Available via license: CC BY-NC 4.0
Content may be subject to copyright.
Page 1 of 28
SARS-CoV-2 reinfection trends in South Africa: analysis of
routine surveillance data
Juliet R.C. Pulliam1,*, Cari van Schalkwyk1, Nevashan Govender2, Anne von Gottberg2,3,
Cheryl Cohen2,4, Michelle J. Groome2,3, Jonathan Dushoff1,5, Koleka Mlisana6,7,8, Harry
Moultrie2
1 SACEMA, Stellenbosch University, South Africa
2 National Institute for Communicable Diseases, Division of the National Health Laboratory
Service, South Africa
3 School of Pathology, Faculty of Health Sciences, University of the Witwatersrand,
Johannesburg, South Africa
4 School of Public Health, Faculty of Health Sciences, University of the Witwatersrand,
Johannesburg, South Africa
5 McMaster University, Canada
6 National Health Laboratory Service, South Africa
7 School of Laboratory Medicine and Medical Sciences, University of KwaZulu-Natal, South
Africa
8 Centre for the AIDS Programme of Research in South Africa (CAPRISA), South Africa
* corresponding author: pulliam@sun.ac.za
Abstract
Objective To examine whether SARS-CoV-2 reinfection risk has changed through time in
South Africa, in the context of the emergence of the Beta and Delta variants
Design Retrospective analysis of routine epidemiological surveillance data
Setting Line list data on SARS-CoV-2 with specimen receipt dates between 04 March 2020
and 30 June 2021, collected through South Africa’s National Notifiable Medical Conditions
Surveillance System
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.
Page 2 of 28
Participants 1,551,655 individuals with laboratory-confirmed SARS-CoV-2 who had a
positive test result at least 90 days prior to 30 June 2021. Individuals having sequential
positive tests at least 90 days apart were considered to have suspected reinfections.
Main outcome measures Incidence of suspected reinfections through time; comparison of
reinfection rates to the expectation under a null model (approach 1); empirical estimates of
the time-varying hazards of infection and reinfection throughout the epidemic (approach 2)
Results 16,029 suspected reinfections were identified. The number of reinfections
observed through the end of June 2021 is consistent with the null model of no change in
reinfection risk (approach 1). Although increases in the hazard of primary infection were
observed following the introduction of both the Beta and Delta variants, no corresponding
increase was observed in the reinfection hazard (approach 2). Contrary to expectation, the
estimated hazard ratio for reinfection versus primary infection was lower during waves
driven by the Beta and Delta variants than for the first wave (relative hazard ratio for wave 2
versus wave 1: 0.75 (CI
!"
: 0.59-0.97); for wave 3 versus wave 1: 0.70 (CI
!"
: 0.55-0.90)).
Although this finding may be partially explained by changes in testing availability, it is also
consistent with a scenario in which variants have increased transmissibility but little or no
evasion of immunity.
Conclusion We conclude there is no population-wide epidemiological evidence of immune
escape and recommend ongoing monitoring of these trends.
Box 1
What is already known on this topic
• Prior infection with SARS-CoV-2 is estimated to provide at least an 80% reduction in
infection risk (1,2).
• Laboratory-based studies indicate reduced neutralization by convalescent serum for
the Beta and Delta variants relative to wild type virus (3–6); however, the impact of
these reductions on risk of reinfection is not known.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 3 of 28
What this study adds
• We provide two methods for monitoring reinfection trends to identify signatures of
changes in reinfection risk.
• We find no evidence of increased reinfection risk associated with circulation of Beta
or Delta variants compared to the ancestral strain in routine epidemiological data
from South Africa.
Introduction
As of 30 June 2021, South Africa had more than two million cumulative laboratory-
confirmed cases of SARS-CoV-2, concentrated in three waves of infection. The first case
was detected in early March 2020 and was followed by a wave that peaked in July 2020
and officially ended in September. The second wave, which peaked in January 2021 and
ended in February, was driven by the Beta (B.1.351 / 501Y.V2 / 20H) variant, which was
first detected in South Africa in October 2020 (7). The third wave, which peaked in July and
ended in September 2021, was dominated by the Delta (B.1.617.2 / 478K.V1 / 21A) variant
(8).
Following emergence of the Beta and Delta variants of SARS-CoV-2 in South Africa,
a key question remains of whether there is epidemiologic evidence of increased risk of
SARS-CoV-2 reinfection with these variants (i.e., immune escape). Laboratory-based
studies suggest that convalescent serum has a reduced neutralizing effect on these
variants compared to wild type virus in vitro (3–6); however, this finding does not
necessarily translate into immune escape at the population level.
To examine whether reinfection risk has changed through time, it is essential to
account for potential confounding factors affecting the incidence of reinfection: namely, the
changing force of infection experienced by all individuals in the population and the growing
number of individuals eligible for reinfection through time. These factors are tightly linked to
the timing of epidemic waves. We examine reinfection trends in South Africa using two
approaches that account for these factors to address the question of whether circulation of
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 4 of 28
the Beta or Delta variants was associated increased reinfection risk, as would be expected
if their emergence was driven by immune escape.
