge to assess the relative risk of populations in SSA and identify 32 those at greatest risk. For example, individuals in lower socio-economic settings have been 33 disproportionately affected in high latitude countries, 8, 9 indicating poverty as an important 34 determinant of risk. Widespread disruptions to routine health services have been reported 10-12 35 and are likely to be an important contributor to the burden of the pandemic in SSA 13 . The role of 36 other factors from demography 2-4 to health system context 14 and intervention timing 15,16 is also 37 increasingly well-characterized. 38 39

Anticipating the trajectory of ongoing outbreaks in SSA requires considering variability in known 41 drivers, and how they may interact to increase or decrease risk across populations in SSA and 42 relative to non-SSA settings (Figure 1 ). For example, while most countries in SSA have 'young' 43 populations, suggesting a decreased burden (since SARS-CoV-2 morbidity and mortality 44 increase with age 2-4 ), prevalent infectious and non-communicable comorbidities may 45 counterbalance this demographic 'advantage' 14, [17] [18] [19] . Similarly, SSA countries have health 46 systems that vary greatly in their infrastructure, and dense, resource-limited urban populations 47 may have fewer options for social distancing 20 . Yet, decentralized, community-based health 48 systems that benefit from recent experience with epidemic response (e.g., to Ebola 21, 22 ) can be 49 mobilized. Climate is frequently invoked as a potential mitigating factor for warmer and wetter 50 settings 1 , including SSA, but climate varies greatly between population centers in SSA and 51 large susceptible populations may counteract any climate forcing during initial phases of the 52 epidemic 23 . Connectivity, at international and subnational scales, also varies greatly 24,25 and 53 the time interval between viral introductions and the onset of interventions such as lockdowns 54 will modulate the trajectory 7 . Finally, burdens of malnutrition, infectious diseases, and many 55 Rice et al | 2020 07 23 | Page 3 likely to interact with social contact rates among the elderly in determining exposure and clinical 90 outcomes (e.g., for variation in household size see Figure 2E -F). Relative ranking across 91 variables is also uneven among countries with the result that this diversity cannot be easily 92 reduced (e.g., the first two principal components explain only 32.6%, and 13.1% of the total 93 variance as shown in Figure S5 ), motivating a more holistic approach to projecting burden. 94

Severity of infection outcome 96

To first evaluate variation in the burden emerging from the severity of infection outcome, we 97 consider how demography, comorbidity, and access to care might modulate the age profile of 98 SARS-CoV-2 morbidity and mortality [2] [3] [4] . Subnational variation in the distribution of high risk age 99 groups indicates considerable variability, with higher burden expected in urban settings in SSA 100 ( Figure 3A) , where density and thus transmission are likely higher 27 . 101

Comorbidities and access to clinical care also vary across SSA (e.g., for diabetes prevalence 103 and hospital bed capacity see Figure 3B ). In comparison to settings where previous SARS-104

CoV-2 infection fatality ratio (IFR) estimates have been reported, mortality due to 105 noncommunicable diseases in SSA increases more rapidly with age ( Figure S6) . Consequently, 106 we explore scenarios where the SARS-CoV-2 IFR increases more rapidly with age than the 107 baseline expected from other settings. Small shifts (e.g., of 2-10 years) in the IFR profile result 108 in large effects on expected mortality for a given level of infection. For example, Chad, Burkina 109 Faso, and the Central African Republic, while among the youngest SSA countries, have a 110 relatively high prevalence of diabetes and relatively low density of hospital beds. A five year shift 111 younger in the IFR by age profile of SARS-CoV-2 in these settings would result in nearly a 112 doubling of mortality, to a rate that would exceed the majority of other, 'older' SSA countries at 113 the unshifted baseline ( Figure 3C , see supplement for details of methods). Although there is 114 greater access to care in older populations by some metrics (Figure 2A , correlation between 115 age and the number of physicians per capita, r = 0.896, p < 0.001), access to clinical care is 116 highly variable overall ( Figure 3D ) and maps poorly to indicators of comorbidity ( Figure 3E) . 117

Empirical data are urgently needed to assess the extent to which the IFR-age-comorbidity 118 associations observed elsewhere are applicable to SSA settings with reduced access to 119 advanced care. Yet both surveillance and mortality registration 28 are frequently under-120 resourced in SSA, complicating both evaluating and anticipating the burden of the pandemic, 121 and underscoring the urgency of strengthening existing systems 22 . 122 123 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Rice et al | 2020 07 23 | Page 4

