Assessing The Burden of COVID-19 in Developing Countries BMJ Global Health

Original research
Assessing the burden of COVID-19 in
BMJ Glob Health: first published as 10.1136/bmjgh-2022-008477 on 26 May 2022. Downloaded from http://gh.bmj.com/ on July 1, 2022 by guest. Protected by copyright.
developing countries: systematic review,
meta-analysis and public
policy implications
Andrew T Levin ‍ ‍ ,1,2 Nana Owusu-Boaitey,3 Sierra Pugh,4 Bailey K Fosdick,5
Anthony B Zwi,6 Anup Malani ‍ ‍ ,7 Satej Soman ‍ ‍ ,8 Lonni Besançon,9
Ilya Kashnitsky,10 Sachin Ganesh,11 Aloysius McLaughlin,11 Gayeong Song,11
Rine Uhm,11 Daniel Herrera-Esposito,12 Gustavo de los Campos,13
Ana Carolina Peçanha Antonio ‍ ‍ ,14 Enyew Birru Tadese,15
Gideon Meyerowitz-Katz ‍ ‍ 16,17
To cite: Levin AT, ABSTRACT

Owusu-Boaitey N, Pugh S, Introduction The infection fatality rate (IFR) of COVID-19
WHAT IS ALREADY KNOWN ON THIS TOPIC
et al. Assessing the burden has been carefully measured and analysed in high-income ⇒ Prior meta-analyses of data from high-income coun-
of COVID-19 in developing tries have shown that the COVID-19 infection fatality
countries, whereas there has been no systematic analysis of
countries: systematic review, rate (IFR) increases exponentially with age while
age-specific seroprevalence or IFR for developing countries.
meta-analysis and public policy
Methods We systematically reviewed the literature to identify seroprevalence (as measured by antibodies against
implications. BMJ Global Health
2022;7:e008477. doi:10.1136/ all COVID-19 serology studies in developing countries that SARS-CoV-2) has been markedly lower for older
bmjgh-2022-008477 were conducted using representative samples collected by adults relative to younger adults.
February 2021. For each of the antibody assays used in these
WHAT THIS STUDY ADDS
Handling editor Seye Abimbola serology studies, we identified data on assay characteristics,
including the extent of seroreversion over time. We analysed ⇒ We analyse serology and mortality data from 62
► Additional supplemental studies of 25 developing countries, and we find that
the serology data using a Bayesian model that incorporates
material is published online only. age-stratified IFRs are about two times higher than
conventional sampling uncertainty as well as uncertainties
To view, please visit the journal the benchmark metaregression for high- income
online (http://dx.d oi.org/10.
about assay sensitivity and specificity. We then calculated
IFRs using individual case reports or aggregated public health countries.
1136/bmjgh-2022-0 08477).
updates, including age-specific estimates whenever feasible. ⇒ Indeed, population IFR in developing countries is
Results In most locations in developing countries, similar to that of high-income countries, because
Received 9 January 2022 seroprevalence among older adults was similar to that of differences in population age structure are roughly
Accepted 5 May 2022 younger age cohorts, underscoring the limited capacity offset by disparities in healthcare access and elevat-
that these nations have to protect older age groups. ed infection rates among older age cohorts.
Age-specific IFRs were roughly 2 times higher than in high-
HOW THIS STUDY MIGHT AFFECT RESEARCH,
income countries. The median value of the population IFR
PRACTICE OR POLICY
was about 0.5%, similar to that of high-income countries,
because disparities in healthcare access were roughly ⇒ Our findings underscore the urgency of disseminat-
offset by differences in population age structure. ing vaccines and effective medications throughout
Conclusion The burden of COVID-19 is far higher in the developing world.
developing countries than in high-income countries, reflecting
a combination of elevated transmission to middle-aged and
older adults as well as limited access to adequate healthcare. appeared to be low in the early stages of
© Author(s) (or their These results underscore the critical need to ensure medical the pandemic, suggesting that the relatively
employer(s)) 2022. Re-use equity to populations in developing countries through provision younger age structure of these countries
permitted under CC BY-NC. No of vaccine doses and effective medications. might have protected them against the harms
commercial re-use. See rights
and permissions. Published by of the disease. More recently, however, it has
BMJ. become clear that the perceived differences
For numbered affiliations see INTRODUCTION in mortality may have been illusory, reflecting
end of article. An important unknown during the COVID-19 poor vital statistics systems leading to under-
pandemic has been the relative severity of the reporting of COVID-19 deaths.1 2 Moreover,
Correspondence to
Gideon Meyerowitz-Katz;
disease in developing countries compared relatively low mortality outcomes in devel-
gideon.meyerowitzkatz@health. with higher- income nations. The incidence oping countries would be starkly different
nsw.g ov.au of fatalities in many developing countries from the typical pattern observed for many
Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 1

BMJ Global Health
far more dangerous for middle-aged and older adults

Table 1 Confirmed COVID-19 deaths as of 20 March 2022
compared with children and young people.12 13 15 Two
Cumulative Mortality rate prior meta-analytical studies have considered variations
Country deaths per million
in IFR by age but did not consider the possibility that IFR
USA 971 162 2917.1 in developing locations might differ systematically from
Brazil 657 495 3072.5 high-income countries due to healthcare quality, access
India 516 510 370.7 and other socioeconomic factors.12 16
Russia 357 234 2448.3 Objectives
Mexico 322 072 2472.5 1. Determine overall prevalence of COVID-19 infection
Peru 211 865 6351.0 in locations in developing countries.
UK 163 658 2399.4 2. Assess age-specific patterns of seroprevalence in these
locations.
Italy 157 785 2613.7
3. Estimate age-specific IFRs and compare to benchmark
Indonesia 153 738 556.3 values for high-income countries.
France 141 002 2091.3 4. Investigate possible reasons for differences in popula-
Iran 139 610 1641.9 tion IFR between locations.
Colombia 139 452 2720.2
Source: Our World In Data.56 METHODS
To perform this meta-analysis, we collected published
papers, preprints and government reports of COVID-19
other communicable diseases, reflecting the generally serology studies for which all specimens were collected
lower access to good- quality healthcare in these loca- before 1 March 2021 and that were publicly disseminated
tions.3 4 by 17 December 2021. The full search methodology is
As shown in table 1, mortality attributable to COVID-19 given in online supplemental appendix 1A. The study was
in many developing locations exceeds 2000 deaths per registered on the Open Science Foundation: https://osf.
million. Of the 12 nations with the highest number of io/edpwv/
deaths attributed to COVID- 19, eight are developing We restricted the scope of our analysis to locations
countries. Furthermore, these statistics may understate in developing countries using the classification system
the true death toll in a number of lower-income and of the International Monetary Fund (IMF); that is, we
middle- income countries. Numerous studies of excess excluded locations that the IMF classifies as ‘high-income
mortality have underscored the limitations of vital regis- countries’.17 In some contexts developing countries are
tration and death reporting, particularly in developing also described as low-income to middle-income countries
countries.1 2 5–9 For example, recent studies of India have or as emerging and developing economies.
found that actual deaths from COVID-19 were about 10
times higher than those in official reports.2 5 Similarly, Inclusion/exclusion criteria
a study in Zambia found that only 1 in 10 of those who Our analysis only included studies that had a random
died with COVID-19 symptoms and whose postmortem selection of participants from a sample frame repre-
COVID-19 test was positive were recorded as COVID-19 sentative of the general population.18 19 Consequently,
deaths in the national registry.10 Strikingly, the contin- studies of convenience samples—such as blood donors
uation of that study has demonstrated the catastrophic or residual sera from commercial laboratories—were
impact of COVID- 19 in Zambia, raising the overall excluded. Such samples are subject to intrinsic selec-
mortality by as much as 5–10 times relative to a normal tion biases that may vary across different settings and
year.11 hence would detract from systematic analysis of the data.
There has, however, been a relative dearth of systematic Indeed, there is abundant evidence from the pandemic
research concerning the early experience of COVID-19 that convenience samples provide inaccurate estimates of
and the associated infection fatality rate (IFR) in devel- seroprevalence, with assessments indicating that they are
oping countries. Previous evaluations have largely focused likely to overestimate the true proportion infected.20 21
on assessing these patterns in high- income countries, A crucial part of our analysis entailed adjusting raw
where high-quality data on seroprevalence and fatalities seroprevalence to reflect the sensitivity and specificity of
has been readily available throughout the pandemic.12 13 the particular assay used in each serology study, and to
In particular, seroprevalence studies conducted in high- construct credible intervals that reflect uncertainty about
income countries in 2020 found low overall prevalence of assay characteristics as well as conventional sampling
antibodies to COVID-19 (generally less than 10%),14 with uncertainty. Where a reported study did not include that
much lower prevalence among older adults compared information, we requested it from study authors. Other
with younger cohorts. Analysis of these data has clearly data needed and extracted for the analysis included start
underscored the extent to which the IFR of COVID-19 and end dates of specimen collection, the specific assay
increases exponentially with age; that is, the disease is used and age-specific serology data.
2 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

BMJ Global Health
See online supplemental appendix 1b for further prevalence for each location at the resolution of the
details on inclusion and exclusion criteria. serology data reported. Then, we model the number of
people that test positive in a given study location and age
Deaths group as coming from a binomial distribution with a test
For locations with publicly available databases of all indi- positivity probability that is a function of the true preva-
vidual cases, we tabulated the fatality data to match the lence, sensitivity and specificity, accounting for serocon-
age brackets of that serology study, using cumulative version and seroreversion (see the online supplemental
fatalities as of 14 days after the midpoint date of spec- appendix 2B).
imen collection to reflect the time lags between infec- As in Carpenter and Gelman (2020),23 we consider
tion, seropositivity and fatal outcomes. In the absence of sensitivity and specificity to be unknown and directly
individual case data, we searched for contemporaneous model the lab validation data (eg, true positives, true
public health reports and tabulated cumulative deaths as negatives, false positives, false negatives) for each test.
of 28 days after the midpoint date of specimen collection Independent weakly informative priors are placed on
to incorporate the additional time lags associated with the seroprevalence parameters, and independent, infor-
real-time reporting of COVID-19 fatalities (see online mative priors akin to those in Carpenter and Gelman23
supplemental appendix 1d for discussion of death lags). are placed on the sensitivity and specificity parameters.
Matching prevalence estimates with subsequent fatal- To avoid assumptions about the variability of prevalence
ities is not feasible if a serology study was conducted in across age within a serology age bin, we aggregate deaths
the midst of an accelerating outbreak. Therefore, as in for each location to match their respective serology age
previous work,15 we estimated seroprevalence but did not bins. Independent mildly informative priors are assumed
analyse IFRs for locations where the cumulative death on the age-group-specific IFR parameters.
toll increased by three times or more over the 4-week Prevalence for a given age group and location is esti-
period following the midpoint date of specimen collec- mated by the posterior mean and equal-tailed 95% cred-
tion. For details, see the online supplemental appendix ible interval. Uniform prevalence across age is deemed
1d. In instances where we were not able to match deaths plausible for locations where the 95% credible intervals
to serology data, or there were accelerating outbreaks, we for the ratio of seroprevalence for age 60 years and older
used this information to look at serology only. over the seroprevalence estimate for ages 20 years to 60
Additionally, we extracted data on excess deaths for years contains 1.
all countries that were included in our IFR analysis.
We used two primary sources of estimates on excess IFR calculation and comparison
mortality: the Institute for Health Metrics and Evaluation We model the number of individuals at a given location
(IHME)22 and the World Mortality Dataset (WMD).1 The and age group that are reported as dying of COVID-19
IHME produces national or regional estimates of excess as Poisson distributed with rate equal to the product of
mortality for every location included in this review, while the age group IFR, age group population and age group
the WMD has estimates for a subset of those locations. We prevalence. For locations where deaths were reported
then computed the ratio of excess mortality to reported separately for different age bins this model provides IFR
fatalities for each location in order to assess the impact of estimates for specific age groups and for broader popula-
potential death under-reporting, and calculated adjusted tion cohorts, including adults aged 18–65 years. For loca-
IFRs using excess mortality as the numerator, as well as tions where death data were not disaggregated by age the
the ratio between IFRs calculated using reported and model provides a population IFR. The model was imple-
excess deaths. We used excess mortality as it is likely to mented in the programming language R, with posterior
better represent the true burden of COVID-19 accurately sampling computation implemented with the Stan soft-
in developing nations.1 ware package.24
To perform a meta-analysis of age-specific IFRs across
Adjustment for seroreversion locations, we conduct a metaregression with random
For those assays used in serology studies included in our effects. In the metaregression, the dependent variable is
analysis, we classified each assay’s risk of seroreversion the estimated IFR for a specific age group in a specific
(high, medium or low) based on longitudinal serology geographical location, the explanatory variable is the
studies and serological analysis of prior RT- PCR posi- median age of that particular age group, and the SD of
tive cases. For each location for which the assay used in each idiosyncratic error is taken from the Bayesian anal-
serology was classified as having high risk of serorever- ysis described above. We used a random-effects proce-
sion, we made adjustments to the data on assay sensitivity. dure to allow for residual heterogeneity between studies
See online supplemental appendix 2a for further details. and across age groups by assuming that these divergences
are drawn from a Gaussian distribution. We also allowed
Statistical analysis for fixed effects by location, to account for locations that
We use a Bayesian modelling framework to simulta- deviate from the norm. Since the metaregression used
neously estimate age-
specific prevalence and IFRs for IFR estimates based on reported deaths, we compared the
each location in our study. First, we model age-specific location-specific fixed effects to two estimates of the ratio

BMJ Global Health
of excess mortality to COVID-19 deaths in each location. studies (representing a total of 25 developing countries)
We also compared these metaregression results to a prior could be used to produce IFR estimates; see online
metaregression of age-specific IFR for high-income coun- supplemental appendix 1c for details. The geographical
tries;15 further details are given in online supplemental distribution of these studies is shown in figure 1, while
appendix 2d and table A5. This was performed using table 2 provides a list of the studies used in producing
the meta regression procedure in Stata V.17. Finally, we IFR estimates, including the specimen collection dates
computed population IFRs adjusted for COVID-19 death and the assay used in each study. Further details are
undercounting and compared these estimates to the provided in our GitHub repository https://covid-ifr.
proportion of well-certified deaths. github.io/.
Covariates
We selected covariates that were judged likely to have an Seroprevalence
impact either on the IFR of COVID-19 itself or on the accu- As shown in figure 2A and 2B, seroprevalence reached
racy of official data on COVID-related mortality based on relatively high levels in numerous locations in developing
prior research and expertise. Such covariates included countries during the time frame covered by our analysis.
GDP per capita and measures of healthcare capacity; The upper panel shows estimates from studies where spec-
the complete list is provided in online supplemental imens were collected between April and September 2020,
appendix 1f. Where possible, we extracted these covari- while the lower panel shows corresponding estimates for
ates at a state or regional level within a country; other- the period from October 2020 to February 2021.
wise, they were identified at a national level. In instances In most developing country locations, seroprevalence
where a covariate was only available at the national level, was roughly uniform across age strata. Figure 3 shows
we aggregated location-specific seroprevalence and IFRs the heatmap of age-specific seroprevalence across all age
by weighting each location using the square root of the cohorts. As shown in figure 4, the ratio of seroprevalence
number of serology specimens collected in that location. for older adults (ages 60+ years) compared with middle-
aged adults (ages 40–59 years) is indistinguishable from
RESULTS unity in most of these locations. While many locations
We identified a total of 2384 study records, with 2281 had a ratio below 1, the majority of the areas were very
records identified from online databases and a further substantially above the ratio for higher- income areas
124 from Twitter, Google Scholar and a prior publica- (green shaded region), and the point estimates were
tion.25 After excluding 2062 records, we assessed 343 not markedly below 1, indicating minimal difference
records and determined that 97 studies satisfied the in infection rates between older and younger adults in
criteria for inclusion in the final analyses, of which 62 developing nations.
Figure 1 Map of study locations. IFR, infection fatality rate.

BMJ Global Health
Table 2 Included studies for infection fatality rate (IFR)
Location Date range Assay used Citation
Latin America
68
Argentina Buenos Aires City 10 September to 18 October 2020 COVIDAR IgG
69
Municipality of Hurlingham 26 November to 10 December 2020 COVIDAR IgG
70
Bolivia Santa Cruz 22 August to 13 September 2020 Standard Q IgG/IgM
71
Brazil Cuiabá 16 September to 15 October 2020 DiaSorin Liaison IgG
72
Distrito Federal 2–17 December 2020 CTK Biotech Onsite IgG/IgM
73
Foz do Iguaçu 14 May to 9 June 2020 Proprietary
74
Maranhao 27 July to 8 August 2020 Roche Elecsys IgG/IgM
71
Mato Grosso 16 September to 15 October 2020 DiaSorin Liaison IgG
75
Pitangueiras 24 August to 29 September 2020 ECO IgG/IgM
76
Rio Grande do Sul 5–7 February 2021 University of Rio de Janeiro
77
Sao Paulo City 1–10 October 2020 Roche Elecsys IgG/IgM
71
Várzea Grande 16 September to 15 October 2020 DiaSorin Liaison IgG
78
Chile Santiago/ Coquimbo/Talca 26 September to 25 November 2020 Roche Elecsys IgG/IgM
79
Colombia Barranquilla 20–30 September 2020 Siemens Advia IgG/IgM
79
Bogotá 10 October to 5 November 2020 Siemens Advia IgG/IgM
79
Bucaramanga 27 September to 9 October 2020 Siemens Advia IgG/IgM
79
Cali 18–28 November 2020 Siemens Advia IgG/IgM
80
Córdoba (eight cities) 1 July to 29 October 2020 INgezim DR IgG/IgM/IgA
79
Cucuta 5–15 October 2020 Siemens Advia IgG/IgM
79
Ipiales 3–11 December 2020 Siemens Advia IgG/IgM
79
Leticia 15–25 September 2020 Siemens Advia IgG/IgM
79
Medellín 5 October to 20 December 2020 Siemens Advia IgG/IgM
79
Villavicencio 20–30 October 2020 Siemens Advia IgG/IgM
81
Ecuador Cuenca 11 August to 1 November 2020 Standard Q IgG/IgM
82
Mexico Nationwide 18 August to 13 November 2020 Roche Elecsys IgG/IgM
83
Paraguay Asunción & Central Dept. 23 December to 16 February 2021 Beijing Kewei IgG/IgM
84
Peru Cusco Province 12–27 September 2020 Roche Elecsys IgG/IgM
85
Iquitos, Loreto 13–18 July 2020 Orient Gene Biotech IgG/IgM
86
Lambayeque 24 June to 10 July 2020 Coretest IgG/IgM
40
Lima and Callao 28 June to 9 July 2020 Standard Q IgG/IgM
Africa
87
Ethiopia Addis Ababa 22 July to 10 August 2020 Core Technology IgG
88
Dire Dawa 15 June to 30 July 2020 Abbott Architect IgG
89
Kenya Nairobi County 2–23 November 2020 Wantai IgG/IgM
90
Mozambique Maputo city 3–21 August 2020 Abbott PanBio IgG/IgM
91
Senegal Nationwide 24 October to 26 November 2020 Wantai IgG/IgM
92
South Africa Gauteng 4 November to 22 January 2021 Luminex S IgG
93
Mitchells Plain 8 December to 31 January 2021 Wantai IgG/IgM
94
Zambia Lusaka & Ndola 4–27 July 2020 Euroimmun IgG
Middle East
95
Iran Nationwide 3 August to 31 October 2020 Pishtaz Teb IgG/IgM
96
Jordan Nationwide 27 December to 6 January 2021 Wantai IgG/IgM
97
Oman Nationwide 12–19 July 2020 DiaSorin Liaison IgG
Europe
Continued

BMJ Global Health
Table 2 Continued
Location Date range Assay used Citation
25
Bosnia & Republika Srpska 4 November to 16 December 2020 Wantai IgG/IgM
Herzegovina
98
Hungary National Study 1–16 May 2020 Abbott Architect IgG
99
Poland Katowice region 1 October to 30 November 2020 Euroimmun IgG
21
Russia St. Petersburg 25 May to 28 June 2020 Genetico CoronaPass Total
South Asia
100
India Berhampur 6–6 August 2020 Roche Elecsys IgG/IgM
100
Bhubaneswar 10–10 July 2020 Roche Elecsys IgG/IgM
101
Chennai 17–28 July 2020 Abbott Architect IgG
102
Delhi 1–7 August 2020 Zydus Kavach IgG
103
Karnataka 15 June to 29 August 2020 THSTI IgG
104
Malegaon 25 July to 20 August 2020 Karwa Kavach IgG
42
Mumbai 29 June to 19 July 2020 Abbott Architect IgG
105
Paschim Medinipur 27 July to 7 August 2020 ErbaLisa IgG
106
Pimpri-Chinchwad 7–17 October 2020 Abbott Architect IgG
107
Puducherry 10–16 September 2020 Roche Elecsys IgG/IgM
108
Srinagar 17–20 October 2020 Abbott Architect IgG
109
Tamil Nadu 19 October to 30 November 2020 iFlash IgG & Vitros IgG
110
Nepal Nationwide 9–22 October 2020 Wantai IgG/IgM
111
Pakistan Karachi 15–31 July 2020 Roche Elecsys IgG/IgM
111
Lahore 15–31 July 2020 Roche Elecsys IgG/IgM
East Asia
112
China Hubei (excluding Wuhan) 10–18 April 2020 Bioscience IgG/IgM
112
China Wuhan 10–18 April 2020 Bioscience IgG/IgM
Infection fatality rates the metaregression prediction, while those for Chennai
Our statistical analysis produced age-specific IFRs and CIs and Karnataka are 1/5 and 1/20, respectively.
for 28 locations, and population IFRs for those locations This metaregression analysis uses age- specific IFRs
as well as an additional 27 places. The full results of this based on reported COVID-19 deaths in each location. As
analysis are shown in the online supplemental appendix a crosscheck, table 3 reports the ratio of excess mortality
3. We obtain the following metaregression results: to reported deaths for each of these locations. For nearly
log10 (IFR) = −2.75 + 0.0478 ∗ age all of these locations, the ratio is indistinguishable from
‍ (0.10)(0.0023) ‍ unity; that is, reported COVID- 19 deaths are broadly
consistent with the evidence from excess mortality assess-
where IFR is expressed in percentage points, and the SE
for each estimated coefficient is given in parentheses. ments. There were three exceptions (Chennai, Karna-
These estimates are highly significant with t-statistics of taka and Nairobi, Kenya), two of which had significant
−28.7 and 21.0, respectively, and p values below 0·0001. location-specific effects in the metaregression.
The residual heterogeneity is τ2=0.039 (p<0.0001) and The precision of IFR estimates varied by age. At lower
I2=92.5, confirming that the random effects are essential age groups, the number of deaths becomes very small,
for capturing unexplained variations across studies and and thus the uncertainty is large regarding the IFR.
age groups. The adjusted R2 is 91.1%. Location-specific Conversely, at older ages the number of infections and
fixed effects are only distinguishable from zero for three deaths can be very small in countries with extremely
locations: Maranhão, Brazil (−0.50); Chennai, India small populations of those aged over 65 years, and thus
(−0.68); and Karnataka, India (−1.29). these estimates are also uncertain. The detailed analysis
The metaregression results can be seen in figure 5. of age-specific IFR for each location is provided in online
Nearly all of the observations fall within the 95% predic- supplemental appendix figure A6.
tion interval. The importance of the location- specific Figure 6 shows that these age-specific IFRs are system-
effects is readily apparent. Indeed, these effects imply atically higher than those of a prior metaregression esti-
that the age-specific IFRs for Maranhão are about 1/3 of mated using studies of high- income countries.15 That