Methods
Data sources
Data analysed in this study come from two sources maintained by the National Institute for
Communicable Diseases (NICD): the outbreak response component of the Notifiable
Medical Conditions Surveillance System (NMC-SS) deduplicated case list and the line list of
repeated SARS-CoV-2 tests. All positive tests conducted in South Africa appear in the
combined data set, regardless of the reason for testing or type of test (PCR or antigen
detection).
Civil unrest during July 2021 severely disrupted testing in Gauteng and KwaZulu-
Natal, the two most populous provinces in the country. As a result, case data became
unreliable and a key assumption of our models - that the force of infection is proportional to
the number of positive tests - was violated. Increasing vaccination rates from August 2021
could also introduce bias. We therefore limited the anlysis to data with specimen receipt
dates between 04 March 2020 and 30 June 2021.
A combination of deterministic (national ID number, names, dates of birth) and
probabilistic linkage methods were utilized to identify repeated tests conducted on the same
person. In addition, provincial COVID-19 contact tracing teams identify and report repeated
SARS-Cov-2 positive tests to the NICD, whether detected via PCR or antigen tests. The
unique COVID-19 case identifier which links all tests from the same person was used to
merge the two datasets. Irreversibly hashed case IDs were generated for each individual in
the merged data set.
Primary infections and suspected repeat infections were identified using the merged
data set. Repeated case IDs in the line list were identified and used to calculate the time
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 5 of 28
between consecutive positive tests for each individual, using specimen receipt dates. If the
time between sequential positive tests was at least 90 days, the more recent positive test
was considered to indicate a suspected new infection. We present a descriptive analysis of
suspected third infections, although only suspected second infections (which we refer to as
“reinfections”) were considered in the analyses of temporal trends. Incidence time series for
primary infections and reinfections are calculated by specimen receipt date of the first
positive test associated with the infection, and total observed incidence is calculated as the
sum of first infections and reinfections. The specimen receipt date was chosen as the
reference point for analysis because it is complete within the data set.
All analyses were conducted in the R statistical programming language (R version
4.0.5 (2021-03-31)).
Data validation
To assess validity of the data linkage procedure and thus verify whether individuals
identified as having suspected reinfections did in fact have positive test results at least 90
days apart, we conducted a manual review of a random sample of suspected second
infections occurring on or before 20 January 2021 (n=585 of 6017; 9.7%). This review
compared fields not used for linkages (address, cell-phone numbers, email addresses,
facility, and health-care providers) between records in the NMC-SS and positive test line
lists. Where uncertainty remained and contact details were available, patients or next-of-kin
were contacted telephonically to verify whether the individual had received multiple positive
test results.
Descriptive analysis
We calculated the time between successive positive tests as the number of days between
the last positive test associated with an individual’s first identified infection (i.e., within 90
days of a previous positive test, if any) and the first positive test associated with their
suspected second infection (i.e., at least 90 days after the most recent positive test).
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 6 of 28
We also compared the age, gender, and province of individuals with suspected
reinfections to individuals eligible for reinfection (i.e., who had a positive test result at least
90 days prior to 30 June 2021).
We did not calculate overall incidence rates by wave because the force of infection is
highly variable in space and time, and the period incidence rate is also influenced by the
temporal pattern of when people become eligible for reinfection. Incidence rate estimates
would therefore be strongly dependent on the time frame of the analysis and not
comparable to studies from other locations or time periods.
Statistical analysis of reinfection trends
We analysed the NICD national SARS-CoV-2 routine surveillance data to evaluate whether
reinfection risk has changed since emergence of the Beta or Delta variants. We evaluated
the daily numbers of suspected reinfections using two approaches. First, we constructed a
simple null model based on the assumption that the reinfection hazard experienced by
previously diagnosed individuals is proportional to the incidence of detected cases and fit
this model to the pattern of reinfections observed before the emergence of the Beta variant
(through 30 September 2020). The null model assumes no change in the reinfection hazard
coefficient through time. We then compared observed reinfections after September 2020 to
expected reinfections under the null model.
Second, we evaluated whether there has been a change in the relative hazard of
reinfection versus primary infection, to distinguish between increased overall transmissibility
of the variants and any additional risk of reinfection due to potential immune escape. To do
this, we calculated an empirical hazard coefficient at each time point for primary infections
and reinfections and compared their relative values through time.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 7 of 28
Approach 1: Catalytic model assuming a constant reinfection hazard coefficient
Model description For a case testing positive on day
!
(by specimen receipt date), we
assumed the reinfection hazard is
"
for each day from
! # $
to
! # %"
and
&'
(
#
for each day
) * ! # %"
, where
'
(
#
is the 7-day moving average of the detected case incidence (first
infections and reinfections) for day
)
. The probability of a case testing positive on day
!
having a diagnosed reinfection by day
+
is thus
,-!.+/ 0 $ 12$
%
&
!"#
!"$%&' '
(
!