Next, we turn to the pace of the pandemic within each country. The frequency of viral 125 introduction to each country, likely governed by international air travel in SSA 29 , determines 126 both the timing of the first infections and the number of initial infection clusters that can seed 127 subsequent outbreaks. The relative importation risk among SSA cities and countries was 128 assessed by compiling data from 108,894 flights arriving at 113 international airports in SSA 129 from January to April 2020 (Figure 4A) , stratified by the SARS-CoV-2 status at the departure 130 location on the day of travel (Figure 4B) . A small subset of SSA countries received a 131 disproportionately large percentage (e.g., South Africa, Ethiopia, Kenya, Nigeria together 132 contribute 47.9%) of the total travel from countries with confirmed SARS-CoV-2 infections, likely 133 contributing to variation in the pace of the pandemic across settings 29, 30 . 134

Once local chains of infection are established, the rate of spread within countries will be shaped 136 by efforts to reduce spread, such as handwashing ( Figure 2D ), population contact patterns 137 including mobility and urban crowding 27 (e.g., Figure 2C) , and potentially the effect of climatic 138 variation 1 . Where countries fall across this spectrum of pace will shape interactions with 139 lockdowns and determine the length and severity of disruptions to routine health system 140 functioning. 141 142 Subnational connectivity varies greatly across SSA, both between subregions of a country and 143 between cities and their rural periphery (e.g., as indicated by travel time to the nearest city over 144 50,000 population, Figure 4C ). As expected, in stochastic simulations using estimates of viral 145 transmission parameters and mobility (assuming no variation in control efforts, see methods), a 146 smaller cumulative proportion of the population is infected at a given time in countries with 147 larger populations in less connected subregions (Figure 4D) . At the national level, susceptibility 148 declines more slowly and more unevenly in such settings (e.g., Ethiopia, South Sudan, 149

Tanzania) due to a lower probability of introductions and re-introductions of the virus locally; an 150 effect amplified by lockdowns. It remains unclear whether the more prolonged, asynchronous 151 epidemics expected in these countries or the overlapping, concurrent epidemics expected in 152 countries with higher connectivity (e.g. Malawi, Kenya, Burundi) will be a greater stress to health 153 systems. Outbreak control efforts are likely to be further complicated during prolonged 154 epidemics if they intersect with seasonal events such as temporal patterns in human mobility 31 155 or other infections (e.g., malaria). 156 Turning to climate, despite extreme variation among cities in SSA (Figure 4E ), large epidemic 158 peaks are expected in all cities ( Figure 4F) , even from models where transmission rate 159 significantly declines in warmer, more humid settings. In the absence of interventions, with 160 transmission rate modified by climate only, peak timing varies only by 4-6 weeks with peaks 161 generally expected earlier in more southerly, colder, drier, cities (e.g., Windhoek and Maseru) 162 and later in more humid, coastal cities (e.g., Bissau, Lomé, and Lagos). Apart from these slight 163 shifts in timing, large susceptible populations overwhelm the effects of climate 23 , and earlier 164 suggestions that Africa's generally more tropical environment may provide a protective effect 1 165 are not supported by evidence. 166

167

Context-specific preparedness in SSA 168

Our synthesis emphasizes striking country to country variation in drivers of the pandemic in SSA 169 Others have suggested that this crisis presents an opportunity to unify and mobilize across 185 existing health programs (e.g., for HIV, TB, Malaria, and other NCDs) 22 . While this may be a 186 powerful strategy in the context of acute, temporally confined crises, long term distraction and 187 diversion of resources 32 may be harmful in settings with extended, asynchronous epidemics. A 188 higher risk of infection among healthcare workers during epidemics 33, 34 may amplify this risk. 189 190 . 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 July 24, 2020. Due to the lag relative to other geographic regions, many SSA settings retain the opportunity to 191 prepare for and intervene in the earlier epidemic phases via context-specific deployment of both 192 routine and pandemic related interventions. As evidenced by failures in locations where the 193 epidemic progressed rapidly (e.g., USA), effective governance and management prior to 194 reaching large case counts is likely to yield the largest rewards. Mauritius 35 and Rwanda 36 , for 195 example, have reported extremely low incidence thanks in part to a well-managed early 196 response. 197