BMJ Global Health
Figure 2 Estimates of seroprevalence.
benchmark metaregression has a slope of 0.0524 (95% Chennai and Karnataka); the estimated intercept and
CI 0.0499 to 0.0549), and a Welch test strongly rejects the slope coefficient of this variant (−2.68 and 0.0480, respec-
hypothesis of equality in the slope parameters for devel- tively) are statistically indistinguishable from the baseline
oping countries versus high-income countries with a value values shown above.
of p<0.0001. This figure also shows a variant of our metare- Figure 7 shows estimates of population IFR at ages
gression, estimated using studies of developing country 18–65 years, adjusted for excess mortality using the ratios
locations conducted over the same time frame as in the shown in table 3. To facilitate comparability across loca-
benchmark metaregression (April to September 2020) tions, these estimates use a standardised age structure
and excluding the three outlier locations (Maranhão, to aggregate the age-specific prevalence and fatalities in

BMJ Global Health
Figure 3 Age‐specific seroprevalence by location.
each location. Corresponding estimates, using the actual of deaths were well certified (using Sustainable Develop-
population age structure of each location, are shown in ment Goal (SDG) assessments26 conducted prior to the
online supplemental appendix figure A12. pandemic), compared with only 0.05% in countries with
lower proportions of well-certified deaths. In the latter
Assessment of death reporting
For the full set of locations for which population IFR can set of countries, adjustments for excess mortality shift the
be assessed, we found that the adequacy of death certifi- population IFR upwards by an order of magnitude, to a
cation was highly significant in explaining cross-country median of 0.6%. Indeed, the population IFR for Zambia
variations. As shown in figure 8, the median value of popu- increases from 0.23% to 1.96%—the highest value for any
lation IFR was about 0.5% in countries where a majority country in our sample. In contrast, the excess mortality

BMJ Global Health
Figure 4 Ratio of seroprevalence for older adults (60+ years) compared with adults (40–59 years).
adjustments make relatively little difference for countries DISCUSSION

with a majority of well-certified deaths. COVID-19 has had a severe burden on developing coun-
Finally, we considered the extent to which the adjusted tries. Prevalence in developing countries is roughly
measures of population IFR were robust to alternative uniform across age groups, in contrast to the typical
estimates of the ratio of excess mortality to reported pattern in high-income countries where seroprevalence
deaths. As shown in figure 9 the estimates from IHME is markedly lower among middle-aged and older adults
and WMD were generally well aligned, with just a small who are most vulnerable to this disease. Moreover, the IFR
number of exceptions. is substantially higher in developing countries compared
The adjusted population IFRs had a median value of with high-income countries.
0.49% using the IHME estimates and 0.58% using the At 20 years of age, the mean IFR in developing countries
WMD estimates. is 2.7 times higher than that in high-income countries

BMJ Global Health
Figure 5 metaregression results. IFR, infection fatality rate. Figure 6 IFR in developing countries compared to high‐
income countries. IFR, infection fatality rate.
Table 3 Ratio of excess mortality to reported COVID-19 deaths

Country Location Ratio (95% CI)
Argentina Buenos Aires City 1.07 (1.0 to 1.5)
Argentina Municipality of Hurlingham 1.07 (1.0 to 1.5)
Brazil Maranhao 1.41 (1.0 to 2.4)
Brazil Sao Paulo City 1.02 (1.0 to 1.3)
Brazil Cuiaba, Mato Grosso 1.00 (1.0 to 1.0)
Brazil Varzea Grande, Mato Grosso 1.00 (1.0 to 1.0)
Chile Coquimbo-La Serena, Greater Santiago, Talca 1.00 (1.0 to 1.0)
China Wuhan 1.00 (1.0 to 1.0)
Colombia Leticia (Amazonas) 1.09 (1.0 to 1.6)
Colombia Barranquilla (Atlantico) 1.09 (1.0 to 1.6)
Colombia Medellin (Antioquia) 1.09 (1.0 to 1.6)
Colombia Bucaramanga (Santander) 1.09 (1.0 to 1.6)
Colombia Cucuta (Norte Santander) 1.09 (1.0 to 1.6)
Colombia Villavicencio (Meta) 1.09 (1.0 to 1.6)
Colombia Bogota 1.09 (1.0 to 1.6)
Colombia Cali (Valle del Cauca) 1.09 (1.0 to 1.6)
Colombia Ipiales (Narino) 1.09 (1.0 to 1.6)
Colombia Cordoba: 8 cities 1.09 (1.0 to 1.6)
Ecuador Cuenca (Azuay) 1.01 (1.0 to 1.1)
Hungary National Study 1.04 (1.0 to 1.4)
India Karnataka 4.89 (2.6 to 8.2)
India Chennai 4.80 (2.7 to 7.9)
Jordan National Study 1.57 (1.0 to 3.0)
Kenya Nairobi County 13.29 (7.1 to 23.1)
Paraguay Asuncion+Central Department 1.10 (1.0 to 1.6)
Peru Lambayeque 1.09 (1.0 to 1.6)
Peru Lima (Metropolitana)+Callao 1.09 (1.0 to 1.6)
Peru Iquitos, Loreto 1.09 (1.0 to 1.6)
Note: This table shows Institute for Health Metrics and Evaluation (IHME) estimates of the ratio of excess mortality to reported COVID-19 deaths
(constrained to be 1.0 or greater).22 The 95% CIs, enclosed in parentheses, are also taken directly from IHME, with a one-tailed interval for each
location where the estimated undercount ratio is constrained by the lower bound of unity.

BMJ Global Health
Figure 7 Population IFR for ages 18–65 years. IFR, infection fatality rate.
and at age 60 years the risk is doubled. At the oldest specific places such as Santiago, Chile.7 This warrants
ages, this discrepancy is reduced, with only a modestly further research to understand why access to healthcare
increased risk at age 80 years. These relationships have and other socioeconomic issues appear to have a larger
also been found with socioeconomic status within impact on survival at younger ages. This finding does not
Figure 8 Population IFR and well‐certified death registrations. IFR, infection fatality rate.

BMJ Global Health
Another important facet of our results is that sero-

prevalence was both higher and consistent across age
groups in developing countries—a striking contrast to
the typical pattern in high- income countries, where
prevalence among older adults was markedly lower than
among younger adults.15 37 38 Evidently, it is very difficult
to insulate elderly people from the virus in a slum or a
rural village. This is likely also impacted by the higher
proportion of multigenerational families in developing
countries,39 a known risk-factor for COVID-19 infection
and death.40 41 For example, seroprevalence in slum
neighbourhoods of Mumbai was about four times higher
than in non-slum neighbourhoods.42 Our analysis indi-
cates that the relatively uniform prevalence of COVID-19
Figure 9 Excess mortality adjusted population IFRs. IFR,
in developing countries has dramatically increased the
infection fatality rate.
number of fatalities in these locations.
Our findings reinforce the conclusions of previous
rely on any specific modelling assumptions such as log- studies that have assessed the IFR of COVID-19.13 43 In
linearity, which is shown by the readily apparent disparity particular, COVID- 19 is dangerous for middle- aged
in figure 7, showing the age-standardised IFR for each adults, not just the elderly and infirm.15 Our metaregres-
individual location compared with the benchmark of sion results are well aligned with IFR estimates produced
high-income countries. for specific locations in developing countries (see supple-
The elevated IFR in developing nations only becomes mentary online supplemental appendix table A8).
apparent when stratifying by age and adjusting for death Our analysis underscores that incomplete death
under-reporting. Indeed, the quality of the vital statistics reporting is a crucial source of apparent differences in
system tends to be linked to the overall level of economic COVID-19 death rates. In particular, this is related to
development, and hence some previous studies of unad- the proportion of deaths that are assigned to so-called
justed data have incorrectly inferred that population IFRs ‘garbage codes’.26 44 45 These deaths are, by definition,
are lower in developing countries than in high-income not included in national tallies of the population that has
countries.27–30 died from COVID-19. As shown in figure 8, the IFR is
Our results are important for addressing questions that on average 10 times higher in locations with reasonably
have arisen about whether COVID-19 was less dangerous adequate vital statistics compared with other locations
for populations in sub- Saharan Africa compared with where a majority of deaths are not well certified.
locations elsewhere.31–33 As shown in figure 7, the age- The divergence between population IFRs for loca-
standardised population IFR of Nairobi County, Kenya tions is similar whether adjusted for death certification
is about five times higher than the high-income country or excess mortality. Adjustment for estimates of excess
benchmark. Likewise, figure 8 shows that the population mortality produced location population IFRs that were
IFR (adjusted for under-reporting of COVID-19 fatalities) consistent with IFRs produced in the age-stratified anal-
exceeds 0.5% for locations in Ethiopia, Mozambique and ysis, aside from a few minor outliers. As shown in figure 8,
South Africa; the sole exception is Senegal, perhaps due the median of these population IFRs for developing
to even more severe death undercounting than captured nations, once adjusted for undercounting of COVID-19
by the estimated ratio. These results underscore the deaths, was 0.58%, very similar to the median estimates
importance of drawing inferences from representative of IFR for high-income countries.46
samples rather than from convenience samples.34–36 Excess mortality is a useful metric for adjusting IFR esti-
These results are consistent with the pattern observed mates in areas where deaths are well registered but not well
for most other communicable diseases.3 4 In locations certified; that is: captured in national vital statistics but
with little ability to work from home, where quarantine is without a specific cause of death.1 Nonetheless, caution
difficult or impossible, where opportunities for physical is warranted in applying national estimates of excess
distancing and access to sanitation are poor, with lower mortality to specific regions within a country, recognising
healthcare resources, and where even basic resources that death reporting systems may vary markedly with the
such as supplemental oxygen are in short supply, people degree of urbanisation and other socioeconomic factors.
have fared substantially worse during the pandemic than In the case of Ecuador, for example, the national estimate
in high-income settings. Indeed, in low-income settings for the ratio of excess mortality to reported COVID-19
where fewer hospital beds and healthcare workers are deaths in 2020 was 2.6 (1, 22), whereas that ratio was only
available, COVID-19 has caused great devastation and an 1.01 in the province of Azuay.47
enormous death toll. With a much higher IFR, particu- Moreover, estimates of excess mortality may partly
larly in younger people, the ultimate burden for devel- reflect indirect effects of the pandemic on other sources
oping nations from COVID-19 is likely to be very high. of mortality. On the one hand, non- pharmaceutical

BMJ Global Health
interventions (such as business closures) may reduce Our analysis makes a strong case for swifter action on
mortality from causes such as vehicle accidents.1 48 vaccine and other medication equity. While countries
Conversely, mortality may be elevated by impaired access have largely sought to protect their own populations,
to healthcare for non-infectious diseases such as chronic there is increasing commitment to ensuring that key
cardiovascular disease or cancer,49 or by higher burdens populations in low-income and middle-income coun-
of non-COVID infectious diseases such as malaria, tuber- tries receive protection, at a minimum for their front-
culosis or parasitic infections.50 line health and other personnel. It is widely accepted
Finally, the true burden of COVID-19 may be practically that failing to control the pandemic across the globe
impossible to assess in locations where many deaths are will contribute to the emergence of additional strains
never entered into the national vital statistics system.51 of COVID- 19, potentially undermining the efficacy
For example, total mortality in Kenya was lower in 2020 of available vaccines.54 Current medication distribu-
than in 2019, but those statistics should certainly not tion efforts are grossly inequitable.55 Recent estimates
be interpreted as suggesting that Kenya was unscathed suggest that fewer than 10% of people in low-income
by the pandemic.22 Indeed, assessments of Kenya’s vital countries have received an immunisation, while the
statistics found that only two- thirds of actual deaths majority of people in high-income countries have had
were recorded in the system.51 Such considerations may at least one vaccination.56 Similarly, the availability of
explain other outliers in our analysis, such as Senegal, effective medications such as Paxlovid is grossly inequi-
which remains far below similar locations even when esti- table across the globe.57
mates are adjusted for excess mortality. As with all research, our study is subject to a number of
A useful example in this case is Ethiopia. Despite limitations. First, while we made every effort to capture
national statistics not showing a large increase in deaths seroprevalence data, including corresponding with
in Ethiopia during the pandemic, an epidemiological dozens of researchers and public health officials world-
investigation of burial sites has revealed a huge increase wide, it is possible that some studies have been missed.
in mortality during this period that is not part of the offi- However, it is unlikely that any small number of addi-
cial reporting of COVID-19.52 tional studies would make a material difference to our
In the absence of better death reporting, it is chal- results.
lenging to assess the extent to which differences in IFR Our analysis did not incorporate time series data on the
across locations reflect systematic disparities in health- evolution of COVID-19 deaths. However, some studies of
care access, socioeconomic status and other indicators. high-income countries have shown how such data can be
Nonetheless, such effects have been clearly demonstrated useful in refining assessment of IFR to incorporate the
by studies that have assessed distinct socioeconomic stochastic timing of COVID-19 deaths.13 58 Such analysis
groups within specific regions such as Santiago, Chile.7 should be a priority for future research about IFR in
Moreover, these considerations are almost certainly rele- developing countries.
vant in interpreting our finding that age-stratified IFR is While our analysis excluded convenience samples
markedly higher in developing countries compared with and focused exclusively on representative samples of
high-income countries.53 Indeed, our results underscore the population, we recognise that such studies may also
the tragedy that a Zambian young adult with COVID-19 be susceptible to selection bias. Research conducted at
would be far more likely to die than a Swiss person of various stages of the pandemic has found that individual
similar age. preferences for testing can be associated with substantial
Accounting for seroreversion and other assay charac- bias in estimates of seroprevalence, with corresponding
teristics is crucial for assessing seroprevalence accurately. implications for estimates of population IFR.20 59 Such
Our analysis makes a novel contribution in providing uncertainty can be incorporated into statistical models of
a systematic assessment of the implications of serorev- prevalence and IFR.60 61 However, we did not follow such
ersion; that is, the proportion of people who develop an approach here, because our statistical model already
antibodies but whose tests will fall below the limit of incorporates a number of other substantial sources of
detection at a later date. Prior studies have either ignored uncertainty.
this issue or have assumed that seroreversion occurs at a Our work also did not consider non-mortality harms
fixed geometric rate regardless of the assay used.12 13 In from COVID-19. Recent work has shown that even at
contrast, we have collated detailed information about the younger ages a substantial fraction of infected individuals
characteristics of all assays used in the serology studies will have severe, long-lasting adverse effects from COVID-
included in our analysis, including data on seroreversion 19.62 Consequently, the impact on the healthcare system
as well as test specificity and sensitivity; that information and society may be far greater than would be reflected
is fully described in online supplemental appendix 2a in mortality rates alone. Focusing only on survival rates
and b. Our analysis clearly indicates that the extent of obscures the large number of deaths that occur when
seroreversion differs in magnitude depending on the many people are infected,63 the relatively high fatality
assay used. Moreover, accounting for seroreversion had rate of COVID-19 in comparison to other diseases and
substantial implications for a number of locations in our other causes of death,64 and non- mortality harms of
analysis. COVID-19, such as hospitalisation from serious disease.62

BMJ Global Health
11
Future work should address these non-mortality harms, Department of Economics, Dartmouth College, Hanover, New Hampshire, USA
12
including long COVID-19. Laboratorio de Neurociencias, Universidad de la República, Montevideo, Uruguay
13
Department of Epidemiology & Biostatistics, Michigan State University, East
Another potentially serious limitation of our anal-
Lansing, Michigan, USA
ysis is cross-reactivity in serological tests due to malaria. 14
Adult Intensive Care Unit, Hospital de Clínicas de Porto Alegre, Porto Alegre, Brazil
An investigation in Nigeria found that the commonly 15
Ethiopian Public Health Institute, Addis Ababa, Ethiopia
used Abbott and Euroimmun serological assays had a 16
Western Sydney Diabetes, Western Sydney Local Health District, Blacktown, New
false-positive rate of 6.1% against prepandemic samples South Wales, Australia
17
due to cross- reactivity with malarial antibodies.65 This School of Health and Society, University of Wollongong, Wollongong, New South
Wales, Australia
would substantially lower specificity of the assay in areas
with a high prevalence of past malaria infection, which Twitter Anthony B Zwi @HEARDatUNSW, Anup Malani @anup_malani, Satej Soman
@satejsoman, Ana Carolina Peçanha Antonio @AnaCarolPecanha and Enyew Birru
would have the practical result of producing an upward
Tadese @Enyew54639156
bias of seroprevalence estimates and downward bias of
IFR estimates. Thus, it is plausible that in areas with a Acknowledgements The authors thank Ariel Karlinsky for assistance with death
large burden of malaria, that the IFR we have calculated registration and mortality data.
represents a substantial underestimate. Contributors ATL and GM-K initiated and provided leadership for the project, and
act as guarantors for the project. BKF and SP designated the Bayesian statistical
Finally, our analysis only includes serology studies
framework. NO-B took primary responsibility for the search procedures, and
where specimen collection was completed by the end of performed the review of assay characteristics and seroreversion. ATL and NO-B
February 2021. Consequently, our results do not reflect reviewed each of the studies identified in the initial screening, and assessed and
any potential changes in IFR that may have resulted from applied the exclusion criteria. SS took the lead in designing the data management
procedures and setting up the GitHub repository. LB has developed an interactive
more recent advances in COVID-19 care, most notably,
tool that will be linked to the GitHub repository. SG, AM, GS and RU assisted
the development of novel antiviral medications and with data extraction and verification. ABZ, AM and IK reviewed the methodology
dissemination of vaccines. Of course, the IFR could also and contributed to the discussion of key findings. DH-E, GdlC, ACPA and EBT
shift with the spread of new variants of SARS-CoV-2.66 contributed insights that reflected their experience with health issues in developing
countries. GM-K drafted the main text; NO-B and SP drafted the supplementary
However, given that the first major variant of COVID-19
materials. ATL was responsible for conducting the metaregressions and produced
was only identified in late 2020, and most vaccination all the figures and tables included in the manuscript. ATL, GM-K, NO-B and SP
campaigns in developing nations only began in early edited the text of the manuscript and the supplementary materials.
2021, our time frame limits the impact that these factors Funding The authors have not declared a specific grant for this research from any
should have on the results. funding agency in the public, commercial or not-for-profit sectors.
Map disclaimer The inclusion of any map (including the depiction of any
boundaries therein), or of any geographical or locational reference, does not imply
CONCLUSION the expression of any opinion whatsoever on the part of BMJ concerning the legal
The prevalence and IFR by age of COVID-19 is far higher status of any country, territory, jurisdiction or area or of its authorities. Any such
in developing countries than in high-income countries, expression remains solely that of the relevant source and is not endorsed by BMJ.
Maps are provided without any warranty of any kind, either express or implied.
reflecting a combination of elevated transmission to
middle-aged and older adults, as well as limited access to Competing interests None declared.
adequate healthcare. These results underscore the crit- Patient and public involvement Patients and/or the public were not involved in
the design, or conduct, or reporting, or dissemination plans of this research.
ical need to accelerate the provision of vaccine boosters
and newer effective medications to vulnerable popu- Patient consent for publication Not applicable.
lations in developing countries. Moreover, many devel- Ethics approval This study exclusively used publicly available aggregate data sets
oping countries require ongoing support to upgrade the and published research, and hence no ethics approval was required.
quality of their vital statistics systems to facilitate public Provenance and peer review Not commissioned; externally peer reviewed.
health decisions and actions, not only for the COVID-19 Data availability statement Data are available in a public, open access
pandemic but for future global health concerns. repository. https://covid-ifr.github.io/
Supplemental material This content has been supplied by the author(s). It has
Code and data not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been
All data and code are available publicly online.67 peer-reviewed. Any opinions or recommendations discussed are solely those
of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and
responsibility arising from any reliance placed on the content. Where the content
Author affiliations includes any translated material, BMJ does not warrant the accuracy and reliability
1
Economics, Dartmouth College, Hanover, New Hampshire, USA of the translations (including but not limited to local regulations, clinical guidelines,
2
National Bureau for Economic Research, Cambridge, Massachusetts, USA terminology, drug names and drug dosages), and is not responsible for any error
3
School of Medicine, Case Western Reserve University, Cleveland, Ohio, USA and/or omissions arising from translation and adaptation or otherwise.
4
Statistics, Colorado State University, Fort Collins, Colorado, USA
5 Open access This is an open access article distributed in accordance with the
Department of Statistics, Colorado State University, Fort Collins, Colorado, USA
6 Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which
School of Social Sciences, University of New South Wales, Sydney, New South
permits others to distribute, remix, adapt, build upon this work non-commercially,
Wales, Australia
7 and license their derivative works on different terms, provided the original work is
Law School, University of Chicago, Chicago, Illinois, USA properly cited, appropriate credit is given, any changes made indicated, and the
8
Harris School of Public Policy, University of Chicago, Chicago, Illinois, USA use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
9
Faculty of Information and Technology, Monash University, Clayton, Victoria,
Australia ORCID iDs
10
Interdisciplinary Centre on Population Dynamics, University of Southern Denmark, Andrew T Levin http://orcid.org/0000-0002-1136-6900
Odense, Denmark Anup Malani http://orcid.org/0000-0002-2594-5778