, and the expected
number of cases testing positive on day
!
that have had a diagnosed reinfection by day
+
is
')
*,-!.+/
., where
')
*
is the detected case incidence (first infections only) for day
!
. Thus, the
expected cumulative number of reinfections by day
+
is
3+0
4
')
*,-!.+/
),+
),-
. The expected
daily incidence of reinfections on day
+
is
5+0 3+13+$*
.
Model fitting The model was fitted to observed reinfection incidence through 30
September 2020 assuming data are negative binomially distributed with mean
5+
. The
reinfection hazard coefficient (
&
) and the inverse of the negative binomial dispersion
parameter (
6
) are fitted to the data using a Metropolis-Hastings Monte Carlo Markov Chain
(MCMC) estimation procedure implemented in the R Statistical Programming Language.
We ran 4 MCMC chains with random starting values for a total of 1e+05 iterations per
chain, discarding the first 2,000 iterations (burn-in). Convergence was assessed using the
Gelman-Rubin diagnostic (9).
Model-based projection We used 1,500 samples from the joint posterior distribution
of fitted model parameters to simulate possible reinfection time series under the null model,
generating 100 stochastic realizations per parameter set. We then calculated projection
intervals as the middle 95% of daily reinfection numbers across these simulations.
We applied this approach at the national level, as well as to Gauteng, KwaZulu-
Natal, and Western Cape Provinces, which were the only provinces with a sufficient number
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 8 of 28
of reinfections during the fitting period to permit estimation of the reinfection hazard
coefficient.
Approach 2: Empirical estimation of time-varying infection and reinfection hazards
We estimated the time-varying empirical hazard of infection as the daily incidence per
susceptible individual. This approach requires reconstruction of the number of susceptible
individuals through time. We distinguish between three “susceptible” groups: naive
individuals who have not yet been infected (
7*
), previously infected individuals who had
undiagnosed infections (
7.
/
), and previously infected individuals who had a prior positive
test at least 90 days ago (
7.
). We estimate the numbers of individuals in each of these
categories on day
!
as follows:
7*-!/ 0 8 1
9
'0
,123
0,)
0,-
7.
/-!/ 0 -$ 1 ,123/
,123
9
'0
0,)
0,-
7.-!/ 0
9
'0
0,)$!-
0,-
1
9
:0
,123(
0,)
0,-
where
8
is the total population size,
'0
is the number of individuals with their first positive
test on day
;
,
,123
is the probability of detection for individuals who have not had a
previously identified infection,
,123(
is the probability of detection for individuals who have
had a previously identified infection, and
:0
is the number of individuals with a detected
reinfection on day
;
. For the main analysis, we assume
,123 0 "<$
and
,123(0 "<=
, although
the conclusions are robust to these assumptions (see Sensitivity Analysis).
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 9 of 28
Individuals in
7.
/
and
7.
are assumed to experience the same daily hazard of
reinfection, estimated as
>.-!/ 0 4
5
$67)*+(
8(9):
. The daily hazard of infection for previously
uninfected individuals is then estimated as
>*-!/ 0 '
(
$67)*+$;(!8(
,9):
8-9):
.
If we assume that the hazard of infection is proportional to incidence (
?)
),
>*-!/ 0
&*-!/?)
and
>.-!/ 0 &.-!/?)
, we can then examine the infectiousness of the virus through
time as:
&*-!/ 0 >*-!/
-')
@
A,123 #:)
B
A,123(/
&.-!/ 0 >.-!/
-')
@
A,123 #:)
B
A,123(/
We also used this approach to construct a data set with the daily numbers of individuals
eligible to have a suspected second infection (
7.-!/
) and not eligible for suspected second
infection (
7*-!/ # 7.
/-!/
) by wave. Wave periods were defined as the time surrounding the
wave peak for which the 7-day moving average of case numbers was above 15% of the
wave peak. We then analyzed these data using a generalized linear mixed model to
estimate the relative hazard of infection in the population eligible for suspected second
infection, compared to the hazard in the population not eligible for suspected second
infection.
Our primary model was a Poisson model with a log link function, groupinc
0
Poisson
-C/
:
DEF-C/ G
group
H
wave
#
offset
-DEF-
groupsize
//# -
day
/
The outcome variable (groupinc) was the daily number of observed infections in the
two groups. Our main interest for this analysis was in whether the relative hazard was
higher in the second and third waves, thus potentially indicating immune escape. This effect
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 10 of 28
is measured by the interaction term between group and wave. The offset term is used to
ensure that the estimated coefficients can be appropriately interpreted as per capita rates.
We used day as a proxy for force of infection and reporting patterns and examined models
where day was represented as a random effect (to reflect that observed days can be
thought of as samples from a theoretical population) and as a fixed effect (to better match
the Poisson assumptions). As focal estimates from the two models were indistinguishable,
we present only the results based on the random effect assumption. Both versions of the
model are included in the code repository.