Conclusions 199

The burden and time-course of SARS-CoV-2 is expected to be highly variable across sub-200 Saharan Africa. As the outbreak continues to unfold, critically evaluating this mapping to better 201 understand where countries lie in terms of their relative risk (e.g., see Figure 5 ) will require 202 increased surveillance, and timely documentation of morbidity and mortality over age. Case 203 counts are rising across SSA, but variability in testing regimes makes it difficult to compare 204 observations to date with expectations in terms of pace ( Figure S7) . The potential to miss large 205 clusters of cases (in contexts with weaker surveillance), combined with the potential that large 206 areas remain unreached by the pandemic for longer (as a result of slower 'pace'), indicate that 207 immunological surveys are likely a powerful lens for understanding the landscape of population 208 risk 37 . When considering hopeful futures with the possibility of a SARS-CoV-2 vaccine, it is 209 imperative that vaccine distribution be equitable, and in proportion with need. Understanding 210 factors that both drive spatial variation in vulnerable populations and temporal variation in 211 pandemic progression could help approach these goals in SSA. 212 . 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 July 24, 2020. .

Factors hypothesized to increase (red) or decrease (blue) mortality burden or epidemic pace within sub-Saharan Africa, relative to global averages, are grouped in six categories or dimensions of risk (A-F). In this framework, epidemic pace is determined by person to person transmissibility (which can be defined as the time-varying effective reproductive number, R t ) and introduction and geographic spread of the virus via human mobility.

SARS-CoV-2 mortality (determined by the infection fatality ratio, IFR) is modulated by demography, comorbidities (e.g., non-communicable diseases (NCDs)), and access to care. Overall burden is a function of direct burden and indirect effects due to, for example, disruptions in health services such as vaccination and infectious disease control. Table S2 contains details and the references used as a basis to draw the hypothesized modulating pathways. 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 July 24, 2020. .

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 July 24, 2020.

. https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint A: Expected mortality in a scenario where cumulative infection reaches 20% across age groups and the infection fatality ratio (IFR) curve is fit to existing age-stratified IFR estimates (see methods, Table S4 ). B: National level variation in comorbidity and access to care variables, for e.g., diabetes prevalence among adults and the number of hospital beds per 100,000 population for sub-Saharan African countries. C: The range in mortality per 100,000 population expected in scenarios where cumulative infection rate is 20% and IFR per age is the baseline (black) or shifted 2, 5, or 10 years younger (gray). Inset, the IFR by age curves for each scenario.

D-E: Select national level indicators; estimates of reduced access to care (e.g., fewer hospitals) or increased comorbidity burden (e.g., higher prevalence of raised blood pressure) shown with darker red for higher risk quartiles (see Figure S4 for all indicators). Countries missing data for an indicator (NA) are shown in gray. For comparison between countries, estimates are agestandardized where applicable (see Table S3 for details). See the [SSA-SARS-CoV-2-tool] for high resolution maps for each variable and scenario. 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 July 24, 2020. Table S5 for all others). C: Connectivity within SSA countries as inferred from average population weighted mean travel time to the nearest urban area greater than 50,000 population. .

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 July 24, 2020. 

Countries are colored by with respect to indicators of their expected epidemic pace (using as an example subnational connectivity in terms of travel time to nearest city) and potential burden (using as an example the proportion of the population over age 50). A: In pink, countries with less connectivity (i.e., less synchronous outbreaks) relative to the median among SSA countries; in blue, countries with more connectivity; darker colors show countries with older populations (i.e., a greater proportion in higher risk age groups).

B: Dotted lines show the median; in the upper right, in dark pink, countries are highlighted due to their increased potential risk for an outbreak to be prolonged (see metapopulation model methods) and high burden (see burden estimation methods).

.

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 July 24, 2020. Data and materials availability 236 All materials are available in the online content 237 238

Competing interests 239 The authors declare no competing interests 240 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Rice et al | 2020 07 23

A1 Reported SARS-CoV-2 case counts, mortality, and testing in sub-Saharan Africa as of June 2020 Table S1 : Sub-Saharan Africa country codes, case counts, and testing Figure S1 : Variation between SSA countries in testing and reporting rates A2 Synthesizing factors hypothesized to increase or decrease SARS-CoV-2 epidemic risk in SSA Table S2 : Dimensions of risk and expected direction of effect on SARS-CoV-2 transmission or burden in sub-Saharan Africa (SSA) relative to higher latitude countries 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint The numbers of reported cases, deaths, and tests for the 48 studied sub-Saharan Africa (SSA) 6 countries (Table S1) were sourced from the Africa Centers for Disease Control (CDC) 7 dashboard on June 30, 2020 (https://africacdc.org/covid-19/). Africa CDC obtains data from the 8 official Africa CDC Regional Collaborating Centre and member state reports. Differences in the 9 timing of reporting by member states results in some variation in recency of data within the 10 centralized Africa CDC repository, but the data should broadly reflect the relative scale of testing 11

and reporting efforts across countries. 12 13

The countries or member states within SSA in this study follow the United Nations Testing rates among SSA countries varied by multiple orders of magnitude: the number of tests 32 completed per 100,000 population ranged from 6.50 in Tanzania to 13,508.13 in Mauritius 33