BMJ Global Health
Satej Soman http://orcid.org/0000-0001-8450-7025 analysis for the global burden of disease study 2019. The Lancet
Ana Carolina Peçanha Antonio http://orcid.org/0000-0001-6146-1247 2020;396:1204–22.
Gideon Meyerowitz-Katz http://orcid.org/0000-0002-1156-9345 27 Ayoub HH, Mumtaz GR, Seedat S, et al. Estimates of global
SARS-CoV-2 infection exposure, infection morbidity, and infection
mortality rates in 2020. Glob Epidemiol 2021;3:100068.
28 Ioannidis JPA. Infection fatality rate of COVID-19 inferred from
seroprevalence data. Bull World Health Organ 2021;99:19–33.
29 IHME COVID-19 Forecasting Team. Variation in the COVID-19
REFERENCES infection–fatality ratio by age, time, and geography during the pre-
1 Karlinsky A, Kobak D. Tracking excess mortality across countries vaccine era: a systematic analysis. The Lancet 2022.
during the COVID-19 pandemic with the world mortality dataset. 30 Campbell H, Gustafson P. Inferring the COVID-19 infection fatality
Elife 2021;10:e69336. rate in the community-dwelling population: a simple Bayesian
2 Ramachandran S, Malani A. All-Cause mortality during SARS- evidence synthesis of seroprevalence study data and imprecise
CoV-2 pandemic in India: Nationally-representative estimates mortality data. Epidemiol Infect 2021;149:e243.
independent of official death registry. medRxiv 2021;2021. 31 Maeda JM, Nkengasong JN. The puzzle of the COVID-19 pandemic
3 Gilks CF, Crowley S, Ekpini R, et al. The who public-health in Africa. Science 2021;371:27–8.
approach to antiretroviral treatment against HIV in resource-limited
32 Mariam SH. The severe acute respiratory syndrome Coronavirus-2
settings. Lancet 2006;368:505–10.
(SARS-CoV-2) pandemic: are Africa's prevalence and mortality
4 Dye C. Global epidemiology of tuberculosis. The Lancet
rates relatively low? Adv Virol 2022;2022:3387784.
2006;367:938–40.
33 Okonji EF, Okonji OC, Mukumbang FC, et al. Understanding varying
5 Deshmukh Y, Suraweera W, Tumbe C. Excess mortality in India
COVID-19 mortality rates reported in Africa compared to Europe,
from June 2020 to June 2021 during the COVID pandemic: death
Americas and Asia. Trop Med Int Health 2021;26:716–9.
registration, health facility deaths, and survey data. medRxiv
34 Escobar LE, Molina-Cruz A, Barillas-Mury C. Bcg vaccine
2021;2021.
protection from severe coronavirus disease 2019 (COVID-19). Proc
6 The true death toll of COVID-19: World health Organization2021
Natl Acad Sci. 2020;117:17720–6.
7 Mena GE, Martinez PP, Mahmud AS, et al. Socioeconomic status
35 Lewis H, Ware H, Whelan M. SARS-CoV-2 infection in Africa:
determines COVID-19 incidence and related mortality in Santiago,
a systematic review and meta-analysis of standardised
Chile. Science 2021;372:eabg5298.
seroprevalence studies, from January 2020 to December 2021.
8 Wetzler HP, Wetzler EA. COVID-19 excess deaths in the United
medRxiv 2022.
States, New York City, and Michigan during April 2020. medRxiv
36 Wolday D, Gebrecherkos T, Arefaine ZG, et al. Effect of co-infection
2020;2020.
with intestinal parasites on COVID-19 severity: a prospective
9 Modi C, Boehm V, Ferraro S. How deadly is COVID-19? A rigorous
observational cohort study. E Clinical Medicine 2021;39
analysis of excess mortality and age-dependent fatality rates in
37 Stringhini S, Wisniak A, Piumatti G. Repeated seroprevalence of
Italy. medRxiv 2020.
10 Mwananyanda L, Gill CJ, MacLeod W, et al. Covid-19 deaths in anti-SARS-CoV-2 IgG antibodies in a population-based sample
Africa: prospective systematic postmortem surveillance study. BMJ from Geneva, Switzerland. medRxiv 2020.
2021;372:n334. 38 Pollán M, Pérez-Gómez B, Pastor-Barriuso R, et al. Prevalence
11 Gill CJ. Latest data from Lusaka morgue analysis shows spike in of SARS-CoV-2 in Spain (ENE-COVID): a nationwide, population-
COVID-19 deaths. The Conversation 2021. based seroepidemiological study. Lancet 2020;396:535–44.
12 O'Driscoll M, Ribeiro Dos Santos G, Wang L, et al. Age-Specific 39 Living arrangements of older persons around the world: department
mortality and immunity patterns of SARS-CoV-2. Nature of economic and social Affairs. 2019.
2021;590:140–5. 40 Reyes-Vega MF, Soto-Cabezas MG, Cárdenas F, et al. SARS-
13 Brazeau N, Verity R, Jenks S. Report 34: COVID-19 infection fatality CoV-2 prevalence associated to low socioeconomic status
ratio: estimates from seroprevalence. Imperial College London, and overcrowding in an LMIC megacity: a population-based
2020. seroepidemiological survey in Lima, Peru. EClinicalMedicine
14 Chen X, Chen Z, Azman AS, et al. Serological evidence of human 2021;34:100801.
infection with SARS-CoV-2: a systematic review and meta-analysis. 41 Ghosh AK, Venkatraman S, Soroka O, et al. Association between
Lancet Glob Health 2021;9:e598–609. overcrowded households, multigenerational households, and
15 Levin AT, Hanage WP, Owusu-Boaitey N, et al. Assessing the age COVID-19: a cohort study. Public Health 2021;198:273–9.
specificity of infection fatality rates for COVID-19: systematic 42 Malani A, Shah D, Kang G, et al. Seroprevalence of SARS-CoV-2
review, meta-analysis, and public policy implications. Eur J in slums versus non-slums in Mumbai, India. Lancet Glob Health
Epidemiol 2020;35:1123–38. 2021;9:e110–1.
16 Meyerowitz-Katz G, Merone L. A systematic review and meta- 43 Verity R, Okell LC, Dorigatti I. Estimates of the severity of
analysis of published research data on COVID-19 infection-fatality coronavirus disease 2019: a model-based analysis. Lancet Infect
rates. medRxiv 2020. Dis 2020.
17 World economic and financial surveys world economic outlook 44 GBD 2016 SDG Collaborators. Measuring progress and
Database—WEO groups and aggregates Information 2021. projecting attainment on the basis of past trends of the health-
18 Community assessment for public health emergency response related sustainable development goals in 188 countries: an
toolkit: CDC 2019. analysis from the global burden of disease study 2016. Lancet
19 Population-Based age-stratified seroepidemiological investigation 2017;390:1423–59.
protocol for COVID-19 virus infection: World health Organization 45 Collaborators GS, GBD 2017 SDG Collaborators. Measuring
2020. progress from 1990 to 2017 and projecting attainment to 2030 of
20 Gajda M, Kowalska M, Zejda JE. Impact of two different the health-related sustainable development goals for 195 countries
recruitment procedures (random vs. volunteer selection) on the and territories: a systematic analysis for the global burden of
results of seroepidemiological study (SARS-CoV-2). Int J Environ disease study 2017. Lancet 2018;392:2091–138.
Res Public Health 2021;18:9928. 46 Meyerowitz-Katz G, Merone L. A systematic review and meta-
21 Barchuk A, Shirokov D, Sergeeva M, et al. Evaluation of the analysis of published research data on COVID-19 infection fatality
performance of SARS--CoV--2 antibody assays for a longitudinal rates. Int J Infect Dis 2020;101:138–48.
population-based study of COVID--19 spread in St. Petersburg, 47 Cuéllar L, Torres I, Romero-Severson E, et al. Excess deaths reveal
Russia. J Med Virol 2021;93:5846–52. the true spatial, temporal, and demographic impact of COVID-19
22 COVID-19 estimate downloads. Seattle: Institute for health metrics on mortality in Ecuador. medRxiv 2021. doi:10.1101/2021.02.25.21
and Evaluation 2021. 252481. [Epub ahead of print: 01 Mar 2021]. 10.1093/ije/dyab163.
23 Gelman A, Carpenter B. Bayesian analysis of tests with unknown 48 Calderon-Anyosa RJC, Kaufman JS. Impact of COVID-19 lockdown
specificity and sensitivity. medRxiv 2020. policy on homicide, suicide, and motor vehicle deaths in Peru. Prev
24 Team SD. RStan: the R interface to Stan. R package version 2.19.3, Med 2021;143:106331.
2020. 49 Meyerowitz-Katz G, Bhatt S, Ratmann O, et al. Is the cure really
25 Bergeri I, Whelan M, Ware H. Global epidemiology of SARS-CoV-2 worse than the disease? the health impacts of lockdowns during
infection: a systematic review and meta-analysis of standardized COVID-19. BMJ Glob Health 2021;6:e006653.
population-based seroprevalence studies, Jan 2020-Oct 2021. 50 Buonfrate D, Bisanzio D, Giorli G, et al. The Global Prevalence of
medRxiv 2021. Strongyloides stercoralis Infection. Pathogens 2020;9:468.
26 Vos T, Lim SS, Abbafati C. Global burden of 369 diseases and 51 Karlinsky A. International completeness of death registration 2015-
injuries in 204 countries and territories, 1990–2019: a systematic 2019. medRxiv 2021.

BMJ Global Health
52 Ahmed Y. COVID-19 mortality. Twitter, 2021. Available: https:// 79 Mercado-Reyes M-R, Jeadran N, Zapata S. Seroprevalence of
twitter.com/yakob_son/status/1450448132047790082?s=20 Anti-Sars-Cov-2 antibodies in Colombia, 2020: a population-based
[Accessed 06/01/2021]. study. SSRN 2021.
53 Demombynes G. COVID-19 age-mortality curves are Flatter in 80 Alvis Guzman N, De la Hoz Restrepo F, Serrano-Coll H, et al.
developing countries, 2021. World Bank Group. Public Research Using serological studies to assess COVID-19 infection fatality
Working Paper(9313) rate in developing countries: a case study from one Colombian
54 COVAX: with a fast-moving pandemic, no one is safe, unless department. Int J Infect Dis 2021;110:4–5.
everyone is safe, 2020. World Health organization. Available: 81 Acurio-Páez D, Vega B, Orellana D, et al. Seroprevalence of SARS-
https://www.who.int/initiatives/act-accelerator/covax CoV-2 infection and adherence to preventive measures in Cuenca,
55 Keith Collins JH. See how rich countries got to the front of the Ecuador, October 2020, a cross-sectional study. Int J Environ Res
vaccine line, 2021. New York times Public Health 2021;18:4657.
56 COVID-19 Data Explorer. Our world in data, 2021. Available: https:// 82 Informe de Resultados de la Encuesta Nacional de Salud y
ourworldindata.org/explorers/coronavirus-data-explorer [Accessed Nutrición - Continua COVID-19: Encuesta Nacional de Salud y
27/09/2021]. Nutrición. 2020.
57 Ledford H, Maxmen A. African clinical trial denied access to key 83 Sequera Guillermo CA, Margarita S, Cynthia V, et al. Infección
COVID drug Paxlovid. Nature 2022;604:412–3. POR COVID 19: estudio seroepidemiológico de cohorte de base
58 Folkhälsomyndigheten. The infection fatality rate of COVID-19 in poblacional estratificado POR edad en Asunción Y central: Consejo
Stockholm – technical report. Sweden: Public Health Agency of Nacional de Ciencia Y Tecnologia, 2020.
Sweden, 2020. 84 Huamaní C, Velásquez L, Montes S, et al. Population-Based
59 Parrott JC, Maleki AN, Vassor VE, et al. Prevalence of SARS- seroprevalence of SARS-CoV-2 antibodies in a high-altitude setting
CoV-2 antibodies in New York City adults, June-October 2020: a in Peru. medRxiv 2021.
population-based survey. J Infect Dis 2021;224:188–95. 85 Álvarez-Antonio C, Meza-Sánchez G, Calampa C, et al.
60 Campbell H, de Valpine P, Maxwell L. Bayesian adjustment Seroprevalence of anti-SARS-CoV-2 antibodies in Iquitos, Peru
for preferential testing in estimating infection fatality rates, as in July and August, 2020: a population-based study. Lancet Glob
motivated by the COVID-19 pandemic. The Annals of Applied Health 2021;9:e925–31.
Statistics 2022;16:436–59. 86 Díaz-Vélez C, Failoc-Rojas VE, Valladares-Garrido MJ. SARS-CoV-2
61 Campbell H, Gustafson P. Inferring the COVID-19 IFR with a simple seroprevalence study in Lambayeque, Peru. June–July 2020.
Bayesian evidence synthesis of seroprevalence study data and PeerJ;2021:e11210.
imprecise mortality data. medRxiv. 2021. 87 Alemu BN, Addissie A, Mamo G. Sero-Prevalence of anti-SARS-
62 Herrera-Esposito D, de los Campos G. Age-Specific rate of severe CoV-2 antibodies in Addis Ababa, Ethiopia. bioRxiv 2020.
and critical SARS-CoV-2 infections estimated with multi-country 88 Shaweno T, Abdulhamid I, Bezabih L, et al. Seroprevalence of
seroprevalence studies. medRxiv 2021. SARS-CoV-2 antibody among individuals aged above 15 years and
63 Moser W, Fahal MAH, Abualas E. Retrospective mortality and residing in congregate settings in dire Dawa City administration,
prevalence of SARS-CoV-2 antibodies in greater Omdurman, Ethiopia. Trop Med Health 2021;49:55.
Sudan: a population–based cross–sectional survey. medRxiv 89 Ngere I, Dawa J, Hunsperger E, et al. High seroprevalence of
2021. SARS-CoV-2 but low infection fatality ratio eight months after
64 Lapidus N, Paireau J, Levy-Bruhl D, et al. Do not neglect SARS- introduction in Nairobi, Kenya. Int J Infect Dis 2021;112:25–34.
CoV-2 hospitalization and fatality risks in the middle-aged adult 90 Inquérito Sero-epidemiológico de SARS-CoV-2 na Cidade
population. Infect Dis Now 2021;51:380–2. de Maxixe e Vila de Massinga (InCOVID 2020) - Resultados
65 Steinhardt LC, Ige F, Iriemenam NC. Cross-Reactivity of two SARS- Preliminares -. República de Moçambique: Ministério da Saúde.
CoV-2 serological assays in a setting where malaria is endemic. J 2020.
Clin Microbiol. 2021;59:e00514–21. 91 Barry A, Roka J, Talla C. Seroprevalence of Anti-SARS-CoV-2
66 Classification of omicron (B.1.1.529): SARS-CoV-2 variant of antibodies in Senegal: a national population-based cross-sectional
concern. Geneva: World health Organization. 2021. survey, between October and November 2020, 2021. SSRN.
67 Levin A, Owusu-Boaitey N, Pugh S. Data from: assessing the Available: http://dx.doi.org/10.2139/ssrn.3890833
burden of COVID-19 in developing countries: systematic review, 92 Mutevedzi PC, Kawonga M, Kwatra G. Estimated SARS-CoV-2
meta-analysis, and public policy implications. Github Repository infection rate and fatality risk in Gauteng Province, South Africa:
April 12, 2022. [dataset] https://covid-ifr.github.io/ a population-based seroepidemiological survey. Int J Epidemiol.
68 Buenos Aires City seroprevalence: city of Buenos Aires2020. 2021.
69 COVID-19 seroprevalence study carried out by the University of the 93 Sue Aitken JY, Fellows T, Makamadi T, et al. COVID-19
Municipality of Hurlingham. seroprevalence during the second wave of the pandemic in three
70 The first phase of the Serovigilance epidemiological study districts of south africa - preliminary findings: national institute for
concludes: Gobierno municipal de Santa Cruz. 2020. communicable diseases, 2021.
71 Research indicates that 12.5% of the Mato Grosso population 94 Mulenga LB, Hines JZ, Fwoloshi S, et al. Prevalence of SARS-
has already been infected by the coronavirus. Governo de Mato CoV-2 in six districts in Zambia in July, 2020: a cross-sectional
Grosso. 2021. http://www.mt.gov.br/-/15990626-pesquisa-aponta- cluster sample survey. Lancet Glob Health 2021;9:e773–81.
que-12-5-da-populacao-mato-grossense-ja-foi-infectada-pelo- 95 Khalagi K, Gharibzadeh S, Khalili D. Prevalence of COVID-19 in
coronavirus Iran: results of the first survey of the Iranian COVID-19 serological
72 Saúde presents partial data from the Covid-19 seroepidemiological surveillance program. medRxiv 2021.
survey in DF: federal district health DEPARTMENT. 2020. 96 Bellizzi S, Alsawalha L, Sheikh Ali S, et al. A three-phase population
73 Serological surveys show a drop in antibody levels against based sero-epidemiological study: assessing the trend in
Covid-19. Federal University of Latin American integration: federal prevalence of SARS-CoV-2 during COVID-19 pandemic in Jordan.
University of Latin American integration, 2020. One Health 2021;13:100292.
74 AAMd S, Lima-Neto LG, CdMPeSd A. Population-Based 97 Al-Abri SS, Al-Wahaibi A, Al-Kindi H, et al. Seroprevalence of
seroprevalence of SARS-CoV-2 is more than halfway through the SARS-CoV-2 antibodies in the general population of Oman: results
herd immunity threshold in the state of Maranhão, Brazil. medRxiv from four successive nationwide sero-epidemiological surveys. Int
2020. J Infect Dis 2021;112:269–77.
75 Crotti Peixoto A, Marques D, de Castro L, et al. Avaliação dA 98 Merkely B, Szabó AJ, Kosztin A, et al. Novel coronavirus epidemic
prevalência de marcadores virológicos E sorológicos dA infecção in the Hungarian population, a cross-sectional nationwide
pelo SARS-CoV-2 Na população de Pitangueiras. Sao Paulo: survey to support the exit policy in Hungary. Geroscience
inquérito epidemiológico populacional - Relatório Final 2020. 2020;42:1063–74.
76 Hartwig FP, Vidaletti LP, Barros AJD. Combining serological assays 99 Zejda JE, Brożek GM, Kowalska M, et al. Seroprevalence of
and official statistics to describe the trajectory of the COVID-19 Anti-SARS-CoV-2 antibodies in a random sample of inhabitants
pandemic: results from the EPICOVID19-RS study in Rio grande do of the Katowice region, Poland. Int J Environ Res Public Health
Sul (southern Brazil). medRxiv 2021. 2021;18:3188.
77 Tess BH, Granato CFH, Porto Alves MCG. SARS-CoV-2 100 Kshatri JS, Bhattacharya D, Praharaj I, et al. Seroprevalence of
seroprevalence in the municipality of São Paulo, Brazil, ten weeks SARS-CoV-2 in Bhubaneswar, India: findings from three rounds of
after the first reported case. medRxiv 2020. community surveys. Epidemiol Infect 2021;149:e139.
78 PAaG V, Claudia L, Ramirez-Santana G. Seroprevalence, spatial 101 Selvaraju SKM, Thangaraj J, Bhatnagar T, et al. Population-Based
distribution, and social determinants of SARS-CoV-2 in three urban serosurvey for severe acute respiratory syndrome coronavirus 2
centers of Chile. SSRN 2021. transmission, Chennai, India. Emerg Infect Dis. 2021;27:586–9.

BMJ Global Health
102 Sharma N, Sharma P, Basu S. The seroprevalence of severe acute 107 Kar SS, Sarkar S, Murali S, et al. Prevalence and time trend of
respiratory syndrome coronavirus 2 in Delhi, India: a repeated SARS-CoV-2 infection in Puducherry, India, August-October 2020.
population-based seroepidemiological study. Trans R Soc Trop Emerg Infect Dis 2021;27:666–9.
Med Hyg Suppl. 2021. 108 Khan SMS, Qurieshi MA, Haq I, et al. Seroprevalence of SARS-
103 Mohanan M, Malani A, Krishnan K. Prevalence of COVID-19 in rural CoV-2-specific IgG antibodies in Kashmir, India, 7 months after the
versus urban areas in a low-income country: findings from a State- first reported local COVID-19 case: results of a population-based
Wide study in Karnataka, India. medRxiv 2020. seroprevalence survey from October to November 2020. BMJ Open
104 Saple P, Gosavi S, Pawar T, et al. Seroprevalence of anti-SARS- 2021;11:e053791.
CoV-2 of IgG antibody by ELISA: community-based, cross- 109 Malani A, Ramachandran S, Tandel V. SARS-CoV-2 seroprevalence
sectional study from urban area of Malegaon, Maharashtra. J in Tamil Nadu in October-November 2020. medRxiv 2021.
Family Med Prim Care 2021;10:1453–8. 110 Enhanced surveillance on sero-prevalence of SARS-COV-2 in
105 Satpati P, Sarangi S, Gantait K. Sero-surveillance (IgG) of SARS- general population government of Nepal. 2020.
CoV-2 among asymptomatic general population of Paschim 111 Haq M, Rehman A, Ahmad J, et al. SARS-CoV-2: big
Medinipur, West Bengal, India. medRxiv 2020. seroprevalence data from Pakistan-is herd immunity at hand?
106 Banerjee A, Gaikwad B, Desale A, et al. Severe acute respiratory Infection 2021;49:983–8.
syndrome-coronavirus-2 seroprevalence study in Pimpri- 112 Li Z, Guan X, Mao N, et al. Antibody seroprevalence in the
Chinchwad, Maharashtra, India coinciding with falling trend - Do epicenter Wuhan, Hubei, and six selected provinces after
the results suggest imminent herd immunity? Indian J Public Health containment of the first epidemic wave of COVID-19 in China.
2021;65:256–60. Lancet Reg Health West Pac 2021;8:100094.

BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance
Supplemental material placed on this supplemental material which has been supplied by the author(s) BMJ Global Health
Supplementary Appendices: Assessing the Burden
of COVID‐19 in Developing Countries: Systematic
Review, Meta‐Analysis & Public Policy Implications
Andrew Levin1,2,3, Nana Owusu‐Boaitey4, Sierra Pugh5,
Bailey K. Fosdick5, Anthony B. Zwi6, Anup Malani 2,7, Satej Soman8,
Lonni Besançon9, Ilya Kashnitsky10, Sachin Ganesh1, Aloysius
McLaughlin1, Gayeong Song1, Rine Uhm1, Daniel Herrera‐Esposito11,
Gustavo de los Campos12, Ana Carolina Pecanha Antiono13,
Enyew Birru Tadese14, Gideon Meyerowitz‐Katz15,16
1. Dartmouth College, Hanover, USA
2. National Bureau for Economic Research, Cambridge, USA
3. Centre for Economic Policy Research, London, United Kingdom
4. Case Western Reserve University School of Medicine, Cleveland, USA
5. Colorado State University, Fort Collins, USA
6. School of Social Sciences, University of New South Wales, Australia
7. Law School, University of Chicago, Chicago, USA
8. Mansueto Institute for Urban Innovation, University of Chicago, Chicago, USA
9. Faculty of Information and Technology, Monash University, Australia
10. Interdisciplinary Centre on Population, University of Southern Denmark, Denmark
11. Laboratorio de Neurociencias, Facultad de Ciencias, Universidad de la República, Uruguay
12. Department of Epidemiology & Biostatistics, Michigan State University, USA
13. Hospital de Clínicas de Porto Alegre, Porto Alegre, Brazil
14. Ethiopian Public Health Institute, Ethiopia
15. School of Health and Society, University of Wollongong, Australia
16. Western Sydney Local Health District, Sydney, Australia
Levin AT, et al. BMJ Global Health 2022; 7:e008477. doi: 10.1136/bmjgh-2022-008477
Contents
1. Data Collection
a. Systematic review methodology
b. Full inclusion / exclusion criteria
c. PRISMA flow diagram
d. Death data
e. Assay characteristics and seroconversion
f. Covariates
2. Statistical Methodology
a. Adjustment for seroreversion
b. Bayesian model of seroprevalence and IFR
c. National seroprevalence for high‐income countries
d. IFR benchmark for high‐income countries
3. Detailed Results
a. Seroreversion estimates
b. Seroprevalence results
c. Age‐specific IFRs by Location
d. Age‐specific IFRs by Age Cohort
e. Metaregression results
f. Population IFR
g. IFRs from other sources
h. Covariates
i. Out‐of‐sample analysis
1. Data Collection
a. Systematic Review Methodology
Search Procedure
We searched Serotracker using “household and community samples” and “persons living in slums”
in the “demographics” field. We also searched MedRxiv, PubMed, Google Scholar, Biorxiv, SSRN,
Twitter, and the Pan American Health Organization database using the pre‐specified search term
“COVID‐19 seroprevalence”. Then we cross‐checked with recently published systematic reviews of
worldwide seroprevalence (1‐4), while identifying further studies by searching the grey literature
and government websites where appropriate. This included searching using the term “COVID‐19
seroprevalence” on Google in languages on Google Translate such as Portuguese, Spanish, English,
French, German, and Italian, with an additional search for "inquérito sorológico, COVID‐19".
Duplicates were reviewed by authors on Google sheets and resolved independently.
We completed searches on October 22nd, 2020, and at least monthly afterwards until July 14, 2021.
We also performed searches monthly until September 22, 2021 during initial drafting of our paper.
After our completing initial draft, we performed additional searches monthly up to December 17,
2021, which represented the final cut‐off date for studies included in our analysis. Only studies with
results from an official source were included, such as a published paper, pre‐print, presentation by
government officials, or the website of the institution that performed the study. If a press report or
another unofficial source was found, we performed more detailed searches using information from
the unofficial source to find a matching official source. Study authors were contacted by email or
Twitter for further information, when needed. We also ran detailed searches after September 22,
2021 on older studies for which preliminary results were found by September 22, but for which
updates were posted after September 22. We include links to the studies at each location in the
appendix folder of our GitHub repository.
Searches were conducted by one member of the team and then repeated to ensure consistency by
another. Data were similarly extracted by one member then cross‐checked by another. No data
collection was automated. This process was recorded by the team working across regions in Google
sheets. Data collection is more fully described below but included extracting seroprevalence
information from included studies by age where available, as well as death data specific to COVID‐19
from each country/region with valid seroprevalence information. Where serology data was not
evident in publicly available reports, we reached out to researchers and public health officials using
both email and social media. For death data we largely relied on publicly available official reports.
Studies were reviewed by two authors and screened for inclusion. Disagreements were resolved
through discussion between all authors at weekly meetings and via email. Where essential data
were missing despite efforts to access them, we excluded the study from our synthesis, as noted in
supplementary appendix section 1.b. Our aim was to provide the most robust estimate of age‐
specific IFR in developing countries, and thus we considered it inappropriate to rely on potentially
flawed assumptions regarding these studies in our analysis.