Results
We identified 16,029 individuals with at least two suspected infections (through 30 June
2021) and 80 individuals with suspected third infections (Figure 1).
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 11 of 28
Figure 1. Daily numbers of detected primary infections, individuals eligible to be considered for reinfection,
and suspected reinfections in South Africa. A: Time series of detected primary infections. Black line indicates
7-day moving average; black points are daily values. Colored bands represent wave periods, defined as the
period for which the 7-day moving average of cases was at least 15% of the corresponding wave peak (purple
= wave 1, pink = wave 2, orange = wave 3). B: Population at risk for reinfection (individuals whose most
recent positive test was at least 90 days ago and who have not yet had a suspected reinfection). C: Time
series of suspected reinfections. Blue line indicates 7-day moving average; blue points are daily values.
Data validation
Of the 585 randomly selected individuals with possible reinfections in the validation sample,
562 (96%) were verified as the same individual based on fields not used to create the
linkages; the remaining 23 (4%) were either judged not a match or to have insufficient
evidence (details captured by the clinician or testing laboratory) to determine whether the
records belonged to the same individual.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 12 of 28
Descriptive analysis
Time between successive positive tests
The time between successive positive tests for individuals with suspected reinfections was
bimodally distributed with peaks near 180 and 360 days (Figure 2A). The shape of the
distribution was strongly influenced by the timing of South Africa’s epidemic waves. The first
peak corresponds to individuals initially infected in wave 1 and reinfected in wave 2 or
initially infected in wave 2 and reinfected in wave 3, while the second peak corresponds to
individuals initially infected in wave 1 and reinfected in wave 3.
Figure 2. Descriptive analysis of suspected reinfections. A: Time in days between infections for individuals
with suspected reinfection. Note that the time since the previous positive test must be at least 90 days. B:
Percentage of eligible primary infections with suspected reinfections, by province. C: Age distribution of
individuals with suspected reinfections (blue) versus eligible individuals with no detected reinfection (yellow),
by sex. Solid lines indicate females; dashed lines indicate males.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 13 of 28
Distribution of suspected reinfections by province
Suspected reinfections were identified in all nine provinces (Figure 2B). The reinfection rate
was highest in Gauteng, where 5,872 of 415,291 eligible primary infections (1.41%) had
suspected reinfections and lowest in Eastern Cape (1,226 of 195,481; 0.63%). For
comparison, the national reinfection rate was 195,481; 1.03% (16,029 of 1,551,655 eligible
primary infections). Numbers for all provinces are provided in Table S1.
Breakdown of suspected reinfections by sex and age group
Among 1,518,044 eligible primary infections with both age and sex recorded, 9,413 of
877,676 females (1.07%) and 6,573 of 640,368 males (1.03%) had suspected reinfections.
Relative to individuals with no identified reinfection, reinfections were concentrated in adults
between the ages of 20 and 55 years (Figure 2C). Numbers for all age group-sex
combinations are provided in Table S2.
Individuals with multiple suspected reinfections
80 individuals were identified who had three suspected infections. Most of these individuals
initially tested positive during the first wave, with suspected reinfections associated with
waves two and three (Figure S1). No individual had more than two suspected reinfections.
Further details are given in the Supplementary Material (Table S1, Table S2, Figure S1).
Reinfection trends
The first individual became eligible for reinfection on 2020-06-02 (i.e., 90 days after the first
case was detected). No suspected reinfections were detected until 23 June 2020, after
which the number of suspected reinfections increased gradually. The 7-day moving
average of suspected reinfections reached a peak of 162.4 during the second epidemic
wave and a maximum of 304.6 during the third wave, as of 30 June 2021 (Figure 1).
Approach 1: Comparison of data to projections from the null model
Under the null model of no change in the reinfection hazard coefficient through time, the
number of incident reinfections was expected to be low prior to the second wave and to
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 14 of 28
increase substantially during the second and third waves, peaking at a similar time to
incident primary infections. The observed time series of suspected reinfections closely
follows this pattern (Figure 3), although it falls slightly below the prediction interval toward
the end of the time series. Provincial-level analyses suggest that this deviation is driven
primarily by the Western Cape, where the observed time series of suspected reinfections
falls below the prediction interval near the peak of both waves two and three (Figure S3). In
contrast, the observed time series of suspected reinfections consistently falls within the
prediction interval for Gauteng and KwaZulu-Natal (Figure S3). This pattern may result from
policies implemented only in the Western Cape that limited testing during the wave peaks.
Figure 3. Observed and expected temporal trends in reinfection numbers. Blue lines (points) represent the 7-
day moving average (daily values) of suspected reinfections. Grey lines (bands) represent mean predictions
(95% projection intervals) from the null model. A: The null model was fit to data on suspected reinfections
through 2021-09-30, prior to the emergence of the Beta vairant. B: Comparison of data to projections from the
null model over the projection period.