( Figure S1A ). The number of reported infections (i.e., positive tests) was strongly correlated 34 with the number of tests completed (Pearson's correlation coefficient, r = 0.9667, p < 0.001) 35 ( Figure S1B) . As of June 30, 2020, no deaths due to SARS-CoV-2 were reported to the Africa 36 CDC for five SSA countries (Eritrea, Lesotho, Namibia, Seychelles, Uganda). Among countries 37

with at least one reported death, CFR varied from 0.22% in Rwanda to 8.54% in Chad ( Figure  38 S1C). Limitations in the ascertainment of infection rates and the rarity of reported deaths (e.g., 39

median number of reported deaths per SSA country was 25.5), indicate that the data are 40 insufficient to determine country specific IFRs and IFR by age profiles. As a result, global IFR by 41 age estimates were used for the subsequent analyses in this study. 42 . 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 July 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 July 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint 

. 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint 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 July 24, 2020. To characterize epidemic risk, defined as potential SARS-CoV-2 related morbidity and mortality, 70 we first synthesized factors hypothesized to influence risk in SSA settings (Table S2) . Early 71 during the pandemic, evidence suggested that age was an important risk factor associated with 72 morbidity and mortality associated with SARS-CoV-2 infection 39 , a pattern subsequently 73 confirmed across settings 2, 9, 40 . Associations between SARS-CoV-2 mortality and comorbidities 74

including hypertension, diabetes, and cardiovascular disease emerged early 39 ; and have been 75 observed across settings, with further growing evidence for associations with obesity 9,41 , severe 76 asthma 9 , and respiratory effects of pollution 42 . 77 78

Many possible sources of bias complicate interpretation of these associations 43 , and while they 79 provide a useful baseline, inference is also likely to change as the pandemic advances. To 80 reflect this, our analysis combines a number of high level variables likely to broadly encompass 81 these putative risk factors (e.g., non-communicable disease (NCD) related mortality and health 82 life expectancy) with more specific measures encompassed in evidence to date (e.g., 83

prevalence of diabetes, obesity, and respiratory illness such as Chronic Obstructive Pulmonary 84

Disease (COPD)). We also include measures relating to infectious diseases, undernourishment, 85 and anemia given their interaction and effects in determining health status in these settings Comparisons of national level estimates sourced from WHO and other sources are affected by 98 variation within countries and variation in the uncertainty around estimates from different 99 geographical areas. To assess potential differences in data quality between geographic areas 100

we compared the year of most recent data for variables ( Figure S2) . available for a single variable, we also include multiple variables per category (e.g., 104

demographic and socio-economic factors, comorbidities, access to care) to avoid reliance on a 105 single metric. This allows exploring variation between countries across a broad suite of 106 variables likely to be indicative of the different dimensions of risk. 107 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Although including multiple variables that are likely to be correlated (see PCA methods below 109 for further discussion) would bias inference of cumulative risk in a statistical framework, we do 110 not attempt to quantitatively combine risk across variables for a country, nor project risk based 111 on the variables included here. Rather, we characterize the magnitude of variation among 112 countries for these variables (see Figure 2 in the main text for a subset of the variables; Figure  113 3B for bivariate risk maps following 46 ) and then explore the range of outcomes that would be 114 expected under scenarios where IFR increases with age at different rates (see Figure 3 in the 115 main text include climatic factors (e.g., specific humidity), access to prevention measures (e.g., 124

handwashing), and human mobility (e.g., international and domestic travel). International flight data was obtained from a custom report from OAG Aviation Worldwide (UK) 132

and included the departure location, airport of arrival, date of travel, and number of passenger 133 seats for flights arriving to 113 international airports in SSA (see Section A5).