1

Exclusion of Convenience Samples
Blood donor studies are widely used as blood donors are a convenient population from which to
draw a population estimate – donors already have blood taken, can be tested easily, and tend to
include people from a relatively wide area (5). However, as has been noted in research prior to the
pandemic, donors are a highly selected sample and donor studies often have a large bias in terms of
estimates of seroprevalence for other diseases (6, 7). Moreover, in many areas, particularly at initial
stages of the pandemic, donating blood was one of few methods available to access a serological
test. It is unclear which direction this bias generally runs, especially considering the dynamics of a
novel pathogen in the community (8).
Residual sera studies examine clinical blood samples taken initially for other reasons. These samples
have an obvious bias in that they are representative of people going to have blood taken for reasons
other than SARS‐CoV‐2 tests, a group that may not be representative of the general population (9).
Bayesian procedures can be used to incorporate studies of convenience samples (such as residual
sera from blood donors or commercial lab tests) in producing estimates of population infection rates
by accounting for uncertainty about the magnitude and direction of bias (10); however, we excluded
such studies from our analysis to avoid introducing these additional sources of bias. Convenience
sampling of such populations may be sufficient for other purposes, but probabilistic selection from a
representative sample frame better facilitates accurate estimation of population‐wide infection
rates (3, 11).
Risk of Bias
In assessing the risk of bias for each location, we considered three specific factors: (1) the serology
study’s rate of non‐response; (2) the risk of bias due to seroreversion if the study used an assay with
high risk of seroreversion but information was not sufficient for adjusting sensitivity accordingly; and
(3) the risk of death undercounting was elevated due to low proportion of well‐certified deaths.
These risk of bias assessments are provided in the appendix folder of our GitHub repository.
Publication Bias
In this context, publication bias in which studies exhibiting certain findings are more likely (or not) to
be published, is very unlikely to have an impact, as studies with both high and low seroprevalence
estimates are of interest to the scientific literature. Consistent with this, in prior work we found no
evidence of publication bias for seroprevalence studies from high‐income countries (12). However,
to mitigate the risk of publication bias influencing our results, we included lengthy searches of grey
literature, following up on media reports of seroprevalence studies to ensure that every age‐
stratified that we were able to identify was in our metasynthesis.
b. Full Inclusion / Exclusion Criteria
We included only studies that met both of the following conditions:
1. Report seroprevalence from a representative sample in developing countries, meaning:
random selection of participants from a sample frame representative of the general
population, such as household sampling, or sampling >50% of the general population by
census (13‐15), conducted in countries classified by the International Monetary Fund as
“Emerging and Developing Economies” (16).
2. Available online and accessible in English, or via translatable text if not in English.
2

For studies with no reported age‐stratified seroprevalence, but sufficient information to otherwise
calculate age‐stratified IFRs, we calculated these IFRs assuming equal seroprevalence across age‐
groups instead of excluding the study. When total sample size and age‐specific seroprevalence were
known, but age‐specific sample sizes were not precisely reported, age‐specific sample size was
imputed based on the age‐distribution of the general population. We excluded:
1. Convenience samples (3), including those utilizing residual sera from clinical specimens and
blood donors (see the “Blood Donors and Residual Sera” section below), dialysis centres,
healthcare workers, and actively recruited participants constituting less than 50% of the
total population sampled (3, 12, 17).
2. Studies sampling a high‐income country, as classified by the International Monetary Fund
(16), or a wealthy micronation such as Andorra or Monaco.
3. Studies in which sampling extended after February 2021, to help mitigate the risk of
seroreversion on longer timeframes (see supplementary appendix section 2.a).
4. If gender ratios were reported and less than 35% of the sample reported as male or female,
in the absence of cited evidence that the study’s gender ratio matched the general
population.
5. Studies that used the Wondfo serology assay, for the reasons discussed in section 2.b below.
6. Studies that did not report the total number of individuals tested or seroprevalence.
7. Studies using serology assays with insufficient data for estimating sensitivity and specificity
from a known number of tested samples.
8. IFR estimate excluded if: A) the sampling start‐week or end‐week was not known to allow
for accurate determination of the corresponding number of COVID‐19 deaths, B) test‐
adjusted population‐wide seroprevalence overlapped with 0%, or C) samples were taken
during an accelerating outbreak in which reported COVID‐19 deaths increased by a factor of
three or more from the midpoint date of sampling to 4 weeks later (12).
9. Seroprevalence estimate excluded if both of the following conditions were met: A) IFR
estimate was excluded for other reasons listed above, and B) the study overlapped
geographically with another included study. This geographical exclusion avoided
oversampling the same location (12). IFR estimates that met condition B but not condition A
are discussed in the out‐of‐sample analysis below.
Our “out‐of‐sample” analysis included studies that met at least one of the following conditions:
1. IFR estimate for a location that geographically overlapped with an included study, and thus
inclusion of both estimates risked oversampling the same location. IFR estimates for those
locations are provided in supplementary appendix section 3.i.
2. Zero COVID‐19 deaths reported for the sampled location, making the calculated IFR non‐
robust (18). Consequently, our GitHub repository includes seroprevalence estimates for
these out‐of‐sample locations, but IFR estimates were not computed.
3. The total population from which the sample was drawn was less than 30,000, which may not
reflect the wider population of the region. Consequently, our GitHub repository includes
seroprevalence estimates for these out‐of‐sample locations, but IFR estimates were not
computed.
4. Seroprevalence data from five cities in Pakistan did not become available until after our final
cutoff.
3

c. PRISMA Flow Diagram
4

d. Death Data
Building on our prior work (12), we assess the length of the lag between the midpoint of serology
sampling and the time at which COVID‐19 deaths were reported. The time interval between
symptom onset and death had an interquartile range (IQR) of:
‐ 7 to 22 days for Argentina (19)
‐ 9 to 24 days for Colombia (20)
‐ 10 to 26 days for the Brazilian states of Espírito Santo (21) and Parana (22)
These intervals largely agree with IQRs reported for the USA as:
‐ 9 to 24 days for ages 18‐64
‐ 7 to 19 days for ages >64 (12)
The IQR for the interval between death and official reporting for the USA was 2 to 19 days (12).
This largely matches the interval ranges for Argentina (19), Colombia (20), and Paraguay (23) before
March 2021 when included seroprevalence studies stopped collecting samples (see section 1.b),
as shown below:
Figure A1 – Death Reporting Lags
# Days

Argentina, Colombia, and Paraguay may be outliers with respect to the systematic collection and
publication of vital statistics during the pandemic (24); so other developing countries may have
substantially longer reporting lags that may not be documented in the absence of detailed case data.
The time interval between symptom onset and official death reporting thus appears roughly similar
in developing countries as in our previous analysis of high‐income countries such as the USA (12).
For some study locations, we were able to extract COVID‐19 fatality data from case databases that
specified the actual date of death; in those instances, we used the cumulative number of fatalities as
of two weeks after the midpoint date of serology specimen collection. In other locations, fatality
data was only available from official epidemiological bulletins, in which the official number of
cumulative deaths announced at a particular date reflected reporting lags. In those instances, as in
5

our previous work (12), we extracted death information four weeks after the midpoint of specimen
collection. These timing specifications reflect approximate 95th percentiles as follows:
‐ 2‐week interval between symptom onset and seropositivity,
‐ 4‐week interval between symptom onset and death,
‐ 2‐week interval between death and official reporting.
There is also the question of what is the most appropriate death data to use. In most countries there
are two sets of COVID‐19 deaths: confirmed or suspected. In some countries the government will
also present a third tally of deaths, which is modelled using excess mortality statistics or similar.
Confirmed COVID‐19 deaths may under‐estimate the total number of COVID‐19 deaths due to
insufficient testing (25‐28). This may be detected by comparing reported COVID‐19 deaths with
excess deaths, as reflected in the Peruvian government increasing their tally of reported deaths in a
manner that better approximated total excess deaths (24, 29). Nepal’s government also later
substantially increased their reported tally of COVID‐19 deaths (30).
Problems with death reporting are well‐illustrated by the case of Mexico, a country whose vital
statistics system has notable gaps and which experienced a huge number of COVID‐19 deaths. When
looking at the raw data on individuals, 90% of those who died did not have a date of death entered
into the publicly available data. Previous research also demonstrated that large numbers of people
who died from COVID‐19 in Mexico failed to access a test and thus are not included in the country’s
mortality statistics (28). This means that the reported death data available for Mexico is not
sufficient to derive a high‐quality COVID‐19‐related IFR. We therefore instead used an alternative
official source in Mexico that accounted for this COVID‐19 death under‐estimation (31).
So for the purposes of the primary analysis, we included the confirmed + suspected death figures
where available instead of only confirmed deaths, as confirmed + suspected is the more robust
estimate of reported COVID‐19 deaths in developing countries. Death data were extracted from
national datasets in each country where possible, with alternative sources noted where applicable.
Where death data were not immediately available, we contacted the national or local authority
through email or social media. We also attempted to confirm death data using the most robust
source, and in most cases took the estimate directly from the relevant health authority rather than
data aggregation websites. We include our informal assessment of risk of COVID‐19 death under‐
estimation in the appendix folder of our GitHub repository. This is assessment is based on
percentage of deaths well‐certified in the past decade (32), and on comparison of reported COVID‐
19 deaths to excess deaths.
e. Assay Characteristics and Seroconversion
For the assays used in the serology studies that were included in our analysis, we catalogued the
assay manufacturer’s estimates of sensitivity and specificity as well as third‐party assessments of its
performance characteristics. In addition, we conducted a review to assess serological assays for risk
of seroreversion. This review was restricted to studies that tested the same individuals at two
different time‐points separated by at least two months, or tested individuals at least two months
after their first known positive test for SARS‐CoV‐2. We placed emphasis on commercial assays or
assays used in seroprevalence studies.

The search used the terms “COVID‐19 seroreversion” and “COVID‐19 longitudinal, antibody waning”
in Medrxiv, Biorxiv, Google Scholar, and SSRN. Searches were completed at least monthly from
March 2021 to June 2021, with the final search performed on June 30, 2021. We supplemented this
with seroreversion studies found during searches up to July 14 for seroprevalence studies with
representative sampling, and for which further information was released after July 14 (see
“Systematic Review Methodology”). Finally, we selected the studies that contained information
6

about assays that had been used in the serology studies included in our analysis (as described in the
preceding subsections of this appendix).
Our systemic review of assay characteristics revealed that the Wondfo assay exhibited extreme
variations in test sensitivity across batches, apparently reflecting defects in its manufacturing
process (33‐35); consequently, any serology study which used this assay was excluded from our
analysis.
For seroprevalence to approximate the number of people infected, almost all infected people need
to seroconvert by increasing antibody levels after infection. Studies of large populations suggest that
>85% or >90% of SARS‐CoV‐2‐infected individuals seroconvert by approximately 2 weeks after
infection (36‐39). This increases confidence in the accuracy of seroprevalence‐based infection
estimates that use tests with sufficiently high sensitivity (36, 40, 41). Moreover, >50% of the total
population seroconverted in several locations, which would not occur if a substantial proportion of
infected individuals failed to seroconvert. Table A1 illustrates this with studies reporting >50%
seroprevalence before the onset of widespread SARS‐CoV‐2 vaccination:
These high seroprevalence estimates may shed light on high vs. low herd immunity thresholds (42‐
44). For example, Leticia suffered another wave of SARS‐CoV‐2 infections after reported
seroprevalence of 62%, as did Delhi after reported seroprevalence of 56%, the state of Maranhão
after reported seroprevalence of 40%, and Jordan after reported seroprevalence of 34% (20, 45, 46).
Cross‐reactivity also likely does not account for elevated seroprevalence in many of the regions
listed in table A1, since cross‐reactivity did not significantly reduce test specificity in locations such
as Colombia, Ethiopia, and Iran (47‐49). These high seroprevalence estimates instead imply that the
vast majority of infected individuals seroconverted, increasing the reliability of seroprevalence‐
based infection estimate (50).
Some serological assays exhibit significantly lower specificity in African populations, possibly due to
cross‐reactivity with other pathogens (48, 51). This may contribute to divergent seroprevalence
estimates between two studies performed in Addis Ababa, Ethiopia (49, 52) (see supplementary
appendix section 3.i). However, specificity likely remains high in African populations for many of the
assays used in our included studies (49, 53, 54).
Finally, it should be noted that the Gladen‐Rogan procedure of adjusting for assay specificity and
sensitivity (55)) assumes that those characteristics are precisely known, without accounting for the
uncertainty that comes with inferring characteristics from a limited number of tested samples (56).
By contrast, our statistical model uses Bayesian methods that incorporate this form of uncertainty.

7

Table A1 ‐ Locations with Seroprevalence Exceeding 50%
Region Location Reported seroprevalence Number tested for

seroprevalence estimate
Argentina: Buenos Aires 53.4% 873
(Barrio Padre Mugica) (CI: 52.8 ‐ 54.1%)
Argentina: Metropolitan Area 56.7% 300
of Buenos Aires (CI: 55.8 ‐ 57.6%)
(17 de Noviembre)
Colombia, 10 cities: 53% 1426
Barranquilla (CI: 41 ‐ 65%)
Latin Colombia, 10 cities: Guapi 78% 721
America (CI: 65 ‐ 91%)
Colombia, 10 cities: Leticia 62% 1417
(CI: 51 ‐ 73%)
Colombia, Córdoba: Montería 55.3% 1368
(CI: 52.5 ‐ 57.8%)
Peru: Iquitos; July, 70% (CI: 67 ‐ 73%) 716
August 66% (CI: 62 ‐ 70%) 621
Africa Ethiopia: 54.2% 218
Addis Ketema (CI: 47.5 ‐ 60.7%)
Afghanistan: Kabul 53% (CI: < +/‐6%) ‐
Middle Iran, 18 cities: Qom, 58.5% (CI: 37.2 ‐ 83.9%) 108
East Rasht 72.6% (CI: 53.9 ‐ 92.8%) 99
Iraq: Duhok city 62.6% 743
Bangladesh: Dhaka (“slums”) 74% ‐
India: Delhi 56.1% 28,169
(CI: 55.5 ‐ 56.8%)
India: Hyderabad 54.2% 9363
(CI: 53.2 ‐ 55.2%)
India: Karnataka (urban 53.8% 453
areas) (CI: 48.4 ‐ 59.2%)
South India: Mumbai, 3 “slums” (in: 56.4% 4202
Asia Chembur West, 55.1% (CI: 52.4 ‐ 57.8%) 1511
Dahisar, 51.1% (CI: 46.4 ‐ 55.8%) 570
Matunga) 57.0% (CI: 54.7 ‐ 59.2%) 2121
India: Pune, 5 subwards 51.3% (CI: 39.9% ‐ 62.4%) 1659
(Lohiyanagar‐Kasewadi, 66.4% (CI: 57.8% ‐ 74.1%) 307
Navi Peth‐Parvati, 54.1% (CI: 48.3% ‐ 61.7%) 331
Yerwada) 55.5% (CI: 46.6% ‐ 64.1%) 367
Notes: CI refers to confidence interval. This analysis is restricted to studies with at least 75 people
tested. Links to these studies are provided in the Appendix folder of our GitHub repository.

8

f. Covariates
We extracted data from the most recent year prior to the pandemic, which in most cases was 2019
or earlier. For some estimates such as workforce, we relied on the best available data, some of
which was several years old for some countries.
The covariates are:
1. GDP per capita
2. Healthcare spending
3. GNI per capita
4. Hospital beds per capita
5. Life expectancy at birth
6. Healthy life expectancy at age 60
7. Global health security index
8. Skilled healthcare workers per capita
9. Universal health coverage index
10. % of deaths well‐certified (32)

Briefly, these covariates were chosen because they either relate to the expected quality of the
health system itself (i.e. doctors/nurses per population) or to how likely a country was to be
accurately recording the burden of COVID‐19 (WHO indicators, human development index). We also
included the ratio of life expectancy between age 60 and 20 to account for the potential for
survivorship bias – if there was a significant element of survivorship bias in the countries examined,
we would expect the ratio to be higher as more elderly people survived longer periods in places with
higher mortality in youth.

9

2. Statistical Methodology
a. Adjustment for Seroreversion
Seroreversion occurs when the specific antibodies a serological assay tests decline to below the
assay’s level of detection, preventing the assay from identifying infected individuals. As many studies
conducted in developing countries were performed long after initial COVID‐19 waves passed, the risk
of seroreversion could be high. This could lead to underestimation of the proportion of infected
people and thus unreliable estimates in our computed IFRs. Moreover, any modelling using
assumptions about seroreversion for one serological test would almost certainly lead to errors in
other places as different tests can substantially differ in characteristics (57).
For all other assays that were used in the serology studies included in our analysis, we classified
each assay’s risk of seroreversion (high, medium, or low) based on two sources of data:
 Longitudinal serology studies, in which specimens were collected periodically from a given
sample of individuals over an extended period of time.
 Serology analysis of prior RT‐PCR positive cases, i.e., collection of specimens from individuals
who had previously tested positive for COVID‐19.
Some seroprevalence studies tested a representative sample of the general population, including
those with a prior positive SARS‐CoV‐2 PCR test weeks or months before serological testing, a
previous COVID‐19 diagnosis weeks or months before serology, etc. If many of these prior‐positive
individuals later tested seronegative, then that is unlikely to represent failed seroconversion, as
previously discussed. It instead likely indicates a high risk of seroreversion during the time following
their initial positive test (58). A threshold of <75% sensitivity was selected for this risk of
seroreversion because at least 75% of prior‐positives tested seropositive using the Roche assay that
is at low risk of seroreversion (see Table A2), and the vast majority of sources reported sensitivity of
at least 75% before seroreversion, as shown in the input data of our GitHub repository.
Table A2 indicates our assessment of the seroreversion risk of each assay for which sufficient
information was available from longitudinal data or analysis of prior confirmed RT‐PCR positive
cases. For each assay, this table also shows the locations for which we have estimated IFR from
serology results obtained using that assay.
Although not shown in the table, three of these assays were also used to estimate seroprevalence in
“Sero‐Only” locations where IFR could not be estimated due to lack of corresponding fatality data:
(1) Roche Elecsys (anti‐nucleocapsid) was used in Duhak, Iraq; Hyderabad and Rourkela, India; Gaza
and West Bank, Palestine; and Jourberton, South Africa. (2) Euroimmun IgG was used in Tirana,
Albania; Pune, India. (3) Wantai IgG/IgM was used in Cox’s Bazar Rohingya camps, Bangladesh;
Georgia (4 districts); Malaysia (nationwide); Mongolia (nationwide), Klerksdorp & Pietermaritzburg,
South Africa; and Phuket, Thailand.
For every location for which the assay used in serology was classified as having high risk of
seroreversion, we made corresponding adjustments to the data on assay sensitivity as follows:
 Abbott Architect assay. We used information from prior studies to assess how the sensitivity
of this assay diminishes over time following the onset of infection at each monthly interval
from 0 to 6 months. For each of the seven locations where this assay was used, we
10

computed its weighted sensitivity as of the midpoint date of the serology study, where the
weights were determined by the time path of confirmed SARS‐CoV‐2 cases in that location.
 Other assays with high risk of seroreversion. For each of the three locations that used such
assays, we extracted information about the seropositivity of specimens from individuals with
a prior positive RT‐PCR test. This approach automatically accounts for the variation in time
intervals since infection, because each of these serology studies used a representative
sample of the general population, and is consistent with prior work on how waning of
antibodies reduces the proportion of prior‐positives who test seropositive (58, 59).

Table A2 – Seroreversion Assessments and Sources
Seroreversion Data
Risk Sequential Prior
Category Assay IFR Locations Tests Positives Citations
High Abbott Architect IgG Ethiopia: Dire Dawa X
Hungary (nationwide) X
Bosnia & Herzegovina: X
(37, 60‐70)
Republika Sprska X
India: Chennai, Mumbai, X
Pimpri‐Chinchwad, Srinagar X
ErbaLisa IgG India: Paschim Medinipur X (71, 72)
Zydus Kavach IgG India: Delhi X (71)
Luminex S South Africa: Gauteng X (58)
Moderate Euroimmun IgG Zambia: Lusaka & Ndola X
X (67, 73‐77)
Poland: Katowice X
Roche Elecsys IgG/IgM Brazil: Maranhao, Sao Paulo X
(anti‐nucleocapsid) Chile: 3 urban areas X X
(37, 38, 64,
India: Berhampur, X X
69, 70, 73,
Bhubaneswar, Puducherry X X 78‐83)
Mexico (nationwide) X
Pakistan: Karachi, Lahore X
Low COVIDAR IgG Argentina: Buenos Aires City,
X (84, 85)
Hurlingham
DiaSorin Liaison IgG Brazil: Cuiabá, Mato Grosso, X
(37, 57, 60,
Pitangueiras, Várzea Grande X
69, 70, 86)
Oman (nationwide) X
Genetico CoronaPass
Russia: St. Petersburg X (63, 65)
Total
Ortho Vitros IgG India: Tamil Nadu X (57)
Roche Elecsys IgG/IgM
N/A X (64, 87)
(anti‐spike)
Siemens Advia IgG/IgM Colombia: Barranquilla, X
Bogotá, Bucaramanga, X
(69, 70)
Cali, Cucuta, Ipiales, Leticia, X
Medellin, Villavicencio X
University
Brazil: Rio Grande do Sul X (33, 88)
of Rio de Janeiro
Wantai SARS‐CoV‐2 Total Jordan (nationwide) X
Kenya: Nairobi X
Nepal (nationwide) X (64, 89, 90)
Senegal (nationwide) X
South Africa: Mitchells Plain X
Note: This table shows the seroreversion risk category assigned to each assay for which sufficient
information was available.
11

Assays with medium risk of seroreversion. For locations that used either of these assays,
we extracted information about the seropositivity of specimens from individuals with a prior
positive RT‐PCR test, and we utilized that data if the seropositivity rate was less than 75%
(corresponding to a significant degree of seroreversion in that location.)