Approach 2: Empirical estimation of time-varying infection and reinfection hazards
The estimated hazard coefficient for primary infection increases steadily through time, as
expected under a combination of relaxing of restrictions, behavioural fatigue, and
introduction of variants with increased transmissibility. The estimated hazard coefficient for
reinfection, in contrast, remains relatively constant, with the exception of an initial spike in
mid-2020, when reinfection numbers were very low. The mean ratio of reinfection hazard to
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 15 of 28
primary infection hazard has decreased slightly with each subsequent wave, from 0.15 in
wave 1 to 0.12 in wave 2 and 0.1 in wave 3. The absolute values of the hazard coefficients
and hazard ratio are sensitive to assumed observation probabilities for primary infections
and reinfections; however, temporal trends are robust (Figure S5).
These findings are consistent with the estimates from the generalized linear mixed
model based on the reconstructed data set. In this analysis, the relative hazard ratio for
wave 2 versus wave 1 was 0.75 (CI
!"
: 0.59-0.97) and for wave 3 versus wave 1 was 0.70
(CI
!"
: 0.55-0.90).
Figure 4. Empirical estimates of infection and reinfection hazards. A: Estimated time-varying hazard
coefficients for primary infection (black) and reinfections (green). Colored bands represent wave periods,
defined as the period for which the 7-day moving average of cases was at least 15% of the corresponding
wave peak (purple = wave 1, pink = wave 2, orange = wave 3). B: Ratio of the empirical hazard for
reinfections to the empirical hazard for primary infections
Discussion
Our analyses suggest that the cumulative number of reinfections observed through June
2021 is consistent with the null model of no change in reinfection risk through time.
Furthermore, our findings suggest that the relative hazard of reinfection versus primary
infection has decreased with each wave of infections, as would be expected if the risk of
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 16 of 28
primary infection increased without a corresponding increase in reinfection risk. Based on
these analyses, we conclude there is no population-level evidence of immune escape at
this time. We recommend ongoing monitoring of these trends.
Differences in the time-varying force of infection, original and subsequent circulating
lineages, testing strategies, and vaccine coverage limit the usefulness of direct
comparisons of rates of reinfections across countries or studies. Reinfection does however
appear to be relatively uncommon. The PCR-confirmed reinfection rate ranged from 0% –
1.1% across eleven studies included in a systematic review (10). While none of the studies
included in the systematic review reported increasing risk of reinfection over time, the
duration of follow-up was less than a year and most studies were completed prior to the
identification of the Beta and Delta variants of concern. Our findings are consistent with
results from the PHIRST-C community cohort study conducted in two locations in South
Africa, which found that infection prior to the second wave provided 84% protection against
reinfection during the second (Beta) wave (11), comparable to estimates of the level of
protection against reinfection for wild type virus from the SIREN study in the UK (1).
A preliminary analysis of reinfection trends in England suggested that the Delta
variant may have a higher risk of reinfection compared to the Alpha variant (12); however,
this analysis did not take into account the temporal trend in the population at risk for
reinfection, which may have biased the findings.
Our findings are somewhat at odds with in vitro neutralization studies. Both the Beta
and Delta variants are associated with decreased neutralization by some anti-receptor
binding-domain (anti-RBD) and anti-N-terminal domain (anti-NTD) monoclonal antibodies
though both Beta and Delta each remain responsive to at least one anti-RBD (4,5,13). In
addition, Beta and Delta are relatively poorly neutralized by convalescent sera obtained
from unvaccinated individuals infected with non-VOC virus (3–5,13). Lastly sera obtained
from individuals after both one and two doses of the BNT162b2 (Pfizer) or ChAdOx1
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 17 of 28
(AstraZeneca) vaccines displayed lower neutralization of the Beta and Delta variants when
compared to non-VOC and Alpha variant (5); although this does not have direct bearing on
reinfection risk it is an important consideration for evaluating immune escape more broadly.
Non-neutralizing antibodies and T-cell responses could explain the apparent disjuncture
between our findings and the in vitro immune escape demonstrated by both Beta and Delta.
Strengths of this study
Our study has two major strengths. Firstly, we analyzed a large routine national data set
comprising all confirmed cases in the country, allowing a comprehensive analysis of
suspected reinfections in the country. Secondly, we found consistent results using two
different analytical methods, both of which accounted for the changing force of infection and
increasing numbers of individuals at risk for reinfection.
Limitations of this study
The primary limitation of this study is that changes in testing practices, health-seeking
behavior, or access to care have not been accounted for in these analyses. Estimates
based on serological data from blood donors suggests substantial geographic variability in
detection rates (14), which may contribute to the observed differences in reinfection
patterns by province. Detection rates likely also vary through time and by other factors
affecting access to testing, which may include occupation, age, and socioeconomic status.