As an estimate of connectivity within subregions of countries, the population weighted mean 136 travel time to the nearest city with a population greater than 50,000 was determined; details are 137 provided in Section A6. To obtain a set of measures that broadly represent connectivity within 138 different countries in the region, friction surfaces from ref 24 were used to obtain estimates of the 139 connectivity between different administrative level 2 units within each country. Details of this, 140 alongside the metapopulation model framework used to simulate viral spread with variation in 141 connectivity are in Section A6. 142 143 Figure 2 in the main text shows variation among SSA countries for four of the variables; Figure  144 S3 shows variation for all variables. Figure 3 in the main text shows variation for a subset of the 145 comorbidity and access to care indicators as a heatmap; Figure S4 shows variation for all the 146 variables (both also available online at the [SSA-SARS-CoV-2-tool]). 147 148 . 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 July 24, 2020. Table S2 149

Hypothesized 

Lower rates of some comorbidities that have been associated with risk of worse outcomes, e.g., obesity 9,41

Higher rates of NCDs such as hypertension or COPD 39 , which are associated with worse outcomes; and a potential role for as yet undescribed interactions e.g., with anemia, or high prevalence infectious diseases Decreased rate of internal spread due to less connectivity within countries 56 152 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Rice et al | 2020 07 23 | Page 8 Table S3 153

Africa 155 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint (Table S3 continued . 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 July 24, 2020. 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 July 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Rice et al | 2020 07 23 | Page 12

168

Dotted vertical line shows regional median; solid vertical line shows regional mean. Note that most data comes from 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint 

A: Select national level indicators; estimates of increased comorbidity burden (e.g., higher prevalence of raised blood 187 pressure) shown with darker red for higher risk quartiles Countries missing data for an indicator (NA) are shown in 188 gray. For comparison between countries, estimates are age-standardized where applicable (see Table S3 for details) 189 190 191 . 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 July 24, 2020. Table S3 for details) 194 195 . 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 July 24, 2020. In some cases, data were missing for a country for an indicator; in these cases, missing data 213

were replaced with a zero value. This is a conservative approach as zero values (i.e., outside 214 the range of typical values seen in the data) inflate the total variance in the data set and thus, if 215

anything, deflate the percent of the variance explained by PCA. Therefore, this approach avoids 216 mistakenly attributing predictive value to principal components due to incomplete data. See 217 Table S3 for data sources for each variable. 218 219

3.2 Principal Component Analysis 220 221

The PCA was conducted on each of the three subsets described above, using the scikitlearn 222 library 64 . In order to avoid biasing the PCA due to large differences in magnitude and scale, 223 each feature was centered around the mean, and scaled to unit variance prior to the analysis. 224

Briefly, PCA applies a linear transformation to a set of n features to output a set of n orthogonal 225 principal components which are uncorrelated and each explain a percentage of the total 226 variance in the dataset 65 . A link to the code for this analysis is available online at the shiny app 227

[SSA-SARS-CoV-2-tool]. 228 229

The principal components were then analyzed for the percentage of variance explained, and 230 compared to: (i) the number of COVID-19 tests per 100,000 population as of the end of June, 231 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint The first two principal components from the analysis of 29 variables explain 32.6%, and 13.1% 241 the total variance, respectively, in the dataset. Countries with higher numbers of completed 242 SARS-CoV-2 tests reported tended to associate with an increase in principal component 1 243

(Pearson correlation coefficient, r = 0.67, p = 1.1e-7, Figure S5A) . Similarly, high GDP 244 countries seem to associate with an increase in principal component 1 (Pearson correlation  245 coefficient, r = 0.80, p = 6.02e-12), Figure S5B ). In contrast, countries with greater wealth 246 inequality (as measured by the GINI index) are associated with a decrease in principal 247 component 2 (Pearson correlation coefficient, r = -0.42, p = .0042, Figure S5C ). Despite these 248 correlations, a relatively low percentage of variance is explained by each principal component: 249

for the 29 variables, 13 of the 29 principal components are required to explain 90% of the 250 variance ( Figure S5D) . When only the access to care subset of variables is considered, the first 251 two principal components explain 50.7% and 19.1% of the variance, respectively, and five of 252 eight principal components are required to explain 90% of the variance. When only the 253 comorbidities subset is considered, the first two principal components explain 27.9% and 17.8% 254 of the variance, respectively, and nine of 14 principal components are required to explain 90% 255 of the variance (Figure S4D ).