Given the seroreversion‐adjusted sensitivity for each location, we imputed the corresponding
number of seropositive specimens that would be obtained using the actual sample size for that
serology study, and then those values serve as inputs to the Bayesian model described below.
This approach is conceptually similar to prior studies that have imputed the number of specimens
and the number of confirmed cases by inverting seroprevalence confidence intervals (10, 91).
Finally, Table A3 lists the assays for which seroreversion could not be assessed due to insufficient
information. For each assay, this table shows the IFR and “Sero‐Only” locations for which we relied
on the baseline characteristics of that assay.
Table A3 – Assays with Unknown Seroreversion
Assay IFR Locations Sero‐Only Locations

Mozambique: Beira, Chókwè,
Abbott PanBio IgG/IgM Mozambique: Maputo Matola, Quelimane, Xai‐Xai
Pakistan: Islamabad
Beijing Kewei IgG/IgM Paraguay: Asuncion & Central Dept.
Bioscience IgG/IgM China: Wuhan, Hubei ex. Wuhan China: 3 provinces
Core Technology IgG Ethiopia: Addis Ababa Ethiopia: 3 towns
Coretest IgG/IgM Peru: Lambayeque
CTK Biotech Onsite IgG/IgM Brazil: Distrito Federal
ECO IgG/IgM Brazil: Pitangueiras
GenBody IgG Dominican Republic: 10 provinces
Healgen IgG/IgM Yemen: Aden
INgezim DR IgG/IgM/IgA Colombia: Córdoba
Karwa Kavach IgG India: Malegaon India: Indore, Jabalpur
Luminex N Dem. Rep. of Congo: Kinshasa
Orient Gene Biotech IgG/IgM Brazil: Iquitos, Loreto
Pishtaz Teb IgG/IgM Iran (nationwide)
Proprietary assay #1 Brazil: Foz do Iguaçu
Proprietary assay #2 Laos: 5 provinces
Proprietary assay #3 Zimbabwe: Budiriro & Highfield
Proprietary assay #4 South Sudan: Juba
Mozambique: Chimoio, Tete,
Qingdao Hightop IgG/IgM
Massinga, Maxixe, Pemba
RightSign IgG/IgM Libya: Benghazi
Shenzhen iFlash IgG India: Tamil Nadu
Ecuador: Cuenca
Standard Q IgG/IgM Peru: Lima & Callao Mozambique: Nampula
Bolivia: Santa Cruz
THSTI IgG India: Karnataka
UNCOV‐40 IgG/IgM Nigeria: Niger State

12

b. Bayesian Model for Estimating Seroprevalence and IFR
Model for COVID‐19 infections
Let 𝑅 ⋆, be the number of individuals who tested seropositive in age group A at location 𝑙, and 𝑛 ,
give the number of individuals tested in that age group for this location. We model the number of
individuals with a positive serology test in the study as
𝑅 ⋆, ∼ Binomial 𝑛 , , 𝑝 , , where (1)
𝑝, sens 𝜋 , 1 spec 1 𝜋 , . (2)
To account for the error rates of the test, the test positivity probability, 𝑝 , , is defined as a function
of test sensitivity (sens ), test specificity (spec ), and the true seroprevalence (𝜋 , ) for the
associated location and age group at the time of the study. For many studies, we did not have
seropositivity by age, in which case A represented all ages.
To account for uncertainty in the test characteristics, we model the lab validation data directly. Let
nsens,t denote the number of positive specimens tested with test t, and xsens,t the number of positive
specimens that correctly tested positive. Similarly, let nspec,t and xspec,t denote the number of negative
specimens tested and the number of negative specimens that correctly tested negative with test t,
respectively. We model these quantities as follows:
xsens,t ∼ Binomial(nsens,t, senst) (3)
xspec,t ∼ Binomial(nspec,t, spect). (4)
Model for COVID‐19 deaths
Let 𝐷 ⋆, give the number of recorded COVID‐19 deaths, for age group A at location l. Note that if only
a single death record is available, then A represents the entire age range. We model the recorded
COVID‐19 deaths as
𝐷 ⋆, ∼ Poisson 𝑁 , 𝜋, IFR , (5)
where 𝑁 , gives the number of individuals at location 𝑙 in age group A. Then 𝑁 , 𝜋 , gives the

expected number of infected individuals, and IFR , is the infection fatality rate for location l and age
group A, representing the probability an individual dies from COVID‐19, given the individual had
COVID‐19. Note, the Poisson distribution reflects the relative rarity of a COVID‐19 fatality relative to
the entire population.
Accounting for data collected in varying age bins
Notice that the models above for deaths and infections in (1) and (5) are functions of prevalence and
IFR, respectively, defined on discrete age bins. However, the discrete age bins are not necessarily the
same for the death data and the seroprevalence studies. The following adjustments were made to
match serology and death age bins:
• Death bins nested within a serology bin: We aggregate deaths for each location to match the
respective serology age bins to avoid placing assumptions about the variability of prevalence
across ages within a single serology age bin.
• Serology bins nested within a death bin: The average seroprevalence for the death age bin is
calculated as an average of the serology age bins, weighted by the percent of the population in
each age bin.
13

• Bin endpoints slightly off: When age bins were within one or two years of matching, serology
age bins were adjusted to match the corresponding death age bins.
All modifications to age bins are documented in a spreadsheet in the data folder.
Population Age Distribution
Let 𝑓 𝑎 denote the number of individuals of age a at location 𝑙 for a ∈ (0,1,...,84+). Note, if
population age structure is only available in 5 year age bins, then define
𝑓 𝑏, 𝑏 5
𝑓 𝑎 𝐼 , 𝑎
5
∈ , ,…,
where 𝑓 𝑏, 𝑏 5 is the proportion of the population ages 𝑏, 𝑏 5 .
In cases where the location specific age structure is only available in large bins, but the national age
structure is available in 5 year age bins, we leverage the national age structure to inform the location
specific age structure as follows. Let A denote an interval the location specific age structure is
available for (e.g., [0,18)). If f(A) is the proportion of the population at location 𝑙 with an age in A and
fn(a) is the proportion of the population aged a at the national level, then we estimate 𝑓 𝑎 , the
proportion at location l that is age a, as
𝑓 𝑎 𝑓 𝐴 ∑ ∈ ∩
. (6)
Essentially, we rescale fn(a) such that the total mass in A matches the observed total mass in A at
location 𝑙, f(A). Since we model seroprevalence as constant past age 85, we let 𝑓 85 represent the
proportion of the population aged 85 or older, rather than just the proportion aged 85.
Calculating Average Seroprevalence within a Death Age Bin
Define the population age density for age bin A as
𝑓, 𝑎 ∑ ∈ ∩
, 𝑎 ∈ 0,1, … ,84 (7)
in order to truncate 𝑓 𝑎 to age bin A.
The prevalence for age bin 𝐵 ∪ ∈𝒜 𝐴 is then defined
𝜋, ∑ ∈𝒜 𝜋, ∑ ∈ ∩ 𝑓 , 𝑏 . (8)
For the locations where we only have serology study information with no corresponding fatality
data, the proportion of all study participants that were in a given age bin was assumed
representative of the proportion of the population in each age bin since the studies were designed
to have representative samples. For the locations with both serology and fatality data, population
data were recorded in the Population Distributions tabs with citations.

14

Priors
Because there are infinitely many combinations of prevalence, sensitivity, and specificity that can
result in the same test positivity rate, we used weakly informative priors for the seroprevalence
parameters and informative priors for sensitivity and specificity to avoid a multimodal posterior,
similar to Gelman and Carpenter (2020). For the seroprevalence parameters, 𝜋 , , we used
independent, weakly informative priors:
𝜋 , ∼ Beta 2,6 for all l, 𝐴. (9)
These priors assume a mode around 0.15 with a prior probability of 0.8 that 𝜋l,𝐴 falls between
0 and 0.8.
We also used independent priors for the test sensitivities and specificities. For each test assay t, the
priors on the sensitivity and specificity were
senst ∼ Beta(10,1) (10)
spect ∼ Beta(50,1). (11)
To further narrow the seroprevalence, sensitivity, and specificity combinations, we used
independent, mildly informative priors for each IFR parameter based on expert knowledge. IFR for
COVID‐19 is known to increase with age. We also expect IFR to be more extreme (smaller than
average or larger than average) when the age bin is small. For example, we would expect an age bin
from 20‐80 to look similar to the country average, but we would expect an age bin from 70‐80 to be
much higher than the country average. To formulate a prior that reflects these characteristics, we
modeled
prior
IFR , ∼ Beta 1, IFR , (12)
where
prior
IFR , 30 20 ,
1 , ,
. (13)
The lower and upper bounds of age bin A at location 𝑙 are given by 𝐿 , and 𝑈 , , respectively. For

prior prior
open‐ended upper ages, we set 𝑈 , 100. As an example, IFR , , 30, while IFR , , 14,
allowing for larger IFR estimates when focusing on the older individuals.
Model Implementation
The model was implemented in version 4.0.2 of the programming language R, and posterior samples
were obtained via the software package Stan (version 2.21.1). We ran three chains for 10,000
iterations, where the first 5,000 iterations were discarded as warm‐up samples. All parameters had
an effective sample size greater than 1,200. Additionally, the 𝑅 value was within 0.0016 of 1 for each
parameter, suggesting convergence. Examination of traceplots also suggested convergence.
Out‐of‐sample observation were run as a separate model, so information on test sensitivity and
specificity was not pooled between in‐sample observations and out‐of‐sample observations.
Traceplots, effective sample size (minimum of 2100), and 𝑅 values (within 0.0029 of 1) suggested
convergence of the out‐of‐sample model as well.
We compared plugin estimates for parameters to the posterior distribution for each parameter to
check model fit. In each case there was good agreement, or the Bayesian estimate was superior. For
example, in Figure A2, we compare the posterior distribution of the sensitivity and specificity
estimates to the raw estimate. In most cases, the middle 50% of the posterior distribution contains
15

the raw estimate. However, for test kit ID 11 (Qingdao Hightop Biotech IgM/IgG Duo), the raw
estimate of specificity is outside the range of the posterior draws. In the case of this test, it was used
in locations with extremely low prevalence, such that the Gladen‐Rogan (55) adjustment results in a
negative estimate of prevalence (meaning the expected number of false positives is greater than the
number that tested positive.) Since this is unreasonable, the Bayesian model raises the specificity
estimate, lowering the expected number of false positives.

Figure A2 – Sensitivity and Specificity Posterior Distributions

A: Sensitivity B: Specificity
1.00
1.0
0.95
0.8
0.90
Sensitivity
Specificity
0.85
0.6
0.80
0.4
0.75
0.70
0.2
2 5 7 12 17 22 27 31 35 41 45 48 52 119 182 2 5 7 12 17 22 27 31 35 41 45 48 52 119 182
Test kit ID Test kit ID

Note: These boxplots show the posterior distribution of each assay’s sensitivity (panel A) and
specificity (panel B) compared to the raw estimate (red dot) based on lab validation data. Further
details (including the name of each assay) are given in the appendix folder of our GitHub repository.

Model Outputs
For each model parameter, we use the posterior mean as the point estimate and produce 95%
equal‐tail credible intervals to describe uncertainty.
Total seroprevalence
Similar to calculating the average seroprevalence for a death age bin, we estimate total
seroprevalence for a location by taking an average of the age bin seroprevalences, weighting by the
population distribution at that location:
𝜋, , ∑ ∈𝒜ℓ 𝜋, ∑ ∈ ∩ 𝑓, 𝑏 (14)
where 𝒜ℓ are the serology age bins associated with location l.
Assessing Uniformity of Seroprevalence Across Age
We calculated total seroprevalence for younger adults and middle aged adults, compared to older
adults. The age bins used for each location were selected as follows:
• Younger adults (approximately 18 to 59): Any age bins such that 15 ≤ lower age < 60 and 20 <
upper age ≤ 65 were included
16

• Middle aged adults (approximately 40 to 59): Any age bins such that 40 < lower age < 60 and
40 < upper age ≤ 65 were included
• Older adults (approximately 60 and older): Any age bins such that 60 ≤ lower age were
included.
This resulted in sets of bins where the 18‐59 bins and the 40‐59 bins did not overlap the 60+ bins.
We then calculated total seroprevalence from age a to b using appropriate age bins, 𝒜, similar to
equation (8)
𝜋, ∑ ∈𝒜 𝜋 , ∑
,
(15)
∈𝒜 ,
where ∑ ,
estimates the percent of the total population in that age bin, assuming
∈𝒜 ,
representative age distributions in the serology studies.
For each draw from the posterior distribution, we calculated 𝜋 , for each of our three age
intervals of interest. We then calculated the ratios , and ,
for each draw.
, ,
Total IFR and Comparison to High‐income (EJE) Prediction
Total IFR
Suppose location 𝑙 has death age bins 𝒜ℓ . Let ∑ ∈ ∩ 𝑓, 𝑏 pop , for 𝐴 ∈ 𝒜ℓ . Then
number of deaths
IFRtotal
number of infections
∑ ∈𝒜ℓ IFR , 𝜋, pop ,

∑ ∈𝒜ℓ 𝜋 , pop ,
pop ,
∑ (16)
,
∈𝒜ℓ IFR , ∑ ∈𝒜ℓ pop ,
,
estimates the total IFR for location l. By calculating IFRtotal for each posterior sample, we can then
obtain posterior mean and credible intervals for IFRtotal.
High‐income country benchmark
We compare the IFR estimate to a high income country benchmark based on results from Levin et.
al. (4) which found a log‐linear relationship between age and IFR. Define
10 . .
𝑑𝑎 if 𝑎 85
HICB . (17)
10 . .
if 𝑎 85
Then HICBa represents the IFR predicted by the high‐income countries line averaged over the interval
[a,a+ 1) for ages less than 85 and assumes the high‐income countries line flattens out and becomes
uniform for ages 85 and older.
Then if we assume uniform prevalence for a location, the total IFR estimate over age bin A for high‐
income countries is
∑ ∈ EJE
HICB ∑ ∈
. (18)
17

Subsetting to ages 18‐65
To estimate the IFR between ages 18 and 65, we used the same strategy in picking age bins as we did
when testing uniform prevalence. That is, we selected ℬℓ to be the death age bins in 𝒜 such that the
lower age of the bin is greater than or equal to 18 and the upper age of the bin is less than 66. We
then applied equation (16), replacing 𝒜ℓ with ℬℓ . We were not able to calculate the IFR between 18
and 65 for locations there were no age bins in ℬℓ .
Baseline population
In order to compare the impact of the age specific IFR while controlling for population age
distribution, we calculated the Total IFR substituting 𝑓 𝑎 for a baseline population age distribution,
𝑓 ̅ 𝑎 in (16). We calculated 𝑓 𝑎 following (7). The baseline population was calculated as a median
across locations for each age, then rescaled to sum to one:
𝑓⋆ 𝑎 median 𝑓 𝑎 ∣ l is one of the observed locations with fatality data (19)
𝑓̅ 𝑎 ∑
⋆
. (20)
⋆
We also considered taking the mean across locations for each age and taking the mean across
locations for each age after removing the top five and bottom five locations for that age (a censored
mean). All three approaches gave similar values as shown in Figure A3.
Baseline age distribution
Various estimates of baseline age distribution (mean, median, and censored mean) plotted on age
distribution for observed locations in grey:
Figure A3 – Estimates of Baseline Age Distribution
Mean
Median
Censored Mean
3.0
2.5
2.0
% Population
1.5
1.0
0.5
0.0
0 20 40 60 80
Age

Country average
To get an average total IFR estimate for each country, we took a weighted average of the total IFR
estimate of the locations within that country. We chose to weight by 1/√𝑛 in order to give more
weight to locations with more certain seroprevalence estimates. In locations with multiple age bins,
we took the average across 𝑛 , as 𝑛 . We weighted by certainty in the seroprevalence estimates
rather than the IFR estimates because larger IFR estimates tend to be due to small seroprevalence
18

estimates and consequently have more uncertainty (i.e., small differences in the denominator,
seroprevalence, can result in large changes in the IFR estimate when seroprevalence is small). We
did not want to bias the average by down weighting all of the larger IFR estimates.
We followed the same process to estimate the country average IFR between 18 and 65, with the
added step of removing any locations in the country where ℬℓ was empty.
19

c. National Serology Studies of High‐Income Countries

As shown in Table A4, many high‐income countries succeeded in limiting SARS‐CoV‐2 transmission
during 2020 and early 2021 (92‐94). Japan and South Korea (as well as several other East Asian
countries) were particularly successful at limiting infection rates (95, 96). Some subnational high‐
income country locations reported higher seroprevalence, e.g., about 21% in New York City, USA
(97), up to 25% in some Swiss cantons (59), and 42% at an Austrian ski area (18). However, the
available serology data, based on representative samples of the general population, indicates that
no location in any high‐income country experienced non‐vaccine‐induced seroprevalence above 45%
prior to March 2021.
Table A4 –National Seroprevalence Estimates for High‐Income Countries
Timeframe Location Reported seroprevalence Midpoint date
April 2020 to Slovenia 0.9% (CI: 0.4‐1.4%) April 25
Sept. 2020 Spain 5.0% (CI: 4.7‐5.4%) May 4
France 4.5% (CI: 3.9 ‐ 5.0%) May 17
Italy 2.5% (CI: 2.3 ‐ 2.6%) June 19
United Kingdom 6.0% (CI: 5.8 ‐ 6.1%) July 1
Canada 1.9% (CI: 1.4 ‐ 2.0%) July 15
South Korea 0.01% July 19
Germany 0.7% July 21
Denmark 2.0% (CI: 1.7‐2.4%) Sept. 19
Netherlands 4.7% (CI: 4.0‐5.5%) Sept. 28
Oct. 2020 to USA 11.9% (CI: 10.5 ‐ 13.5%) Oct. 30
January 2021 England 5.6% (CI: 5.4‐5.7%) Nov. 3
Germany 1.1% (CI: 0.9 ‐ 1.3%) Nov. 6
Slovenia 4.1% (CI: 3.0 ‐ 5.2%) Nov. 11
Austria 4.7% (CI: 3.8 ‐ 5.6%) Nov. 13
South Korea 0.1% Nov. 21
Spain 9.9% (CI: 9.4 ‐ 10.4%) Nov. 22
France 6.2% (CI:5.9‐6.6%) Nov. 26
Denmark 4.1% (CI: 3.1 ‐ 4.9%) Dec. 16
Norway 0.9% (CI: 0.7 ‐ 1.0%) Jan. 5

20

Table A5 – Variants of Benchmark Metaregression for High‐Income Countries
Description # Observations Intercept Slope Coefficient
‐3.27 0.0524
Benchmark 104
(0.073) (0.0013)
Exclusion of ‐3.32 0.0532
68
Convenience Samples (0.089) (0.0015)
Adjustment for ‐3.22 0.0516

104
Seroreversion (0.070) (0.0012)
Adjustment for Death ‐3.18 0.0526

104
Undercounting (0.075) (0.0013)

d. Metaregression benchmark for high‐income countries
To provide a benchmark for our analysis of IFR in developing countries, we consider the findings
from a prior meta‐analysis of age‐specific IFRs for high‐income countries (12). That study conducted
a metaregression using 104 observations on age‐specific IFRs from 28 locations (using samples
collected between April and July 2020) and obtained the following results:
𝑙𝑜𝑔 𝐼𝐹𝑅 3.27 0.0524 ∗ 𝑎𝑔𝑒

0.07 0.0013
where the standard error for each estimated coefficient is given in parentheses.
To determine whether those results can serve as a suitable benchmark, we must consider several
distinct methodological issues. First, the prior study used serology data from convenience samples as
well as from representative samples of the general population, whereas our present analysis
excludes convenience samples. Second, the prior study computed assay‐adjusted seroprevalence
using the baseline characteristics of each assay, whereas our present analysis incorporates
adjustments for seroreversion over time. Third, the prior study used official reports on confirmed
COVID‐19 deaths, without incorporating any information about underreporting of COVID‐19
fatalities, but subsequent analysis has shown that such underreporting has been substantial in some
locations in high‐income countries.
To assess the significance of these methodological issues, we have replicated the prior
metaregression analysis along with three variants. In the first variant, the metaregression excludes
36 observations from convenience samples. In the second variant, we make seroreversion
adjustments for two locations (Italy and Spain) that utilized the Abbott Architect assay, using the
same approach as in our present analysis described above. In the third variant, we adjust fatalities
using IHME estimates of COVID‐19 death undercounts. Table A5 reports the results of this sensitivity
analysis. Evidently, the results for each variant are nearly identical to those of the prior benchmark
regression, with no statistically significant differences in the estimates of the intercept or slope
coefficient. These results underscore the robustness of this metaregression for high‐income
countries and support its use as a benchmark for assessing IFR in developing countries.

21

3. Additional Results
a. Seroreversion Estimates
Table A6: Seroreversion Adjustments and Assay Sensitivity
Country Location Assay Baseline (%) Adjusted (%) Ratio
Chile 3 urban areas Elecsys 99.5 68.8 0.69
Ethiopia Diredawa Abbott 100 86.8 0.87
Hungary National Abbott 100 87.2 0.87
India Delhi Kawach 92.1 65.2 0.71
Pimpri‐Chinchwad Abbott 100 77.9 0.78
Paschim Medinipur Erbalisa 98.3 36.7 0.37
Chennai Abbott 100 83.2 0.83
Srinagar Abbott 100 78.1 0.78
Mumbai Abbott 100 76.9 0.77
South Africa Gauteng Luminex 100 46.7 0.47
Note: This table shows the characteristics of the assay used in the serology study of each of the
specified locations, including the assay sensitivity at baseline, the seroreversion‐adjusted sensitivity,
and the ratio of adjusted to baseline sensitivity. In denoting these assays, Elecsys refers to the
Elecsys Anti‐SARS‐Cov‐2 Roche assay, Abbott refers to the Abbott Architect IgG assay, Kavach refers
to the Kawach IgG assay, Erbalisa refers to the ErbaLisa IgG assay, and Luminex refers to the Luminex
protein trimer assay.
Table A7: Implications for Seroprevalence and IFR
Seroprevalence (%)
Baseline Seroreversion‐ Seroprevalence IFR
Country Location Sensitivity Adjusted Sensitivity Ratio Ratio
Chile 3 urban areas 10.1 13.8 1.4 0.73
Ethiopia Diredawa 4.4 5.0 1.2 0.87
Hungary National 0.4 0.5 1.4 0.79
India Delhi 31.2 43.2 1.4 0.72
Pimpri‐Chinchwad 32.9 40.7 1.2 0.83
Paschim Medinipur 6.8 12.6 1.9 0.55
Chennai 22.1 26.3 1.2 0.84
Srinagar 40.2 50.4 1.3 0.80
Mumbai 40.1 52.0 1.3 0.78
South Africa Gauteng 18.8 33.2 1.8 0.56
Note: For each location, this table reports the seroprevalence estimate obtained using the baseline
sensitivity of the assay used in that serology study as well as the corresponding estimate obtained
using the seroreversion‐adjusted sensitivity for that assay. The penultimate column shows the ratio
of seroreversion‐adjusted to baseline‐adjusted seroprevalence, while the final column shows the
ratio for the corresponding estimates of population IFR for that location.