In particular, rapid antigen tests, which were introduced in South Africa in late 2020, may be
under-reported despite mandatory reporting requirements. If under-reporting of antigen
tests was substantial and time-varying it could influence our findings. However, comparing
temporal trends in infection risk among those eligible for reinfection with the rest of the
population, as in approach 2, mitigates against potential failure to detect a substantial
increase in risk.
Reinfections were not confirmed by sequencing or by requiring a negative test
between putative infections. Nevertheless, the 90-day window period between consecutive
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 18 of 28
positive tests reduces the possibility that suspected reinfections were predominantly the
result of prolonged viral shedding. Furthermore, due to data limitations, we were unable to
examine whether symptoms and severity in primary episodes correlate with protection
against subsequent reinfection.
Lastly, while vaccination may increase protection in previously infected individuals
(15–18), vaccination coverage in South Africa was very low during the time of the study
(e.g., <3% of the population was fully vaccinated by 30 June 2021 (19)). Vaccination is
therefore unlikely to have substantially influenced our findings. Increased vaccination
uptake may reduce the risks of both primary infection and reinfection moving forward and
would be an important consideration for application of our approach to other locations with
higher vaccine coverage.
Conclusion
To date, we find no evidence that reinfection risk is higher as a result of the emergence of
Beta or Delta variants of concern, suggesting the selective advantage that allowed these
variants to spread derived primarily from increased transmissibility, rather than immune
escape. The discrepancy between the population-level evidence presented here and
expectations based on laboratory-based neutralization assays highlights the need to
identify better correlates of immunity for assessing immune escape in vitro.
References
1. Hall VJ, Foulkes S, Charlett A, Atti A, Monk EJM, Simmons R, et al. SARS-CoV-2
infection rates of antibody-positive compared with antibody-negative health-care
workers in England: a large, multicentre, prospective cohort study (SIREN). The
Lancet. 2021 Apr 17;397(10283):1459–69.
2. Hansen CH, Michlmayr D, Gubbels SM, Mølbak K, Ethelberg S. Assessment of
protection against reinfection with SARS-CoV-2 among 4 million PCR-tested
individuals in Denmark in 2020: a population-level observational study. The Lancet.
2021 Mar 27;397(10280):1204–12.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 19 of 28
3. Cele S, Gazy I, Jackson L, Hwa S-H, Tegally H, Lustig G, et al. Escape of SARS-CoV-
2 501Y.V2 from neutralization by convalescent plasma. Nature. 2021
May;593(7857):142–6.
4. Wibmer CK, Ayres F, Hermanus T, Madzivhandila M, Kgagudi P, Oosthuysen B, et al.
SARS-CoV-2 501Y.V2 escapes neutralization by South African COVID-19 donor
plasma. Nat Med. 2021 Apr;27(4):622–5.
5. Planas D, Veyer D, Baidaliuk A, Staropoli I, Guivel-Benhassine F, Rajah MM, et al.
Reduced sensitivity of SARS-CoV-2 variant Delta to antibody neutralization. Nature.
2021 Jul 8;1–7.
6. Liu C, Ginn HM, Dejnirattisai W, Supasa P, Wang B, Tuekprakhon A, et al. Reduced
neutralization of SARS-CoV-2 B.1.617 by vaccine and convalescent serum. Cell. 2021
Aug 5;184(16):4220-4236.e13.
7. Tegally H, Wilkinson E, Giovanetti M, Iranzadeh A, Fonseca V, Giandhari J, et al.
Detection of a SARS-CoV-2 variant of concern in South Africa. Nature. 2021
Apr;592(7854):438–43.
8. Shu Y, McCauley J. GISAID: Global initiative on sharing all influenza data – from vision
to reality. Eurosurveillance. 2017 Mar 30;22(13):30494.
9. Gelman A, Rubin DB. Inference from Iterative Simulation Using Multiple Sequences.
Statistical Science. 1992 Nov;7(4):457–72.
10. Murchu EO, Byrne P, Carty PG, De Gascun C, Keogan M, O’Neill M, et al. Quantifying
the risk of SARSICoVI2 reinfection over time. Rev Med Virol [Internet]. 2021 May 27
[cited 2021 Nov 3]; Available from:
https://onlinelibrary.wiley.com/doi/10.1002/rmv.2260
11. Cohen C, Kleynhans J, von Gottberg A, McMorrow ML, Wolter N, Bhiman JN, et al.
SARS-CoV-2 incidence, transmission and reinfection in a rural and an urban setting:
results of the PHIRST-C cohort study, South Africa, 2020-2021 [Internet].
Epidemiology; 2021 Jul [cited 2021 Nov 10]. Available from:
http://medrxiv.org/lookup/doi/10.1101/2021.07.20.21260855
12. Public Health England. SARS-CoV-2 variants of concern and variants under
investigation - Technical briefing 19 [Internet]. Public Health England; 2021 Jul p. 55.
(Technical briefing 19). Report No.: 19. Available from:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachme
nt_data/file/1005517/Technical_Briefing_19.pdf
13. Wang P, Nair MS, Liu L, Iketani S, Luo Y, Guo Y, et al. Antibody resistance of SARS-
CoV-2 variants B.1.351 and B.1.1.7. Nature. 2021 May 6;593(7857):130–5.