3.4 PCA Discussion 258 259

These data suggest that inter-country variation in this dataset is not easily explained by a small 260 number of variables. Moreover, though correlations exist between principal components and 261

high-level explanatory variables (testing capacity, wealth), their magnitude is modest. These 262 results highlight that dimensionality reduction is unlikely to be an effective analysis strategy for 263 the variables considered in this study. Despite this overall finding, the PCA on the access to 264 care subset of variables highlights that the variance in these variables is more easily explained 265 by a small number of principal components, and hence may be more amenable to 266 dimensionality reduction. This finding is unsurprising as, for example, the number of hospital 267 beds per 100,000 population is likely to be directly related to the number of hospitals per 268 100,000 population (indeed r = 0.60, p = 5.7e-6 for SSA). In contrast, for comorbidities, the 269 relationship between different variables is less clear. Given the low percentages of variation 270 captured by each principal component, and the high variability between different types of 271 variables, these results motivate a holistic approach to using these data for assessing relative 272 SARS-CoV-2 risk across SSA. 273 274 275 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Table S3 )), and 300 access to care indicators (orange, 14 variables, Section E in Table S3) 301 302 303 . 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 July 24, 2020. Estimates of the infection fatality ratio (IFR) that account for asymptomatic cases, 309

underreporting, and delays in reporting are few, however, it is evident that IFR increases 310 substantially with age 66 . We use age-stratified estimates of IFR from three studies (two 311 published 2,4 , one preprint 3 ) that accounted for these factors in their estimation ( We assume a given level of cumulative infection (here 20% in each age class, i.e., a constant 323 rate of infection among age classes) and then apply IFRs by age to the population structure of 324 each country to generate estimates of burden. Age structure estimates were taken from the 325 UNPOP (see Table S3 ) country level estimates of population in 1 year age groups (0 -100 326 years of age) to generate estimates of burden. 327 328 329 330 . 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 July 24, 2020. Applying these IFR estimates to the demographic structure of SSA countries provides a 333 baseline expectation for mortality, but depends on the assumption that mortality patterns in sub-334

Saharan Africa will be similar to those from where the IFR estimates were sourced (France, 335

China, and Italy). Comorbidities have been shown to be an important determinant of the severity 336 of infection outcomes (i.e., IFR); to assess the relative risk of comorbidities across age in SSA, 337 estimates of comorbidity severity by age (in terms of annual deaths attributable) were obtained 338 from the Institute for Health Metrics and Evaluation (IHME) Global Burden of Disease (GBD) 339 study in 2017 69 . Data were accessed through the GBD results tool for cardiovascular disease, 340 chronic respiratory disease (not including asthma), and diabetes, reflecting three categories of 341 comorbidity with demonstrated associations with risk (Table S2) . We make the assumption that 342 higher mortality rates due to these NCDs, especially among younger age groups, is indicative of 343 increased severity and lesser access to sufficient care for these diseases -suggesting an 344 elevated risk for their interaction with SARS-CoV-2 as comorbidities. While there are significant 345 uncertainties in these data, they provide the best estimates of age specific risks and have been 346 used previously to estimate populations at risk 18 . 347 348

The comorbidity by age curves for SSA countries were compared to those for the three 349 countries from which SARS-CoV-2 IFR by age estimates were sourced. Attributable mortality 350 due to all three NCD categories is higher at age 50 in all 48 SSA countries when compared to 351 estimates from France and Italy and for 42 of 48 SSA countries when compared to China 352 ( Figure S5 ).

Given the potential for populations in SSA to experience a differing burden of SARS-CoV-2 due 355

to their increased severity of comorbidities in younger age groups, we explore the effects of 356

shifting IFRs estimated by the GAM of IFR estimates from France, Italy, and China younger by 357 2, 5, and 10 years (Figure 3 in main text). 358 359 360 . 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 July 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint

The number of passenger seats on flights arriving to international airports were grouped by 370 country and month for January 2020 to April 2020 ( The number of travelers within each category arriving per month is shown in Table S5 . This 406 approach makes the conservative assumption that the probability a traveler is infected reflects 407 the general countrywide infection rate of the source country at the time of travel (i.e., travelers 408

are not more likely to be exposed than non-travelers in that source location) and does not 409 account for complex travel itineraries (i.e., a traveler from a high risk source location transiting 410 . 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 July 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 July 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 July 24, 2020. We make the simplifying assumption that mobility linking locations i and j, denoted 1 ',! ,scales 488

with the inverse of the cost of travel between sites i and j evaluated according to the friction 489

surface provided in 24 . The introduction of an infected individual into location j is then defined by 490 a draw from a Bernouilli distribution following: 491

where L is the total number of administrative 2 units in that country, and the rate of introduction 493

is the product of connectivity between the focal location and each other location multiplied by 494 the proportion of population in each other location that is infected. 495 496