22

b. Seroprevalence Estimates
Figure A4 – Population‐wide Seroprevalence
Latin America Location Start Date End Date Estimate (95% CI)
Argentina Buenos Aires Province 07/16/2020 12/01/2020 14.8 (13.6-16.0)
Buenos Aires City 09/10/2020 10/18/2020 11.4 (9.7-13.1)
Hurlingham 11/26/2020 12/10/2020 14.9 (12.5-17.7)
Bolivia Santa Cruz 08/22/2020 09/13/2020 21.3 (18.4-23.9)
Brazil Cuiabá 09/16/2020 10/15/2020 18.2 (14.9-21.0)
Distrito Federal 12/02/2020 12/17/2020 16.1 (12.8-19.1)
Foz do Iguaçu 05/14/2020 06/09/2020 12.8 (9.6-15.3)
Maranhao 07/27/2020 08/08/2020 38.1 (35.7-40.6)
Mato Grosso (8 cities) 09/16/2020 10/15/2020 11.5 (9.6-13.8)
Pitangueiras 08/24/2020 09/29/2020 7.1 (3.1-11.5)
Rio Grande do Sul 02/05/2021 02/07/2021 9.0 (6.8-10.7)
Sao Paulo City 10/01/2020 10/10/2020 26.0 (23.4-28.6)
Várzea Grande 09/16/2020 10/15/2020 27.9 (23.2-33.4)
Chile 3 urban areas 09/26/2020 11/25/2020 13.8 (11.7-16.2)
Colombia Barranquilla 09/20/2020 09/30/2020 48.7 (45.9-51.5)
Bogotá 10/10/2020 11/05/2020 26.5 (25.1-27.9)
Bucaramanga 09/27/2020 10/09/2020 28.2 (25.9-30.6)
Cali 11/18/2020 11/28/2020 26.5 (24.6-28.6)
Cucuta 10/05/2020 10/15/2020 35.5 (32.9-38.0)
Córdoba 07/01/2020 10/29/2020 37.8 (30.1-43.7)
Ipiales 12/03/2020 12/11/2020 30.4 (28.1-32.9)
Leticia 09/15/2020 09/25/2020 51.1 (48.4-53.9)
Medellín 10/05/2020 12/20/2020 24.0 (22.1-26.0)
Villavicencio 10/20/2020 10/30/2020 29.9 (27.6-32.3)
Dominican Rep. Ten Provinces 04/01/2020 06/30/2020 7.6 (4.7-10.5)
Ecuador Cuenca 08/11/2020 11/01/2020 12.0 (10.1-13.9)
Mexico National Study 08/18/2020 11/13/2020 26.2 (25.1-27.3)
Paraguay Asunción + Central Dept. 12/23/2020 02/16/2021 21.3 (15.2-26.6)
Peru Cusco Province 09/12/2020 09/27/2020 36.7 (34.3-39.3)
Iquitos 07/13/2020 07/18/2020 65.1 (61.0-69.1)
Lambayeque 06/24/2020 07/10/2020 34.6 (30.9-38.0)
Lima + Callao 06/28/2020 07/09/2020 21.6 (19.7-23.4)
Europe
Albania Tirana City 12/21/2020 12/28/2020 51.7 (46.8-57.7)
Bosnia & Herzegovina Republika Srpska 11/04/2020 12/16/2020 39.7 (37.2-42.4)
Georgia 4 districts 12/01/2020 12/14/2020 44.3 (41.3-47.5)
Hungary National Study 05/01/2020 05/16/2020 0.5 (0.2-0.7)
Poland Katowice Region 10/01/2020 11/30/2020 12.2 (10.0-14.6)
Russia St. Petersburg 05/25/2020 06/28/2020 9.9 (7.1-12.6)
Africa
Cameroon Cité Verte 10/14/2020 11/26/2020 31.8 (28.8-35.2)
Congo (Dem. Rep.) Kinshasa 10/22/2020 11/08/2020 15.6 (13.5-17.9)
Ethiopia Addis Ababa 07/22/2020 08/10/2020 3.4 (2.2-4.8)
Dire Dawa 06/15/2020 07/30/2020 5.0 (2.9-7.6)
Jimma 08/19/2020 09/02/2020 0.7 (0.2-1.4)
Kenya Nairobi County 11/02/2020 11/23/2020 32.6 (29.8-35.6)
Mozambique Beira 09/21/2020 10/02/2020 5.5 (4.7-6.3)
Chimoio 11/02/2020 11/17/2020 0.5 (0.3-0.8)
Chókwè 11/16/2020 11/21/2020 5.5 (4.5-6.6)
Maputo 08/03/2020 08/21/2020 3.8 (3.1-4.5)
Massinga 11/26/2020 12/03/2020 7.3 (5.9-8.9)
Matola 09/14/2020 09/30/2020 3.6 (3.0-4.2)
Maxixe 11/07/2020 11/21/2020 5.0 (4.2-5.9)
Nampula 06/17/2020 06/30/2020 4.3 (2.8-5.7)
Pemba 07/06/2020 07/13/2020 2.6 (1.6-3.8)
Quelimane 08/10/2020 08/21/2020 3.0 (2.4-3.7)
Tete 08/31/2020 10/12/2020 0.9 (0.4-1.5)
Xai-Xai 11/02/2020 11/12/2020 3.7 (3.1-4.4)
Nigeria Niger State 06/26/2020 06/30/2020 24.7 (19.2-30.6)
Senegal National Study 10/24/2020 11/26/2020 27.5 (24.8-30.2)
South Africa Gauteng 11/04/2020 01/22/2021 33.2 (24.4-40.5)
Jouberton 07/20/2020 09/17/2020 15.5 (12.5-18.6)
Klerksdorp 11/23/2020 01/31/2021 27.0 (20.8-33.8)
Mitchells Plain 12/08/2020 01/31/2021 46.1 (40.6-51.8)
Pietermaritzburg 11/23/2020 01/31/2021 32.4 (27.9-37.2)
South Sudan Juba 08/10/2020 09/11/2020 32.5 (28.8-36.0)
Zambia Copperbelt 07/04/2020 07/27/2020 2.7 (1.0-3.8)
Zimbabwe Harare 11/20/2020 12/20/2020 18.8 (15.9-21.9)
Middle East
Iran National Study 08/03/2020 10/31/2020 13.0 (11.0-14.9)
Iraq Duhok City 01/10/2021 01/30/2021 62.8 (59.1-66.5)
Jordan National Study 12/27/2020 01/06/2021 34.6 (32.9-36.6)
Libya Benghazi 05/02/2020 05/07/2020 0.3 (0.0-0.0)
Oman National Study 07/12/2020 07/19/2020 5.2 (4.3-6.3)
Palestine Gaza 12/01/2020 12/31/2020 42.0 (40.0-44.1)
West Bank 12/01/2020 12/31/2020 39.5 (37.8-41.2)
United Arab Emirates Abu Dhabi 07/19/2020 08/14/2020 8.9 (7.9-10.2)
Yemen Aden 11/29/2020 12/31/2020 26.2 (22.6-29.4)
South Asia
Bangladesh Rohingya Camp 12/01/2020 12/30/2020 61.1 (59.0-63.8)
India Chennai 07/17/2020 07/28/2020 26.3 (24.9-27.7)
Delhi 08/01/2020 08/07/2020 43.2 (39.8-46.9)
Hyderabad 01/08/2021 01/24/2021 54.5 (53.2-56.3)
Indore City 08/11/2020 08/23/2020 9.2 (7.1-10.7)
Jabalpur City 12/11/2020 12/21/2020 37.0 (34.5-39.6)
Karnataka 06/15/2020 08/29/2020 43.4 (39.1-47.9)
Srinagar District 10/17/2020 10/20/2020 50.4 (47.1-53.0)
Malegaon 07/25/2020 08/20/2020 41.0 (34.1-48.2)
Mumbai 06/29/2020 07/19/2020 52.0 (49.3-54.0)
Berhampur 08/06/2020 08/06/2020 29.4 (24.2-35.4)
Bhubaneswar 07/10/2020 07/10/2020 2.2 (1.2-3.7)
Rourkela 08/31/2020 08/31/2020 24.0 (21.8-26.3)
Paschim Medinipur 07/27/2020 08/07/2020 12.6 (5.6-22.3)
Pimpri-Chinchwad 10/07/2020 10/17/2020 40.7 (38.3-43.2)
Puducherry 09/10/2020 09/16/2020 20.5 (17.3-24.1)
Pune 07/20/2020 08/05/2020 54.8 (50.6-60.1)
South Andaman 12/01/2020 02/28/2020 39.0 (36.0-42.0)
Tamil Nadu (Vitros) 10/19/2020 11/30/2020 25.9 (24.8-27.3)
Tamil Nadu (iFlash) 10/19/2020 11/30/2020 43.9 (41.5-46.4)
Uttar Pradesh 09/04/2020 09/10/2020 26.2 (24.1-28.2)
Nepal National Study 10/09/2020 10/22/2020 13.9 (12.5-15.3)
Pakistan Islamabad 06/01/2020 06/30/2020 16.8 (15.6-18.1)
Karachi 07/15/2020 07/31/2020 38.3 (29.0-48.2)
Lahore 07/15/2020 07/31/2020 46.2 (43.7-48.9)
East Asia
China Hubei excluding Wuhan 04/10/2020 04/18/2020 0.7 (0.5-0.9)
Six Provinces 04/10/2020 04/18/2020 0.1 (0.1-0.2)
Wuhan 04/10/2020 04/18/2020 4.1 (3.6-4.6)
Laos 5 provinces 08/12/2020 09/25/2020 4.8 (2.8-7.1)
Malaysia National Study 08/07/2020 10/06/2020 0.6 (0.3-0.9)
Mongolia National Study 10/13/2020 12/04/2020 1.7 (1.4-2.2)
Thailand 5 cities 12/18/2020 02/02/2021 0.9 (0.4-1.5)
0 10 20 30 40 50 60 70 80
Percent

Notes: The green shading represents the range of national seroprevalence for high‐income countries
in Table A4. Links to these studies are in the Appendix folder of our GitHub repository.

23

Figure A5 – Ratio of Seroprevalence for Older Adults (60+ years) Compared to
Younger Adults (18‐59 years)
Latin America Location Ratio (95% CI)

Argentina Buenos Aires City 1.0 (0.7-1.4)
Hurlingham 0.8 (0.6-1.1)
Brazil Cuiabá 0.8 (0.6-1.2)
Maranhao 0.9 (0.8-1.1)
Mato Grosso (8 cities) 0.7 (0.5-1.0)
Sao Paulo City 0.9 (0.7-1.2)
Várzea Grande 0.4 (0.3-0.6)
Chile 3 urban areas 0.6 (0.4-0.8)
Colombia Barranquilla 0.9 (0.7-1.0)
Bogotá 0.6 (0.5-0.7)
Bucaramanga 0.8 (0.6-1.0)
Cali 0.7 (0.6-0.9)
Cucuta 0.8 (0.6-1.0)
Córdoba 1.0 (0.9-1.2)
Ipiales 0.7 (0.5-0.9)
Leticia 0.8 (0.7-1.0)
Medellín 0.6 (0.5-0.8)
Villavicencio 0.7 (0.6-0.9)
Ecuador Cuenca 1.0 (0.7-1.3)
Mexico National Study 0.6 (0.5-0.7)
Paraguay Asunción + Central Dept. 0.8 (0.5-1.2)
Peru Iquitos 1.1 (0.9-1.3)
Lambayeque 0.7 (0.6-0.9)
Lima + Callao 0.8 (0.7-1.0)
Europe
Bosnia & Herzegovina Republika Srpska 0.6 (0.5-0.8)
Georgia 4 districts 0.9 (0.8-1.1)
Hungary National Study 1.9 (0.7-4.7)
Poland Katowice Region 0.4 (0.2-0.7)
Russia St. Petersburg 0.7 (0.2-1.5)
Africa
Cameroon Cité Verte 1.1 (0.7-1.6)
Kenya Nairobi County 0.6 (0.3-0.9)
Mozambique Beira 0.6 (0.3-1.0)
Chimoio 3.1 (0.7-8.1)
Chókwè 0.8 (0.3-1.5)
Maputo 1.2 (0.8-1.8)
Massinga 0.8 (0.3-1.4)
Matola 1.0 (0.6-1.5)
Maxixe 0.8 (0.4-1.4)
Nampula 0.9 (0.2-2.0)
Pemba 2.0 (0.4-5.2)
Quelimane 0.4 (0.1-1.0)
Tete 3.8 (0.4-13.0)
Xai-Xai 1.2 (0.7-1.8)
Nigeria Niger State 0.9 (0.2-1.9)
Senegal National Study 0.8 (0.6-1.1)
South Africa Gauteng 1.0 (0.8-1.2)
Jouberton 0.4 (0.1-0.8)
South Sudan Juba 1.2 (0.7-1.7)
Zimbabwe Harare 0.7 (0.3-1.3)
Middle East
Iran National Study 1.4 (1.1-1.6)
Jordan National Study 1.0 (0.9-1.1)
Palestine Gaza 0.8 (0.7-1.0)
West Bank 1.0 (0.8-1.1)
United Arab Emirates Abu Dhabi 0.4 (0.2-0.8)
Yemen Aden 0.9 (0.6-1.2)
South Asia
Bangladesh Rohingya Camp 1.0 (0.9-1.1)
India Chennai 0.7 (0.6-0.8)
Hyderabad 0.9 (0.8-1.0)
Indore City 1.1 (0.8-1.5)
Jabalpur City 1.0 (0.9-1.1)
Karnataka 0.9 (0.7-1.1)
Srinagar District 1.2 (0.9-1.5)
Malegaon 0.9 (0.4-1.5)
Mumbai 0.9 (0.8-1.0)
Berhampur 0.9 (0.7-1.1)
Bhubaneswar 1.2 (0.3-2.8)
Rourkela 1.2 (1.0-1.5)
Paschim Medinipur 1.7 (0.3-4.9)
Pimpri-Chinchwad 0.9 (0.8-1.1)
Puducherry 0.8 (0.5-1.1)
Pune 0.7 (0.6-0.9)
South Andaman 0.9 (0.7-1.0)
Uttar Pradesh 1.1 (1.0-1.3)
Nepal National Study 0.9 (0.6-1.1)
Pakistan Karachi 1.2 (0.6-1.8)
Lahore 1.0 (0.8-1.1)
East Asia
China Hubei excluding Wuhan 0.8 (0.3-1.5)
Six Provinces 1.2 (0.1-4.3)
Wuhan 1.2 (0.9-1.5)
Laos 5 provinces 2.8 (1.4-5.6)
Mongolia National Study 1.3 (0.5-2.4)
Thailand 5 cities 1.4 (0.2-4.6)
0 .2 .4 .6 .8 1 1.2 1.4 1.6 1.8 2

Ratio

Note: The green shading represents the range of national seroprevalence for high‐income countries
from our prior work (12). Links to these studies are in the Appendix folder of our GitHub repository.
24

c. Age‐specific IFR curves by location
Figure A6 – Age‐Specific IFR Curves By Location
Percent Buenos Aires City, Argentina Hurlingham, Argentina

10
Percent
10
0.1
0.1
0.01
0.01
0.001 0.001
Posterior Estimate Posterior Estimate
Coverage Interval Coverage Interval
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age

Age
Percent Sao Paulo City, Brazil

Cuiabá, Brazil
Percent 10
10
1
1
0.1
0.1
0.01
0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
Várzea Grande, Brazil

Percent
3 urban areas, Chile
10 Percent
10
1
1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
Barranquilla, Colombia Percent Bogotá, Colombia

Percent 10
10
1
1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
(figure continues on next page)
25

Bucaramanga, Colombia Percent Cali, Colombia

Percent 10
10
1 1
0.1 0.1
0.01 0.01
0.001 0.001

0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age

Percent Córdoba, Colombia Cucuta, Colombia
10
Percent
10
1
1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
Percent Ipiales, Colombia Leticia, Colombia

Percent
10 10
1 1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
Percent Medellín, Colombia Villavicencio, Colombia

Percent
10 10
1 1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
26

Cuenca, Ecuador
Percent Percent Asunción + Central Dept., Paraguay
10 10
1 1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
Percent Iquitos, Peru Percent Lambayeque, Peru

10 10
1 1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
Percent Lima + Callao, Peru

10
Percent Nairobi County, Kenya
10
1
1
0.1
0.1
0.01
0.01
0.001
0.001
Posterior Estimate
Posterior Estimate
Coverage Interval
Coverage Interval
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age
National Study, Jordan Percent National Study, Hungary

Percent 10
10
1
1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age Age

27

Percent Chennai, India Karnataka, India

Percent
10 10
1 1
0.1 0.1
0.01 0.01
0.001 0.001
0.0001 0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85

Age Age
Percent Wuhan, China

10
0.1
0.01
0.001
Posterior Estimate
Coverage Interval
0.0001
0 4 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85

Age
Note: For each location, this figure shows the posterior estimate and 95% credible interval of IFR for each of the age
brackets reported in the serology study of that location; each estimate reflects the reported number of COVID‐19
fatalities for that age bracket in that location and has not been adjusted for death undercounting.
28

d. Age‐specific IFRs by Age Cohort
Figure A7 – IFR Estimates for Children
Country Location Ages Estimate (95% CI)
India Karnataka 10-19 0.000 (0.000-0.001)
India Chennai 0-18 0.001 (0.000-0.001)
Jordan National Study 5-14 0.001 (0.000-0.002)
Colombia Barranquilla 11-18 0.001 (0.000-0.004) IFR Estimate (%) 95% CI
Jordan National Study 0-4 0.002 (0.001-0.004)
Colombia Barranquilla 5-10 0.002 (0.000-0.006)
Paraguay Asunción + Central Dept. 0-19 0.002 (0.001-0.005)
Colombia Cucuta 11-18 0.003 (0.000-0.010)
Colombia Medellín 11-18 0.003 (0.000-0.008)
Colombia Cali 11-18 0.003 (0.000-0.008)
Colombia Bogotá 5-10 0.003 (0.001-0.006)
Colombia Bogotá 11-18 0.003 (0.001-0.006)
Kenya Nairobi County 10-19 0.003 (0.001-0.006)
Brazil Maranhao 10-19 0.003 (0.002-0.005)
Brazil Maranhao 0-9 0.003 (0.002-0.006)
Colombia Córdoba 10-19 0.004 (0.001-0.009)
Colombia Cucuta 5-10 0.004 (0.000-0.016)
Colombia Villavicencio 11-18 0.004 (0.000-0.016)
Colombia Bucaramanga 11-18 0.004 (0.000-0.016)
Chile 3 urban areas 7-14 0.006 (0.002-0.012)
Argentina Buenos Aires City 0-17 0.006 (0.002-0.013)
Peru Lima + Callao 12-17 0.007 (0.004-0.011)
Colombia Cali 5-10 0.007 (0.001-0.018)
Peru Lambayeque 9-20 0.007 (0.003-0.014)
Peru Iquitos 12-17 0.009 (0.000-0.032)
Peru Iquitos 0-11 0.009 (0.001-0.025)
China Wuhan 0-19 0.010 (0.003-0.021)
Peru Lima + Callao 0-11 0.012 (0.009-0.016)
Colombia Leticia 11-18 0.023 (0.001-0.085)
Ecuador Cuenca 0-19 0.023 (0.008-0.048)
Colombia Ipiales 11-18 0.030 (0.001-0.110)
0.000 0.015 0.030 0.045 0.060

Figure A8 – IFR Estimates for Young Adults
India Karnataka 20-29 0.001 (0.001-0.002)
India Karnataka 30-39 0.005 (0.004-0.006)
India Chennai 19-29 0.005 (0.003-0.007)
Brazil Maranhao 20-29 0.006 (0.004-0.009)
IFR Estimate (%) 95% CI
Peru Iquitos 18-29 0.012 (0.001-0.033)
India Chennai 30-39 0.016 (0.013-0.020)
Peru Lambayeque 21-30 0.020 (0.011-0.032)
Brazil Várzea Grande 20-29 0.021 (0.004-0.053)
Brazil Maranhao 30-39 0.033 (0.026-0.041)
Colombia Cali 19-40 0.035 (0.027-0.044)
Colombia Cucuta 19-40 0.037 (0.026-0.050)
Peru Lima + Callao 18-29 0.037 (0.030-0.045)
Colombia Bogotá 19-40 0.037 (0.033-0.043)
Brazil Sao Paulo City 20-34 0.045 (0.036-0.055)
Argentina Hurlingham 18-30 0.045 (0.009-0.116)
China Wuhan 20-39 0.058 (0.043-0.075)
Brazil Cuiabá 20-29 0.058 (0.026-0.106)
Peru Lambayeque 31-40 0.079 (0.057-0.105)
Ecuador Cuenca 20-34 0.082 (0.044-0.136)
Hungary National Study 15-39 0.103 (0.014-0.500)
0.00 0.03 0.06 0.09 0.12

29

Figure A9 – IFR Estimates for Middle‐aged Adults
India Karnataka 40-49 0.01 (0.01-0.01)
India Karnataka 50-59 0.04 (0.03-0.04)
India Chennai 40-49 0.05 (0.04-0.06)
Jordan National Study 15-64 0.07 (0.06-0.07) IFR Estimate (%) 95% CI
Brazil Maranhao 40-49 0.08 (0.06-0.09)
India Chennai 50-59 0.17 (0.15-0.19)
Brazil Maranhao 50-59 0.22 (0.19-0.26)
Colombia Cali 41-60 0.25 (0.21-0.30)
Peru Lambayeque 41-50 0.28 (0.22-0.36)
Ecuador Cuenca 35-49 0.33 (0.23-0.46)
Colombia Bogotá 41-60 0.34 (0.31-0.38)
Colombia Cucuta 41-60 0.37 (0.31-0.45)
China Wuhan 40-59 0.40 (0.33-0.48)
Peru Iquitos 30-59 0.43 (0.36-0.51)
Brazil Cuiabá 30-49 0.44 (0.32-0.60)
Hungary National Study 40-64 0.45 (0.18-1.33)
Peru Lima + Callao 30-59 0.64 (0.57-0.72)
Peru Lambayeque 51-60 0.82 (0.66-1.01)
Brazil Cuiabá 50-59 1.16 (0.82-1.63)

0.0 0.4 0.8 1.2 1.6
Figure A10 – IFR Estimates for Older Adults
India Karnataka 60-69 0.06 (0.05-0.08)
India Karnataka 70-79 0.09 (0.06-0.15)
India Karnataka 80+ 0.14 (0.07-0.29)
Brazil Maranhao 60-69 0.69 (0.60-0.81) IFR Estimate (%) 95% CI
India Chennai 60+ 0.73 (0.65-0.83)
Kenya Nairobi County 60+ 1.55 (0.88-2.79)
Colombia Barranquilla 61+ 1.73 (1.48-2.03)
Colombia Cucuta 61+ 1.99 (1.59-2.49)
Brazil Maranhao 70+ 2.04 (1.77-2.36)
Peru Lambayeque 61-70 2.14 (1.64-2.78)
Colombia Villavicencio 61+ 2.25 (1.73-2.93)
Jordan National Study 65+ 2.27 (1.87-2.79)
Paraguay Asunción + Central Dept. 60+ 2.36 (1.50-4.09)
Colombia Bucaramanga 61+ 2.60 (2.06-3.29)
Peru Iquitos 60+ 2.64 (2.18-3.27)
Colombia Cali 61+ 2.99 (2.42-3.71)
Colombia Medellín 61+ 3.20 (2.58-3.99)
Colombia Córdoba 70+ 3.27 (2.34-4.99)
Colombia Bogotá 61+ 3.69 (3.12-4.38)
Peru Lambayeque 71-80 3.98 (2.78-5.68)
Brazil Sao Paulo City 60+ 4.31 (3.43-5.42)
China Wuhan 60+ 4.32 (3.46-5.40)
Hungary National Study 65+ 4.36 (2.15-9.24)
Colombia Ipiales 61+ 4.41 (3.16-6.05)
Colombia Leticia 61+ 5.01 (3.72-6.61)
Peru Lambayeque 81+ 5.64 (3.46-9.32)
Peru Lima + Callao 60+ 6.11 (5.10-7.38)
Argentina Buenos Aires City 60+ 6.35 (4.69-8.61)
Ecuador Cuenca 65+ 8.49 (6.09-11.87)
Brazil Cuiabá 60+ 8.99 (6.22-12.97)
Chile 3 urban areas 60+ 9.81 (6.82-14.03)
Brazil Várzea Grande 60+ 11.81 (7.65-17.99)
Argentina Hurlingham 81+ 18.34 (9.13-32.92)
0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0
Note: Links to these studies are in the Appendix folder of our GitHub repository.