14. Vermeulen M, Mhlanga L, Sykes W, Coleman C, Pietersen N, Cable R, et al.
Prevalence of anti-SARS-CoV-2 antibodies among blood donors in South Africa during
the period January-May 2021 [Internet]. In Review; 2021 Aug [cited 2021 Nov 3].
Available from: https://www.researchsquare.com/article/rs-690372/v2
15. Stamatatos L, Czartoski J, Wan Y-H, Homad LJ, Rubin V, Glantz H, et al. mRNA
vaccination boosts cross-variant neutralizing antibodies elicited by SARS-CoV-2
infection. Science. 2021 Jun 25;372(6549):1413–8.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 20 of 28
16. Krammer F, Srivastava K, Alshammary H, Amoako AA, Awawda MH, Beach KF, et al.
Antibody Responses in Seropositive Persons after a Single Dose of SARS-CoV-2
mRNA Vaccine. N Engl J Med. 2021 Apr 8;384(14):1372–4.
17. Saadat S, Rikhtegaran Tehrani Z, Logue J, Newman M, Frieman MB, Harris AD, et al.
Binding and Neutralization Antibody Titers After a Single Vaccine Dose in Health Care
Workers Previously Infected With SARS-CoV-2. JAMA. 2021 Apr 13;325(14):1467.
18. Lustig Y, Nemet I, Kliker L, Zuckerman N, Yishai R, Alroy-Preis S, et al. Neutralizing
Response against Variants after SARS-CoV-2 Infection and One Dose of BNT162b2.
N Engl J Med. 2021 Jun 24;384(25):2453–4.
19. Ritchie H, Mathieu E, Rodés-Guirao L, Appel C, Giattino C, Ortiz-Ospina E, et al.
Coronavirus Pandemic (COVID-19). Our World in Data [Internet]. 2020 Mar 5 [cited
2021 Oct 28]; Available from: https://ourworldindata.org/covid-vaccinations
Ethics statements
Ethical approval
Ethical approval: This study has received ethical clearance from University of the
Witwatersrand (Clearance certificate number M160667) and approval under reciprocal
review from Stellenbosch University (Project ID 19330, Ethics Reference Number
N20/11/074_RECIP_WITS_M160667_COVID-19).
Data availability statement
Data and code are available at https://github.com/jrcpulliam/reinfections. The following data
are included in the repository:
• Counts of reinfections and primary infections by province, age group (5-year bands),
and sex (M, F, U)
• Daily time series of primary infections and suspected reinfections by specimen
receipt date (national)
• Model output: posterior samples from the MCMC fitting procedure and simulation
results
Acknowledgements
The authors wish to acknowledge the members of the NICD Epidemiology and Information
Technology teams which curate, clean, and prepare the data utilized in this analysis.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 21 of 28
Epidemiology team: Andronica Moipone Shonhiwa, Genevie Ntshoe, Joy Ebonwu,
Lactatia Motsuku, Liliwe Shuping, Mazvita Muchengeti, Jackie Kleynhans, Gillian Hunt,
Victor Odhiambo Olago, Husna Ismail, Nevashan Govender, Ann Mathews, Vivien Essel,
Veerle Msimang, Tendesayi Kufa-Chakezha, Nkengafac Villyen Motaze, Natalie Mayet,
Tebogo Mmaborwa Matjokotja, Mzimasi Neti, Tracy Arendse, Teresa Lamola, Itumeleng
Matiea, Darren Muganhiri, Babongile Ndlovu, Khuliso Ravhuhali, Emelda Ramutshila,
Salaminah Mhlanga, Akhona Mzoneli, Nimesh Naran, Trisha Whitbread, Mpho Moeti,
Chidozie Iwu, Eva Mathatha, Fhatuwani Gavhi, Masingita Makamu, Matimba Makhubele,
Simbulele Mdleleni, Bracha Chiger, Jackie Kleynhans
Information Technology team: Tsumbedzo Mukange, Trevor Bell, Lincoln Darwin,
Fazil McKenna, Ndivhuwo Munava, Muzammil Raza Bano, Themba Ngobeni
We also thank Carl A.B. Pearson, Shade Horn, Youngji Jo, Belinda Lombard, Liz S.
Villabona-Arenas, and colleagues in the SARS-CoV-2 variants research consortium in
South Africa for helpful discussions during the development of this work.
Footnotes
Author contributions
Conceptualization - JP, CvS, JD, HM
Data collection, management, and validation - NG, KM, AvG, CC
Data analysis - JP, CvS, JD
Interpretation - JP, AvG, CC, MJG, JD, HM
Drafting the manuscript - JP
Manuscript review, revision, and approval - all authors
Guarantor: HM
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 22 of 28
Funding
JRCP and CvS are supported by the South African Department of Science and Innovation
and the National Research Foundation. Any opinion, finding, and conclusion or
recommendation expressed in this material is that of the authors and the NRF does not
accept any liability in this regard. This work was also supported by the Wellcome Trust
(grant number 221003/Z/20/Z) in collaboration with the Foreign, Commonwealth and
Development Office, United Kingdom.