Some countries show rapid spread between administrative units within the country (e.g., a 497 country with parameters that broadly reflect those available for Malawi, Figure S7 ), while in 498 others (e.g., reflecting Madagascar), connectivity may be so low that the outbreak may be over 499

in the administrative unit of the largest size (where it was introduced) before introductions 500 successfully reach other poorly connected administrative units. The result is a hump shaped 501 relationship between the fraction of the population that is infected after 5 years and the time to 502 the first local extinction of the pathogen (Figure S7 , right top). In countries with lower 503 connectivity (e.g., that might resemble Madagascar), local outbreaks can go extinct rapidly 504 before travelling very far; in other countries (e.g., that might resemble Gabon), the pathogen 505 goes extinct rapidly because it travels rapidly and rapidly depletes susceptible individuals 506

everywhere. 507 508

The impact of the pattern of travel between centroids is echoed by the pattern of travel within 509 administrative districts: countries where the pathogen does not reach a large fraction of the 510 administrative 2 units within the country in 5 years are also those where within administrative 511 unit travel is low (Figure S7 , right bottom). 512 513

These simulations provide a window onto qualitative patterns expected for subnational spread 514 of the pandemic virus, but there is no clear way of calibrating the absolute rate of travel between 515 regions of relevance for SARS-CoV-2. Thus, the time-scales of these simulations should be 516 considered in relative, rather than absolute terms. Variation in lockdown effectiveness, or other 517 changes in mobility for a given country may also compromise relative comparisons. Variability in 518 case reporting complicates clarifying this ( Figure S8) . 519 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint Rice et al | 2020 07 23 | Page 30

542

Cases and testing vs. the pace of the outbreak 543

The total number of confirmed cases reported by country (x axis, left, as reported for June 28th 544 by Africa CDC) and the test positivity (x axis, right, defined as the total number of confirmed 545 cases divided by the number of tests run, as reported by Africa CDC, likewise) show no 546 significant relationship with the proportion of the population estimated to be infected after one 547

year using the metapopulation simulation described in A6 (respectively,= = −0.04, : > 0.5, BC = 548 41 and = = 0.02, : > 0.5, BC = 41). All else equal, a positive relationship is expected; however, 549 both uncertainty in case numbers, and uncertainty associated with the simulation might both 550 drive the absence of a signal. 551 552 553 . 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 July 24, 2020. where S is the susceptible population, I is the infected population and N is the total population. 574 D is the mean infectious period, set at 5 days following ref 23, 49 . To investigate the maximum 575 possible climate effect, we use parameters from the most climate-dependent scenario in ref 23  ,  576 based on betacoronavirus HKU1. In this scenario L, the duration of immunity, is found to be 577 66.25 weeks (i.e., greater than 1 year and such that waning immunity does not affect timing of 578 the epidemic peak). 579 580

Transmission is governed by %(E) which is related to the basic reproduction number R0 by 581 F ) (E) = %(E)G. The basic reproduction number varies based on the climate and is related to 582 specific humidity according to the equation: 583 584 -! = ./0 (2 * 4(#) + 678 (-!"#$ − -!"%& ) ) + -!"%& 585 586

where q(t) is specific humidity 47 and a is set at -227.5 based on estimated HKU1 parameters 23 . 587

R0max and R0min are 2.5 and 1.5 respectively. We assume the same time of introduction for all 588 cities, set at March 1 st , 2020 (consistent with the first reported cases in SSA, Figure S1D ) 589 590 591 . 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 July 24, 2020. . https://doi.org/10.1101/2020.07.23.20161208 doi: medRxiv preprint

Effects of temperature and humidity 593 on the spread of COVID-19: A systematic review

Estimating the burden of SARS-CoV-2 in France

An empirical estimate of the infection fatality rate of COVID-19 597 from the first Italian outbreak

Estimates of the severity of coronavirus disease 2019: a model-based 599 analysis

Africa CDC -COVID-19 Daily Updates

Mortality Analyses. Johns Hopkins Coronavirus Resource Center

Seeding COVID-19 across sub-Saharan Africa: an analysis of reported 604 importation events across 40 countries

Case fatality risk of novel coronavirus diseases 2019 in China

OpenSAFELY: factors associated with COVID-19-related hospital 608 death in the linked electronic health records of 17 million adult NHS patients

COVID-19 significantly impacts health services for noncommunicable diseases

Maintaining essential health services: operational guidance for the COVID-19 context