30

e. Metaregression Results
Figure A11 – Age‐specific IFR Metaregression in Levels
Percent

Note: Links to the studies at each location and categorization by percentage of deaths well‐certified
are provided in the appendix folder of our GitHub repository.

31

f. Population IFR
Figure A12 – Population IFR (ages 18 to 64)
Latin America Estimate (95% CI)

Argentina Buenos Aires City 0·32 (0·22-0·42)
Hurlingham 0·27 (0·16-0·39)
Brazil Cuiabá 0·47 (0·36-0·58)
Maranhao 0·08 (0·04-0·11)
Sao Paulo City 0·21 (0·17-0·25)
Várzea Grande 0·37 (0·29-0·46)
Chile 3 urban areas 0·32 (0·25-0·38)
Colombia Barranquilla 0·15 (0·11-0·19)
Bogotá 0·16 (0·12-0·21)
Bucaramanga 0·18 (0·12-0·23)
Cali 0·14 (0·10-0·18)
Cucuta 0·16 (0·11-0·21)
Córdoba 0·15 (0·10-0·21)
Ipiales 0·25 (0·14-0·35)
Leticia 0·28 (0·16-0·40)
Medellín 0·10 (0·07-0·13)
Villavicencio 0·12 (0·08-0·17)
Ecuador Cuenca 0·54 (0·43-0·65)
Paraguay Asunción + Central Dept. 0·13 (0·08-0·19)
Peru Iquitos 0·33 (0·22-0·44)
Lambayeque 0·25 (0·17-0·33)
Lima + Callao 0·46 (0·32-0·59)
Europe
Hungary National Study 0·47 (0·00-1·08)
Africa
Kenya Nairobi County 0·58 (0·22-0·95)
Middle East
Jordan National Study 0·10 (0·04-0·17)
South Asia
India Chennai 0·20 (0·09-0·31)
Karnataka 0·05 (0·02-0·08)
East Asia
China Wuhan 0·23 (0·19-0·26)
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Percent

IFR Estimate (%) 95% Credible Interval High-Income Country Benchmark
Note: This figure shows IFR estimates for the population aged 18‐64 based on the age structure and age‐specific seroprevalence in each location.

32

Figure A13 – Population IFR (all ages)
Latin America Location Estimate (95% CI)

Argentina Buenos Aires City 1.52 (1.10-1.94)
Hurlingham 1.51 (1.04-1.98)
Brazil Cuiabá 1.34 (1.07-1.61)
Distrito Federal 0.97 (0.77-1.17)
Foz do Iguaçu 0.02 (0.01-0.04)
Maranhao 0.19 (0.09-0.28)
Mato Grosso (8 cities) 0.93 (0.75-1.11)
Pitangueiras 0.93 (0.12-1.73)
Rio Grande do Sul 1.97 (1.49-2.44)
Sao Paulo City 0.86 (0.71-1.00)
Várzea Grande 0.94 (0.75-1.12)
Chile 3 urban areas 1.24 (1.03-1.44)
Colombia Barranquilla 0.32 (0.23-0.40)
Bogotá 0.45 (0.33-0.57)
Bucaramanga 0.47 (0.33-0.60)
Cali 0.48 (0.35-0.61)
Cucuta 0.32 (0.23-0.40)
Córdoba 0.37 (0.25-0.49)
Ipiales 0.61 (0.41-0.80)
Leticia 0.53 (0.36-0.70)
Medellín 0.42 (0.30-0.53)
Villavicencio 0.31 (0.22-0.40)
Ecuador Cuenca 0.95 (0.78-1.13)
Mexico National Study 0.63 (0.36-0.89)
Paraguay Asunción + Central Dept. 0.31 (0.19-0.44)
Peru Cusco Province 0.49 (0.34-0.63)
Iquitos 0.60 (0.42-0.78)
Lambayeque 0.54 (0.37-0.70)
Lima + Callao 0.97 (0.69-1.26)
Europe
Bosnia & Herz. Republika Srpska 0.43 (0.28-0.58)
Hungary National Study 1.69 (0.39-2.99)
Poland Katowice Region 0.99 (0.48-1.51)
Russia St. Petersburg 0.54 (0.38-0.69)
Africa
Ethiopia Addis Ababa 1.48 (0.35-2.62)
Dire Dawa 0.42 (0.00-0.90)
Kenya Nairobi County 0.71 (0.27-1.15)
Mozambique Maputo 0.75 (0.17-1.32)
Senegal National Study 0.08 (0.03-0.14)
South Africa Gauteng 0.32 (0.13-0.51)
Mitchells Plain 0.80 (0.35-1.24)
Zambia Copperbelt 1.99 (0.00-4.21)
Middle East
Iran National Study 0.39 (0.18-0.61)
Jordan National Study 0.19 (0.06-0.31)
Oman National Study 0.38 (0.17-0.60)
South Asia
India Berhampur 0.61 (0.17-1.06)
Bhubaneswar 1.19 (0.00-2.38)
Chennai 0.37 (0.17-0.57)
Delhi 0.12 (0.06-0.17)
Karnataka 0.07 (0.03-0.11)
Malegaon 0.19 (0.07-0.31)
Mumbai 0.26 (0.11-0.40)
Paschim Medinipur 0.05 (0.00-0.10)
Pimpri-Chinchwad 0.66 (0.29-1.02)
Puducherry 1.17 (0.55-1.80)
Srinagar District 0.41 (0.16-0.67)
Tamil Nadu (Vitros) 0.31 (0.14-0.48)
Tamil Nadu (iFlash) 0.08 (0.03-0.12)
Nepal National Study 0.35 (0.12-0.58)
Pakistan Karachi 0.40 (0.14-0.67)
Lahore 0.29 (0.11-0.46)
East Asia
China Hubei excluding Wuhan 0.20 (0.15-0.26)
Wuhan 0.86 (0.76-0.97)
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25
Percent
IFR Estimate (%) 95% Credible Interval
33

g. IFR Estimates from Other Sources
Table A8 compares reported IFRs from studies included in our literature search to our meta‐analysis
IFR estimates for the corresponding locations, while Table A9 lists reported IFRs from studies
identified in our literature search that were excluded from our IFR analysis. It should be noted that
IFRs based on reported COVID‐19 deaths may be biased due to death undercounting, especially for
countries that have a low percentage of well‐certified deaths (32).

Table A8 – Comparison of Meta‐analysis IFRs to Reported IFRs
Location IFR stated in the study Meta‐analysis IFR
Brazil: Maranhão* 0.14% (CI: 0.13 ‐ 0.16%) [reported deaths] 0.13% (CI: 0.12 ‐ 0.14%)

0.28% (CI: 0.25 ‐ 0.32%) [excess deaths]
Chile: Coquimbo‐La
Serena, Greater 1.67% (CI: 1.64 ‐ 1.70%) 1.23% (CI: 1.04 ‐ 1.45%)
Santiago, and Talca
Colombia: 0.24% (CI: 0.23 ‐ 0.25%) 0.34% (CI: 0.29 ‐ 0.43%)
Córdoba (8 cities)*
Mexico: National 0.47% (CI: 0.44‐0.50%) 0.46% (CI: 0.44‐0.48%)
Peru: Lambayeque 0.5% 0.49% (CI: 0.43 ‐ 0.56%)
Poland: Katowice 0.62% (CI: 0.53 ‐ 0.74%) 0.62% (CI: 0.50 ‐ 0.76%)
region
Russia: St. Petersburg* 0.83% (CI: 0.62 ‐ 1.00%) [excess deaths] 0.54% (CI: 0.41 ‐ 0.73%)
Ethiopia: ≥0.09% 0.20% (CI: 0.09 ‐ 0.63%)
Addis Ababa #2
Kenya: Nairobi County* 0.04% 0.06% (CI: 0.05 ‐ 0.07%)
South Africa: 0.28% (CI: 0.27 ‐ 0.30%) [reported deaths] 0.12% (CI: 0.10 ‐ 0.17%)
Gauteng province 0.67% (CI: 0.64 ‐ 0.71%) [excess deaths]
South Africa: 0.3% (CI: 0.3 ‐ 0.4%) [in‐hospital deaths] 0.31% (CI: 0.27 ‐ 0.37%)
Mitchells Plain 0.5% (CI: 0.4 ‐ 0.6%) [excess deaths]
India: national 0.08% (CI: 0.07 ‐ 0.09%) to 0.06% (CI: 0.05 ‐ 0.06%)
0.11% (CI: 0.10 ‐ 0.12%)
India: Chennai* 0.17% (CI: 0.14 ‐ 0.22%) 0.08% (CI: 0.07 ‐ 0.08%)
India: Delhi 0.079% (CI: 0.076 ‐ 0.081%) 0.055% (CI: 0.05 ‐ 0.06%)
India: Kashmir 0.03% (CI: 0.03 ‐ 0.04%) 0.03% (CI: 0.026 ‐
0.030%)
India: Madurai district* 0.04% (CI: 0.04 ‐ 0.05%) 0.03% (CI: 0.02 ‐ 0.03%)
India: Mumbai (3 0.12% 0.08% (CI: 0.08 ‐ 0.09%)
wards)*
India: Pimpri‐ 0.17% 0.22% (CI: 0.20 ‐ 0.23%)
Chinchwad
India: Puducherry* 0.08% 0.18% (CI: 0.15 ‐ 0.22%)
Note: IFRs are based on reported deaths, not excess deaths, unless otherwise noted. Studies with
an asterisk also reported age‐specific IFRs. Age‐specific IFRs were also reported for Karnataka,
India (98).

34

Table A9 – Reported IFRs for Sero‐Only Studies
Location IFR stated in the study Type of Reason IFR estimate was
death not generated
Rio das Pedras: 0.2%; Maré: 0.3%;
Brazil: Rio de Rocinha: 0.3%; Cidade de no death data
Janeiro (multiple Deus: 0.4%; Realengo: 1.2%; reported 4 weeks post‐midpoint,
regions of the city) Campo Grande: 1.8% insufficient information
on assay
wave 1: 0.12% (CI: 0.09 – 0.20%) excess
South Africa: wave 1: 0.16% (CI: 0.13 – 0.23%) in‐hospital no death data
Jouberton wave 2: 0.50% (CI: 0.29 – 1.17%) excess 4 weeks post‐midpoint
wave 2: 0.36% (CI: 0.24 – 0.72%) in‐hospital
Klerksdorp:
South Africa: 0.3% (CI: 0.2 – 0.3%) in‐hospital
Klerksdorp, 0.3% (CI: 0.3 – 0.3%) excess no death data
Pietermaritzburg Pietermaritzburg: 4 weeks post‐midpoint
0.3% (CI: 0.3 – 0.3%) in‐hospital
0.6% (CI: 0.5 – 0.6%) excess
no death data
4 weeks post‐midpoint,
Sudan: 0.64% (CI: 0.62 – 0.75%) excess sampling after February
Omdurman* 2021
Iran: Guilan 0.12% reported no death data 4 weeks
province post‐midpoint
no death data 4 weeks
Iran: Mazandaran 0.33% reported post‐midpoint,
province no stated start‐week and
end‐week
Palestine 0.11% reported no stated start‐week and
end‐week
India: Indore (city, 0.17% reported no death data 4 weeks
not district) post‐midpoint
India: Pune, 5 0.21% reported no death data 4 weeks
subwards* post‐midpoint
India: Tamil Nadu* 0.05% reported Tamil Nadu split into 2
regions based on assay
Note: IFRs are based on reported deaths, not excess deaths, unless otherwise noted. Studies with an
asterisk also reported age‐specific IFRs.

35

h. Covariates
Table A10 – Correlations Between Covariates, IFR, and Well‐Certified Deaths
Covariate Population IFR Well‐Certified Deaths
Human Development Index 0.63 (0.27‐0.83) 0.90 (0.78‐0.96)
Log of GDP per capita 0.60 (0.23‐0.82) 0.88 (0.73‐0.95)
Log(Healthcare Spending) 0.60 (0.22‐0.82) 0.93 (0.82‐0.97)
Log of GNI per capita 0.60 (0.22‐0.82) 0.88 (0.72‐0.95)
Hospital Beds Per Capita 0.57 (0.18‐0.80) 0.48 (0.06‐0.76)
Universal Health Coverage Index 0.55 (0.16‐0.79) 0.95 (0.88‐0.98)
Skilled Healthcare Workers Per Capita 0.49 (0.08‐0.76) 0.69 (0.36‐0.86)
Global Health Security Index 0.47 (0.05‐0.75) 0.59 (0.21‐0.81)
Life Expectancy at Birth 0.43 (0.0‐0.73) 0.71 (0.40‐0.88)
Healthy Life Expectancy at Age 60 0.40 (‐0.04‐0.71) 0.83 (0.61‐0.93)

This table demonstrates the relationship between various covariates (32), IFR, and the measure of
well‐certified deaths. This shows that well‐certified death is likely to be a primary explanatory
variable, which is confounded by relationships with GDP and other national measures when these
are used instead. An example of this is shown in the Directed Acyclic Graph below, made using the
Daggity online tool: http://www.dagitty.net/dags.html#

Figure A14 – Directed Acyclic Graph of Covariate Relationships

36

i. Out‐of‐Sample Analysis
Our analysis excludes seroprevalence estimates that geographically overlap with an already included
location, as discussed in supplementary appendix section 1.b. The body of the paper also excludes
IFR estimates that overlap with an included IFR, though we still calculated population‐wide IFRs for
these out‐of‐sample locations. The table below lists these IFR estimates:
Table A11 – IFRs for Out‐of‐Sample Locations with Geographical Overlap
Location Excluded from body of paper Meta‐analysis IFR
Brazil: São Paulo* No 0.84% (CI: 0.76 ‐ 0.93%)
Brazil: São Paulo #2* Yes 0.77% (CI: 0.66 ‐ 0.88%)
Ethiopia: Addis Ababa* No 0.11% (CI: 0.07 ‐ 0.16%)
Ethiopia: Addis Ababa #2 Yes 0.20% (CI: 0.09 ‐ 0.63%)
Ethiopia: Addis Ababa #3* Yes 0.002% (CI: 0.001 ‐
0.005%)
India: national* Yes 0.06% (CI: 0.05 ‐ 0.06%)
India: Kashmir (Srinagar district)* No 0.06% (CI: 0.06 ‐ 0.07%)
India: Kashmir* Yes 0.03% (CI: 0.026 ‐
0.030%)
India: Tamil Nadu (Vitros districts) No 0.07% (CI: 0.06 ‐ 0.07%)
India: Madurai district Yes 0.03% (CI: 0.02 ‐ 0.03%)
(in Tamil Nadu)*
China: Wuhan* No 0.86% (CI: 0.76 ‐ 0.97%)
China: Wuhan #2* Yes 0.71% (CI: 0.59 ‐ 1.06%)
*age‐specific seroprevalence also reported in the paper

IFRs from studies of the same location may differ by sampling time due to factors such as improved
treatment or new SARS‐CoV‐2 variants. Despite this, population‐wide IFRs were relatively similar for
studies that sampled the same location, as illustrated in the table above. These consilient results
increase confidence that methodological differences between studies likely do not strongly bias our
IFR estimates, in contrast to the order of magnitude difference in IFR between locations stratified by
percentage of well‐certified deaths, as shown in the body of the paper. Addis Ababa #3 remains the
only outlier, possibly due to lower test specificity resulting from cross‐reactivity (see supplementary
appendix section 2.a), low sample size in comparison to the other two Addis Ababa studies, or
sampling in late April 2020 when under‐estimation of COVID‐19 deaths may have been greater than
the July/August 2020 time period during which the other two studies sampled.

37

Appendix References
1. Bobrovitz N, Arora RK, Yan T, et al. Lessons from a rapid systematic review of early SARS‐CoV‐2
serosurveys. medRxiv. 2020:2020.05.10.20097451. doi:10.1101/2020.05.10.20097451
2. Chen X, Chen Z, Azman AS, et al. Serological evidence of human infection with SARS‐CoV‐2: a
systematic review and meta‐analysis. The Lancet Global Health. doi:10.1016/S2214‐
109X(21)00026‐7
3. Byambasuren O, Dobler CC, Bell K, et al. Comparison of seroprevalence of SARS‐CoV‐2
infections with cumulative and imputed COVID‐19 cases: Systematic review. PLOS ONE.
2021;16(4):e0248946. doi:10.1371/journal.pone.0248946
4. Bergeri I, Whelan M, Ware H, et al. Global epidemiology of SARS‐CoV‐2 infection: a systematic
review and meta‐analysis of standardized population‐based seroprevalence studies, Jan 2020‐
Oct 2021. medRxiv. 2021:2021.12.14.21267791. doi:10.1101/2021.12.14.21267791
5. Laurette M, Marion V, Eduard G, Alex W. Research Square. 2021. doi:10.21203/rs.3.rs‐
707813/v2
6. Golding J, Northstone K, Miller LL, Davey Smith G, Pembrey M. Differences between blood
donors and a population sample: implications for case‐control studies. International journal of
epidemiology. 2013;42(4):1145‐56. doi:10.1093/ije/dyt095
7. Pham D, Nguyen D, Nguyen T, et al. Seroprevalence of HTLV‐1/2 Among Voluntary Blood
Donors in Vietnam. AIDS Research and Human Retroviruses. 2019;35(4):376‐81.
doi:10.1089/aid.2018.0240
8. He D, Artzy‐Randrup Y, Musa SS, Gräf T, Naveca F, Stone L. The unexpected dynamics of COVID‐
19 in Manaus, Brazil: Was herd immunity achieved? medRxiv. 2021:2021.02.18.21251809.
doi:10.1101/2021.02.18.21251809
9. Boyce RM, Shook‐Sa BE, Aiello AE. A Tale of 2 Studies: Study Design and Our Understanding of
Severe Acute Respiratory Syndrome Coronavirus 2 Seroprevalence. Clinical Infectious Diseases.
2020. doi:10.1093/cid/ciaa1868
10. Campbell H, de Valpine P, Maxwell L, et al. Bayesian adjustment for preferential testing in
estimating infection fatality rates, as motivated by the COVID‐19 pandemic. The Annals of
Applied Statistics. 2022;16(1):436‐59, 24.
11. Shook‐Sa BE, Boyce RM, Aiello AE. Estimation Without Representation: Early Severe Acute
Respiratory Syndrome Coronavirus 2 Seroprevalence Studies and the Path Forward. The Journal
of Infectious Diseases. 2020;222(7):1086‐9. doi:10.1093/infdis/jiaa429
12. Levin AT, Hanage WP, Owusu‐Boaitey N, Cochran KB, Walsh SP, Meyerowitz‐Katz G. Assessing
the age specificity of infection fatality rates for COVID‐19: systematic review, meta‐analysis, and
public policy implications. European Journal of Epidemiology. 2020;35(12):1123‐38.
doi:10.1007/s10654‐020‐00698‐1
13. Community Assessment for Public Health Emergency Response Toolkit: CDC2019.
14. Bobrovitz N, Noël K, Li Z, et al. SeroTracker‐RoB: an approach to automating reproducible risk of
bias assessment of seroprevalence studies. medRxiv. 2022:2021.11.17.21266471.
doi:10.1101/2021.11.17.21266471
15. World Health Organization. Population‐based age‐stratified seroepidemiological investigation
protocol for COVID‐19 virus infection. Geneva: World Health Organization2020 17 March 2020
Contract No.: WHO/2019‐nCoV/Seroepidemiology/2020.1.
16. World Economic and Financial Surveys World Economic Outlook Database—WEO Groups and
Aggregates Information2021.
17. Gajda M, Kowalska M, Zejda JE. Impact of Two Different Recruitment Procedures (Random vs.
Volunteer Selection) on the Results of Seroepidemiological Study (SARS‐CoV‐2). International
Journal of Environmental Research and Public Health. 2021;18(18):9928.
38