Competing interests
All authors have completed the ICMJE uniform disclosure form. CC and AvG have received
funding from Sanofi Pasteur in the past 36 months. JRCP and KM serve on the Ministerial
Advisory Committee on COVID-19 of the South African National Department of Health. The
authors have declared no other relationships or activities that could appear to have
influenced the submitted work.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 23 of 28
Supplementary Material
Distribution of suspected reinfections by province, South Africa, March
2020 to June 2021
Province
No reinfection
One reinfection
Two reinfections
Total
EASTERN CAPE
194,255
1,224
2
195,481
FREE STATE
82,769
842
3
83,614
GAUTENG
409,419
5,832
40
415,291
KWAZULU-NATAL
332,074
2,437
11
334,522
LIMPOPO
62,800
503
1
63,304
MPUMALANGA
74,744
746
4
75,494
NORTH WEST
63,200
892
7
64,099
NORTHERN CAPE
36,126
413
1
36,540
WESTERN CAPE
280,237
3,060
11
283,308
UNKNOWN
2
0
0
2
Total
1,535,626
15,949
80
1,551,655
Breakdown of suspected reinfections by sex and age group (years),
South Africa, March 2020 to June 2021
Sex
Age group
No reinfection
One reinfection
Two reinfections
Total
F
(0,20]
84,241
506
3
84,750
F
(20,40]
359,483
4,776
27
364,286
F
(40,60]
303,546
3,366
8
306,920
F
(60,80]
104,507
620
6
105,133
F
(80,Inf]
16,486
101
0
16,587
M
(0,20]
67,956
369
2
68,327
M
(20,40]
247,359
3,039
15
250,413
M
(40,60]
230,546
2,496
17
233,059
M
(60,80]
79,777
588
2
80,367
M
(80,Inf]
8,157
45
0
8,202
Total
1,502,058
15,906
80
1,518,044
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 24 of 28
Individuals with multiple suspected reinfections
Figure S1. Timing of infections for individuals with multiple suspected reinfections. Circles represent the first
positive test of the first detected infection; triangles represent the first positive test of the suspected second
infection; squares represent the first positive test of the suspected third infection. Colored bands represent
wave periods, defined as the period for which the 7-day moving average of cases was at least 15% of the
corresponding wave peak (purple = wave 1, pink = wave 2, orange = wave 3).
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 25 of 28
Timing of primary infections and reinfections by province
Figure S2. Number of detected primary infections (black) and suspected reinfections (blue), by province.
Lines represent 7-day moving averages. The y-axis is shown on a log scale.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 26 of 28
Province-level comparison of data to projections from the null model
Figure S3. Observed and expected temporal trends in reinfection numbers, for provinces with sufficient
numbers of suspected reinfections. Blue lines (points) represent the 7-day moving average (daily values) of
suspected reinfections. Grey lines (bands) represent mean predictions (95% projection intervals) from the null
model. A and B: Gauteng. C and D: KwaZulu-Natal. E and F: Western Cape.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 27 of 28
Approach 1: Convergence diagnostics
Figure S4. Convergence diagnostics and density of the posterior distribution for MCMC fits. A and B: MCMC
chains for each parameter. C: Gelman-Rubin values (a.k.a. potential scale reduction factors) for each
parameter; values less than 1.1 indicate sufficient mixing of chains to suggest convergence. D, G, I: posterior
density for each parameter and the log likelihood. E, F, H: 2-D density plots showing correlations between
parameters and the log likelihood.
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
Page 28 of 28
Approach 2: Sensitivity analysis
Figure S5. Sensitivity analysis of empirical hazard ratio estimates to assumed observation probabilities for
primary infections and reinfections. Estimates are shown for the full range of probabilities for which the overall
mean relative hazard is between 0 and 1. The white polygon encloses the most plausible estimates
(i.e. consistent with relative reinfection risk observed in the SIREN study (1) and observation probabilities for
primary infection consistent with estimates based on seroprevalence data (14)). Top: Mean relative empirical
hazard for reinfections versus primary infections in each wave, as a function of assumed observation
probabilities for primary infections (
𝑝./0
) and reinfections (
𝑝./0!
). A: wave 1, B: wave 2, C: wave 3. Bottom:
Percent change in the mean relative empirical hazard for reinfections versus primary infections in waves 2 (D)
and 3 (E) relative to wave 1, as a function of assumed observation probabilities for primary infections (
𝑝./0
)
and reinfections (
𝑝./0!
).
. CC-BY-NC 4.0 International licenseIt is made available under a
perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted November 11, 2021. ; https://doi.org/10.1101/2021.11.11.21266068doi: medRxiv preprint