Effects of the COVID-19 Pandemic on Routine Pediatric Vaccine Ordering 617

Administration -United States

Early estimates of the indirect effects of the COVID-19 pandemic on 619 maternal and child mortality in low-income and middle-income countries: a modelling study

The impact of COVID-19 and strategies for mitigation and 622 suppression in low-and middle-income countries

Differential Effects of Intervention Timing on COVID-19

the United States. medRxiv

Assessing the effect of global travel and contact reductions to mitigate the 627

COVID-19 pandemic and resurgence

Besides population age structure, health and other demographic 629 factors can contribute to understanding the COVID-19 burden

Global, regional, and national estimates of the population at increased risk of 632 severe COVID-19 due to underlying health conditions in 2020: a modelling study

Predicted COVID-19 fatality rates based on age, sex, comorbidities, and 635 health system capacity

Is Africa prepared for tackling the COVID-19 (SARS-CoV-2) epidemic

Lessons from past outbreaks, ongoing pan-African public health efforts, and implications for 640 the future

Testing, and Contact Tracing in Africa -Shared Approaches, Innovations, and 644 Challenges to Minimize Morbidity and Mortality

Susceptible supply 647 limits the role of climate in the early SARS-CoV-2 pandemic

A global map of travel time to cities to assess inequalities in accessibility 650 in 2015

Quantifying the impact of US state non-pharmaceutical interventions 655 on COVID-19 transmission

A global assessment of civil registration and vital statistics systems: 657 monitoring data quality and progress

Preparedness and vulnerability of African countries against importations of 659 COVID-19: a modelling study

Passengers' destinations from China: low risk of Novel Coronavirus

nCoV) transmission into Africa and South America

Reduced vaccination and the risk of measles and other childhood 663 infections post-Ebola

Has COVID-19 subverted global health?

Supporting the Health Care Workforce During the COVID-19

Ebola virus disease in health care workers

Covid 19 Mauritius Dashboard. Google Data Studio

Rwanda: WHO Coronavirus Disease (COVID-19) Dashboard

A Global lmmunological Observatory to meet a time of pandemics

World Population Prospects -Population Division -United Nations

The Novel Coronavirus Pneumonia Emergency Response Epidemiology Team. The 679

Epidemiological Characteristics of an Outbreak of 2019 Novel Coronavirus Diseases 680 (COVID-19) -China

Severe Outcomes Among Patients with Coronavirus Disease 2019 682 (COVID-19) -United States

High prevalence of obesity in severe acute respiratory syndrome 685 coronavirus-2 (SARS-CoV-2) requiring invasive mechanical ventilation

Can atmospheric pollution be considered a co-factor in 688 extremely high level of SARS-CoV-2 lethality in Northern Italy?

Collider bias undermines our understanding of COVID-19 disease risk and 691 severity

Nutritional Modulation of Malaria Morbidity and Mortality

Rice et al | 2020 07 23 | Page 36 and nutrition programs

county-level characteristics to inform equitable COVID-19 response

The ERA5 global reanalysis

Report 23 -State-level tracking of COVID-19 in the United States

Projecting the 703 transmission dynamics of SARS-CoV-2 through the postpandemic period

Inequalities in access to water and soap matter for the 706 COVID-19 response in sub-Saharan Africa

Beyond handwashing: Water insecurity undermines 708 COVID-19 response in developing areas

Keeping COVID-19 at bay in Africa

Ranking of elimination feasibility between malaria-endemic countries

Transport networks and inequities in vaccination: remoteness shapes 713 measles vaccine coverage and prospects for elimination across Africa

Africa's struggle with inadequate COVID-19 testing

The demographics of human and malaria movement and migration 718 patterns in East Africa

Impacts of social and economic factors on the transmission of 720 coronavirus disease (COVID-19) in China

Age-Dependent Risks of Incidence and Mortality of COVID-19 in Hubei 722 Province and Other Parts of China

GHDx

Global access to handwashing: 727 implications for COVID-19 control in low-income countries

National and sub-national variation in patterns of febrile case 730 management in sub-Saharan Africa

Intensive care unit capacity in low-income 732 countries: a systematic review

Population distribution, 734 settlement patterns and accessibility across Africa in 2010

Principal component analysis: a review and recent developments

A systematic review and meta-analysis of published 739 research data on COVID-19 infection-fatality rates

The R Project for Statistical Computing

GHDx

An interactive web-based dashboard to track COVID-19 in 747