18. Knabl L, Mitra T, Kimpel J, et al. High SARS‐CoV‐2 seroprevalence in children and adults in the
Austrian ski resort of Ischgl. Communications Medicine. 2021;1(1):4. doi:10.1038/s43856‐021‐
00007‐1
19. COVID‐19 Cases. Ministerio de Salud Argentina. http://datos.salud.gob.ar/dataset/covid‐19‐
casos‐registrados‐en‐la‐republica‐argentina/archivo/fd657d02‐a33a‐498b‐a91b‐2ef1a68b8d16.
20. COVID‐19 in Colombia. Instituto Nacional De Salud. 2021.
https://www.ins.gov.co/Noticias/paginas/coronavirus.aspx.
21. COVID‐19 PANEL ‐ ESPIRITO SANTO STATE. 2021. https://coronavirus.es.gov.br/painel‐covid‐19‐
es.
22. Coronavirus ‐ COVID‐19. Parana Governo Do Estado. 2021.
https://www.saude.pr.gov.br/Pagina/Coronavirus‐COVID‐19.
23. COVID19 Registro Fallecidos. 2021.
https://public.tableau.com/app/profile/mspbs/viz/COVID19PY‐Registros/FALLECIDOS.
24. Karlinsky A, Kobak D. Tracking excess mortality across countries during the COVID‐19 pandemic
with the World Mortality Dataset. eLife. 2021;10:e69336. doi:10.7554/eLife.69336
25. Ramachandran S, Malani A. All‐cause mortality during SARS‐CoV‐2 Pandemic in India:
Nationally‐representative estimates independent of official death registry. medRxiv.
2021:2021.07.20.21260577. doi:10.1101/2021.07.20.21260577
26. Watson OJ, Alhaffar M, Mehchy Z, et al. Leveraging community mortality indicators to infer
COVID‐19 mortality and transmission dynamics in Damascus, Syria. Nature Communications.
2021;12(1):2394. doi:10.1038/s41467‐021‐22474‐9
27. Dyer O. Covid‐19: Russia admits to understating deaths by more than two thirds. BMJ.
2020;371:m4975. doi:10.1136/bmj.m4975
28. Dyer O. Covid‐19: Mexico acknowledges 50 000 more deaths than official figures show. BMJ.
2020;371:m4182. doi:10.1136/bmj.m4182
29. Technical criteria to update the death toll from COVID‐19 in Peru: Peruvian State2021.
30. Nepal Dashboard: World Health Organization2021.
31. COVID‐19 MÉXICO Comunicado Técnico Diario.
32. Fullman N, Barber RM, Abajobir AA, et al. Measuring progress and projecting attainment on the
basis of past trends of the health‐related Sustainable Development Goals in 188 countries: an
analysis from the Global Burden of Disease Study 2016. The Lancet. 2017;390(10100):1423‐59.
doi:10.1016/S0140‐6736(17)32336‐X
33. Silveira MF, Mesenburg MA, Dellagostin OA, et al. Time‐dependent decay of detectable
antibodies against SARS‐CoV‐2: A comparison of ELISA with two batches of a lateral‐flow test.
Braz J Infect Dis. 2021;25(4):101601‐. doi:10.1016/j.bjid.2021.101601
34. Conklin SE, Martin K, Manabe YC, et al. Evaluation of Serological SARS‐CoV‐2 Lateral Flow
Assays for Rapid Point‐of‐Care Testing. Journal of clinical microbiology. 2021;59(2):e02020‐20.
doi:10.1128/JCM.02020‐20
35. Hartwig FP, Vidaletti LP, Barros AJD, et al. Combining serological assays and official statistics to
describe the trajectory of the COVID‐19 pandemic: results from the EPICOVID19‐RS study in Rio
Grande do Sul (Southern Brazil). medRxiv. 2021:2021.05.21.21257634.
doi:10.1101/2021.05.21.21257634
36. Lipsitch M, Grad YH, Sette A, Crotty S. Cross‐reactive memory T cells and herd immunity to
SARS‐CoV‐2. Nature Reviews Immunology. 2020;20(11):709‐13. doi:10.1038/s41577‐020‐
00460‐4
37. Dorigatti I, Lavezzo E, Manuto L, et al. SARS‐CoV‐2 antibody dynamics and transmission from
community‐wide serological testing in the Italian municipality of Vo’. Nature Communications.
2021;12(1):4383. doi:10.1038/s41467‐021‐24622‐7
38. Kshatri JS, Bhattacharya D, Praharaj I, et al. Seroprevalence of SARS‐CoV‐2 in Bhubaneswar,
India: findings from three rounds of community surveys. Epidemiology and infection.
2021;149:e139. doi:10.1017/s0950268821000972
39

39. Laxmaiah A, Rao NM, Arlappa N, et al. SARS‐CoV‐2 seroprevalence in the city of Hyderabad,
India in early 2021. medRxiv. 2021:2021.07.18.21260555. doi:10.1101/2021.07.18.21260555
40. Wagner R, Peterhoff D, Beileke S, et al. Estimates and Determinants of SARS‐Cov‐2
Seroprevalence and Infection Fatality Ratio Using Latent Class Analysis: The Population‐Based
Tirschenreuth Study in the Hardest‐Hit German County in Spring 2020. Viruses.
2021;13(6):1118.
41. Prevention USCfDC. Science Brief: SARS‐CoV‐2 Infection‐induced and Vaccine‐induced
Immunity. 2021.
42. Sharma N, Sharma P, Basu S, et al. Second wave of the Covid‐19 pandemic in Delhi, India: high
seroprevalence not a deterrent? medRxiv. 2021:2021.09.09.21263331.
doi:10.1101/2021.09.09.21263331
43. Álvarez‐Antonio C, Meza‐Sánchez G, Calampa C, et al. Seroprevalence of anti‐SARS‐CoV‐2
antibodies in Iquitos, Peru in July and August, 2020: a population‐based study. The Lancet
Global Health. 2021;9(7):e925‐e31. doi:10.1016/S2214‐109X(21)00173‐X
44. Fox SJ, Potu P, Lachmann M, Srinivasan R, Meyers LA. The COVID‐19 herd immunity threshold is
not low: A re‐analysis of European data from spring of 2020. medRxiv.
2020:2020.12.01.20242289. doi:10.1101/2020.12.01.20242289
45. Bellizzi S, Alsawalha L, Sheikh Ali S, et al. A three‐phase population based sero‐epidemiological
study: Assessing the trend in prevalence of SARS‐CoV‐2 during COVID‐19 pandemic in Jordan.
One Health. 2021;13:100292. doi:https://doi.org/10.1016/j.onehlt.2021.100292
46. Sharma N, Sharma P, Basu S, et al. Second Wave of the COVID‐19 Pandemic in Delhi, India: High
Seroprevalence Not a Deterrent? Cureus. 2021;13(10): e19000. doi:doi:10.7759/cureus.19000
47. Poustchi H, Darvishian M, Mohammadi Z, et al. SARS‐CoV‐2 antibody seroprevalence in the
general population and high‐risk occupational groups across 18 cities in Iran: a population‐
based cross‐sectional study. The Lancet Infectious Diseases. 2021;21(4):473‐81.
doi:10.1016/S1473‐3099(20)30858‐6
48. Emmerich P, Murawski C, Ehmen C, et al. Limited specificity of commercially available SARS‐
CoV‐2 IgG ELISAs in serum samples of African origin. Tropical Medicine & International Health.
2021;26(6):621‐31. doi:https://doi.org/10.1111/tmi.13569
49. Abdella S, Riou S, Tessema M, et al. Prevalence of SARS‐CoV‐2 in urban and rural
Ethiopia: Randomized household serosurveys reveal level of spread during the first wave
of the pandemic. EClinicalMedicine. 2021;35. doi:10.1016/j.eclinm.2021.100880
50. Brazil Civil Registry. COVID‐19 Portal da Transparencia. 2022.
51. Steinhardt LC, Ige F, Iriemenam NC, et al. Cross‐Reactivity of Two SARS‐CoV‐2 Serological Assays
in a Setting Where Malaria Is Endemic. Journal of Clinical Microbiology. 2021;59(7):e00514‐21.
doi:doi:10.1128/JCM.00514‐21
52. Alemu BN, Addissie A, Mamo G, et al. Sero‐prevalence of anti‐SARS‐CoV‐2 Antibodies in Addis
Ababa, Ethiopia. bioRxiv. 2020:2020.10.13.337287. doi:10.1101/2020.10.13.337287
53. Nkuba AN, Makiala SM, Guichet E, et al. High Prevalence of Anti–Severe Acute Respiratory
Syndrome Coronavirus 2 (Anti–SARS‐CoV‐2) Antibodies After the First Wave of Coronavirus
Disease 2019 (COVID‐19) in Kinshasa, Democratic Republic of the Congo: Results of a Cross‐
sectional Household‐Based Survey. Clinical Infectious Diseases. 2021. doi:10.1093/cid/ciab515
54. Wiens KE, Mawien PN, Rumunu J, et al. Seroprevalence of Severe Acute Respiratory Syndrome
Coronavirus 2 IgG in Juba, South Sudan, 2020(1). Emerg Infect Dis. 2021;27(6):1598‐606.
doi:10.3201/eid2706.210568
55. Rogan WJ, Gladen B. Estimating prevalence from the results of a screening test. American
journal of epidemiology. 1978;107(1):71‐6. doi:10.1093/oxfordjournals.aje.a112510
56. Gelman A, Carpenter B. Bayesian analysis of tests with unknown specificity and sensitivity.
medRxiv. 2020:2020.05.22.20108944. doi:10.1101/2020.05.22.20108944
40

57. Peluso MJ, Takahashi S, Hakim J, et al. SARS‐CoV‐2 antibody magnitude and detectability are
driven by disease severity, timing, and assay. Science Advances. 2021;7(31):eabh3409.
doi:doi:10.1126/sciadv.abh3409
58. Mutevedzi PC, Kawonga M, Kwatra G, et al. Estimated SARS‐CoV‐2 infection rate and fatality
risk in Gauteng Province, South Africa: a population‐based seroepidemiological survey.
International Journal of Epidemiology. 2021. doi:10.1093/ije/dyab217
59. Perez‐Saez J, Lauer SA, Kaiser L, et al. Serology‐informed estimates of SARS‐COV‐2 infection
fatality risk in Geneva, Switzerland. medRxiv. 2020:2020.06.10.20127423.
doi:10.1101/2020.06.10.20127423
60. Stefanelli P, Bella A, Fedele G, et al. Longevity of seropositivity and neutralizing titers among
SARS‐CoV‐2 infected individuals after 4 months from baseline: a population‐based study in the
province of Trento. medRxiv. 2020:2020.11.11.20229062. doi:10.1101/2020.11.11.20229062
61. Pérez‐Olmeda M, Saugar JM, Fernández‐García A, et al. Evolution of antibodies against SARS‐
CoV‐2 over seven months: experience of the Nationwide Seroprevalence ENE‐COVID Study in
Spain. medRxiv. 2021:2021.03.11.21253142. doi:10.1101/2021.03.11.21253142
62. NEW STATEWIDE DATA SHOW EVIDENCE OF FOUR‐FOLD INCREASE IN RECENT COVID‐19
INFECTIONS: University of Wisconsin‐Madison2020.
63. Barchuk A, Shirokov D, Sergeeva M, et al. Evaluation of the performance of SARS‐‐CoV‐‐2
antibody assays for a longitudinal population‐based study of COVID‐‐19 spread in St.
Petersburg, Russia. Journal of Medical Virology. 2021;93(10):5846‐52.
doi:https://doi.org/10.1002/jmv.27126
64. Beverland A, Keogan M, Connell J, De Gascun C, Igoe D. Longitudinal Serological Analysis
Following a National Seroprevalence Study to Investigate COVID‐19 Infection in People Living in
Ireland. The Journal of Infectious Diseases. 2021;224(6):1100‐1. doi:10.1093/infdis/jiab346
65. European University of St. Petersburg. By mid‐August, less than 13% of St. Petersburg residents
had contracted the coronavirus. Most recoveries retain antibodies. 2020.
66. Carreño JM, Mendu DR, Simon V, et al. Longitudinal analysis of SARS‐CoV‐2 seroprevalence
using multiple serology platforms. medRxiv. 2021:2021.02.24.21252340.
doi:10.1101/2021.02.24.21252340
67. Kahre E, Galow L, Unrath M, et al. Kinetics and seroprevalence of SARS‐CoV‐2 antibodies – a
comparison of 3 different assays. medRxiv. 2021:2021.03.10.21253273.
doi:10.1101/2021.03.10.21253273
68. Thiruvengadam R, Chattopadhyay S, Mehdi F, et al. Longitudinal serology in SARS‐CoV‐2
infected individuals in India – a prospective cohort study. medRxiv. 2021:2021.02.04.21251140.
doi:10.1101/2021.02.04.21251140
69. Muecksch F, Wise H, Batchelor B, et al. Longitudinal Serological Analysis and Neutralizing
Antibody Levels in Coronavirus Disease 2019 Convalescent Patients. The Journal of Infectious
Diseases. 2020;223(3):389‐98. doi:10.1093/infdis/jiaa659
70. Sim M, Cockcroft C, Darby D, et al. Paired sensitivity analysis of four SARS‐CoV‐2 serological
immunoassays in a longitudinal cohort of convalescent hospital staff. Annals of Clinical
Biochemistry. 2021:00045632211030957. doi:10.1177/00045632211030957
71. Sharma N, Sharma P, Basu S, et al. The seroprevalence of severe acute respiratory syndrome
coronavirus 2 in Delhi, India: a repeated population‐based seroepidemiological study.
Transactions of The Royal Society of Tropical Medicine and Hygiene. 2021.
doi:10.1093/trstmh/trab109
72. Satpati P, Sarangi S, Gantait K, et al. Sero‐surveillance (IgG) of SARS‐CoV‐2 among
Asymptomatic General population of Paschim Medinipur, West Bengal, India. medRxiv.
2020:2020.09.12.20193219. doi:10.1101/2020.09.12.20193219
73. Perez‐Saez J, Zaballa M‐E, Yerly S, et al. Persistence and detection of anti‐SARS‐CoV‐2
antibodies: immunoassay heterogeneity and implications for serosurveillance. medRxiv.
2021:2021.03.16.21253710. doi:10.1101/2021.03.16.21253710
41

74. Ladage D, Rösgen D, Schreiner C, et al. Persisting Antibody Response to SARS‐CoV‐2 in a Local
Austrian Population. Frontiers in Medicine. 2021;8(881). doi:10.3389/fmed.2021.653630
75. Choe PG, Kim K‐H, Kang CK, et al. Antibody Responses 8 Months after Asymptomatic or Mild
SARS‐CoV‐2 Infection. Emerging Infectious Disease journal. 2021;27(3):928.
doi:10.3201/eid2703.204543
76. Robert Koch Institut. Local Corona Monitoring: Key Data for Berlin‐Mitte. 2021.
doi:10.25646/8986
77. Robert Koch Institut. Local Corona Monitoring: Key data for Straubing. 2021.
78. Kar S SS, Murali S, Dhodapkar R, Joseph N, Aggarwal R. Prevalence and Time Trend of SARS‐
CoV‐2 Infection in Puducherry, India. Emerg Infectious Diseases. 2021;27(2):666‐9.
doi:https://doi.org/10.3201/eid2702.204480
79. Vial PAaG, Claudia and Icaza, Gloria and Ramirez‐Santana, Muriel and Quezada‐Gaete, Ruben
and Nuñez‐Franz, Loreto and Apablaza, Mauricio and Vial, M. Cecilia and Rubilar, Paola and
Correa, Juan and Pérez, Claudia and Florea, Andrei and Guzman, Eugenio and Lavin, Maria‐
Estela and Concha, Paula and Najera‐de Ferrari, Manuel and Najera‐de Ferrari, Manuel and
Aguilera, Ximena, . Seroprevalence, Spatial Distribution, and Social Determinants of SARS‐CoV‐2
in Three Urban Centers of Chile. . SSRN. 2021.
80. Radon K, Bakuli A, Pütz P, et al. From first to second wave: follow‐up of the prospective Covid‐
19 cohort (KoCo19) in Munich (Germany). medRxiv. 2021:2021.04.27.21256133.
doi:10.1101/2021.04.27.21256133
81. He Z, Ren L, Yang J, et al. Seroprevalence and humoral immune durability of anti‐SARS‐CoV‐2
antibodies in Wuhan, China: a longitudinal, population‐level, cross‐sectional study. The Lancet.
2021;397(10279):1075‐84. doi:10.1016/S0140‐6736(21)00238‐5
82. Goto A, Go H, Miyakawa K, et al. Sustained Neutralizing Antibodies 6 Months Following
Infection in 376 Japanese COVID‐19 Survivors. Frontiers in Microbiology. 2021;12(1039).
doi:10.3389/fmicb.2021.661187
83. Domènech‐Montoliu S, Puig‐Barberà J, Pac‐Sa MR, et al. Persistence of Anti‐SARS‐CoV‐2
Antibodies Six Months after Infection in an Outbreak with Five Hundred COVID‐19 Cases in
Borriana (Spain): A Prospective Cohort Study. COVID. 2021;1(1):71‐82.
84. Pagotto V, Luna L, Salto J, et al. Long‐Term Duration of Antibody Response to SARS CoV‐2 in
One of the Largest Slums of Buenos Aires. medRxiv. 2021:2021.03.05.21253010.
doi:10.1101/2021.03.05.21253010
85. Rodeles LM PL, Benitez R, Benzaquen N, Serravalle P, Long AK, et al. Seroprevalence of anti‐
SARS‐CoV‐2 IgG in asymptomatic and pauci‐symptomatic people over a 5 month survey in
Argentina. Rev Panama Salud Publica. 2021;45(66).
86. Peluso MJ, Takahashi S, Hakim J, et al. SARS‐CoV‐2 antibody magnitude and detectability are
driven by disease severity, timing, and assay. medRxiv. 2021:2021.03.03.21251639.
doi:10.1101/2021.03.03.21251639
87. Perez‐Saez J, Zaballa M‐E, Yerly S, et al. Persistence of anti‐SARS‐CoV‐2 antibodies:
immunoassay heterogeneity and implications for serosurveillance. Clinical Microbiology and
Infection. 2021;27(11):1695.e7‐.e12. doi:10.1016/j.cmi.2021.06.040
88. Alvim RGF, Lima TM, Rodrigues DAS, et al. Development and large‐scale validation of a highly
accurate SARS‐COV‐2 serological test using regular test strips for autonomous and affordable
finger‐prick sample collection, transportation, and storage. medRxiv.
2021:2020.07.13.20152884. doi:10.1101/2020.07.13.20152884
89. Petersen MS, Hansen CB, Fríðheim Kristiansen M, et al. SARS‐CoV‐2 natural antibody response
persists up to 12 months in a nationwide study from the Faroe Islands. medRxiv.
2021:2021.04.19.21255720. doi:10.1101/2021.04.19.21255720
90. Gudbjartsson DF, Helgason A, Jonsson H, et al. Spread of SARS‐CoV‐2 in the Icelandic
Population. New England Journal of Medicine. 2020. doi:10.1056/NEJMoa2006100
42

91. Campbell H, Gustafson P. Inferring the COVID‐19 infection fatality rate in the community‐
dwelling population: a simple Bayesian evidence synthesis of seroprevalence study data and
imprecise mortality data. Epidemiology and infection. 2021;149:e243.
doi:10.1017/S0950268821002405
92. Lu N, Cheng K‐W, Qamar N, Huang K‐C, Johnson JA. Weathering COVID‐19 storm: Successful
control measures of five Asian countries. American Journal of Infection Control.
2020;48(7):851‐2. doi:10.1016/j.ajic.2020.04.021
93. Okell LC, Verity R, Watson OJ, et al. Have deaths from COVID‐19 in Europe plateaued due to
herd immunity? The Lancet. 2020;395(10241):e110‐e1. doi:10.1016/S0140‐6736(20)31357‐X
94. Fuller JA HA, Victory KR, et al. Mitigation Policies and COVID‐19–Associated Mortality — 37
European Countries, January 23–June 30, 2020. MMWR Morb Mortal Wkly Rep. 2020;70:58‐62.
95. Miyawaki A, Tsugawa Y. Health and Public Health Implications of COVID‐19 in Asian Countries.
Asian Economic Policy Review.n/a(n/a). doi:https://doi.org/10.1111/aepr.12358
96. Lee K, Jo S, Lee J. Seroprevalence of SARS‐CoV‐2 antibodies in South Korea. Journal of the
Korean Statistical Society. 2021;50(3):891‐904. doi:10.1007/s42952‐021‐00131‐7
97. Parrott JC, Maleki AN, Vassor VE, et al. Prevalence of SARS‐CoV‐2 Antibodies in New York City
Adults, June–October 2020: A Population‐Based Survey. The Journal of Infectious Diseases.
2021;224(2):188‐95. doi:10.1093/infdis/jiab296
98. Cai R, Novosad P, Tandel V, Asher S, Malani A. Representative estimates of COVID‐19 infection
fatality rates from four locations in India: cross‐sectional study. BMJ Open.
2021;11(10):e050920. doi:10.1136/bmjopen‐2021‐050920

43


Assessing The Burden of COVID-19 in Developing Countries BMJ Global Health

Uploaded by

Copyright:

Available Formats

Assessing The Burden of COVID-19 in Developing Countries BMJ Global Health

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Assessing The Burden of COVID-19 in Developing Countries BMJ Global Health

Uploaded by

Copyright:

Available Formats

Original research

Assessing the burden of COVID-­19 in

To cite: Levin AT, ABSTRACT

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477  1

far more dangerous for middle-­aged and older adults

2 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 3

Figure 1 Map of study locations. IFR, infection fatality rate.

4 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Table 2 Included studies for infection fatality rate (IFR)

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 5

6 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Figure 2 Estimates of seroprevalence.

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 7

Figure 3 Age‐specific seroprevalence by location.

8 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

adjustments make relatively little difference for countries DISCUSSION

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 9

Table 3 Ratio of excess mortality to reported COVID-­19 deaths

10 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 11

Another important facet of our results is that sero-

12 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 13

14 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 15

16 Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 17

Region Location Reported seroprevalence Number tested for

Assay IFR Locations Sero‐Only Locations

𝑝, sens 𝜋 , 1 spec 1 𝜋 , . (2)

𝐷 ⋆, ∼ Poisson 𝑁 , 𝜋, IFR , (5)

where 𝑁 , gives the number of individuals at location 𝑙 in age group A. Then 𝑁 , 𝜋 , gives the

The lower and upper bounds of age bin A at location 𝑙 are given by 𝐿 , and 𝑈 , , respectively. For

2 5 7 12 17 22 27 31 35 41 45 48 52 119 182 2 5 7 12 17 22 27 31 35 41 45 48 52 119 182

Test kit ID Test kit ID

𝑓⋆ 𝑎 median 𝑓 𝑎 ∣ l is one of the observed locations with fatality data (19)

Adjustment for ‐3.22 0.0516

Adjustment for Death ‐3.18 0.0526

𝑙𝑜𝑔 𝐼𝐹𝑅 3.27 0.0524 ∗ 𝑎𝑔𝑒

Latin America Location Ratio (95% CI)

0 .2 .4 .6 .8 1 1.2 1.4 1.6 1.8 2

Percent Buenos Aires City, Argentina Hurlingham, Argentina

Percent Sao Paulo City, Brazil

Posterior Estimate Posterior Estimate

Coverage Interval Coverage Interval

Várzea Grande, Brazil

Barranquilla, Colombia Percent Bogotá, Colombia

Bucaramanga, Colombia Percent Cali, Colombia

Posterior Estimate Posterior Estimate

Percent Ipiales, Colombia Leticia, Colombia

Percent Medellín, Colombia Villavicencio, Colombia

Percent Iquitos, Peru Percent Lambayeque, Peru

Percent Lima + Callao, Peru

National Study, Jordan Percent National Study, Hungary

Percent Chennai, India Karnataka, India

Percent Wuhan, China

0.000 0.015 0.030 0.045 0.060

0.00 0.03 0.06 0.09 0.12

Assessing the burden of COVID-19 in

Levin AT, et al. BMJ Global Health 2022;7:e008477. doi:10.1136/bmjgh-2022-008477 1

far more dangerous for middle-aged and older adults

Table 3 Ratio of excess mortality to reported COVID-19 deaths