IEA

International Journal of Epidemiology, 2022, 1–10

https://doi.org/10.1093/ije/dyac199
Original article
International Epidemiological Association

Original article

Waning of first- and second-dose ChAdOx1 and

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

BNT162b2 COVID-19 vaccinations: a pooled
target trial study of 12.9 million individuals in
England, Northern Ireland, Scotland and Wales
Steven Kerr,1,† Stuart Bedston,2,† Declan T Bradley,3,4,† Mark Joy,5,†
Emily Lowthian ,2,6,† Rachel M Mulholland,1,† Ashley Akbari ,2
FD Richard Hobbs,5 Srinivasa Vittal Katikireddi ,7
Simon de Lusignan ,5 Igor Rudan,1 Fatemeh Torabi,2 Ruby SM Tsang,5
Ronan A Lyons,2 Chris Robertson 8,9,† and Aziz Sheikh1,10
1
Centre for Medical Informatics, Usher Institute, The University of Edinburgh, Edinburgh, UK,
2
Population Data Science, Swansea University Medical School, Swansea University, Swansea, UK,
3
School of Medicine, Dentistry and Biomedical Sciences, Queen’s University Belfast, Belfast, UK,
4
Public Health Agency, Belfast, UK, 5Nuffield Department of Primary Care Health Sciences, University
of Oxford, Oxford, UK, 6Department of Education and Childhood Studies, Swansea University,
Swansea, UK, 7MRC/CSO Social & Public Health Sciences Unit, University of Glasgow, Glasgow, UK,
8
Public Health Scotland, Glasgow, UK, 9Department of Mathematics and Statistics, University of
Strathclyde, Glasgow, UK and 10BREATHE—The Health Data Research Hub for Respiratory Health, The
University of Edinburgh, Edinburgh, UK
*Corresponding author. Centre for Medical Informatics, Usher Institute, NINE Edinburgh BioQuarter, 9 Little France Road,
Edinburgh, EH16 4UX, UK. E-mail: [email protected]
†
These authors contributed equally to this work.

Received 22 April 2022; Editorial decision 20 September 2022; Accepted 30 September 2022

Abstract
Background: Several SARS-CoV-2 vaccines have been shown to provide protection
against COVID-19 hospitalization and death. However, some evidence suggests that no-
table waning in effectiveness against these outcomes occurs within months of vaccina-
tion. We undertook a pooled analysis across the four nations of the UK to investigate
waning in vaccine effectiveness (VE) and relative vaccine effectiveness (rVE) against
severe COVID-19 outcomes.
Methods: We carried out a target trial design for first/second doses of ChAdOx1(Oxford–
AstraZeneca) and BNT162b2 (Pfizer–BioNTech) with a composite outcome of COVID-19
hospitalization or death over the period 8 December 2020 to 30 June 2021. Exposure
groups were matched by age, local authority area and propensity for vaccination. We
pooled event counts across the four UK nations.

C The Author(s) 2022. Published by Oxford University Press on behalf of the International Epidemiological Association.
V 1
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits
unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
2 International Journal of Epidemiology, 2022, Vol. 00, No. 00

Results: For Doses 1 and 2 of ChAdOx1 and Dose 1 of BNT162b2, VE/rVE reached zero by
approximately Days 60–80 and then went negative. By Day 70, VE/rVE was –25% (95% CI:
–80 to 14) and 10% (95% CI: –32 to 39) for Doses 1 and 2 of ChAdOx1, respectively, and
42% (95% CI: 9 to 64) and 53% (95% CI: 26 to 70) for Doses 1 and 2 of BNT162b2, respec-
tively. rVE for Dose 2 of BNT162b2 remained above zero throughout and reached 46%
(95% CI: 13 to 67) after 98 days of follow-up.
Conclusions: We found strong evidence of waning in VE/rVE for Doses 1 and 2 of
ChAdOx1, as well as Dose 1 of BNT162b2. This evidence may be used to inform policies
on timings of additional doses of vaccine.

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

Key words: COVID-19, vaccine effectiveness, vaccine waning

Key Messages

• We undertook an observational epidemiological analysis of the effectiveness of COVID-19 vaccines across all four

nations of the UK, pooling data from a UK cohort of 12.9 million individuals.
• We carried out a target trial study of waning in vaccine effectiveness (VE)/relative vaccine effectiveness (rVE) against

COVID-19 hospitalization or death for first/second doses of ChAdOx1 (Oxford–AstraZeneca) and BNT162b2 (Pfizer–
BioNTech).
• For Doses 1 and 2 of ChAdOx1, as well as Dose 1 of BNT162b2, VE/rVE reached zero by approximately Days 60–80

after vaccination, whereas for Dose 2 of BNT162b2, rVE remained above zero throughout 98 days of follow-up.
• Our methodology provides proof of concept for carrying out pooled studies where data are stored in different

locations and sharing of individual-level data is not permitted.

Introduction 4 months.8 A cross-country study found evidence of wan-

In December 2019, a novel coronavirus (SARS-CoV-2) ing in effectiveness of second-dose ChAdOx1 against se-
emerged in Wuhan, China.1 The World Health Organization vere COVID-19 outcomes, with vaccine effectiveness (VE)
declared the outbreak a Public Health Emergency of of 63.7% in Scotland and 42.2% in Brazil at 18–19 weeks
International Concern on 30 January 2020 and then a pan- after the second dose.9 A meta-analysis of 18 studies on VE
demic on 11 March 2020. In the UK as of 28 June 2022, found that protection against severe COVID-19 disease de-
there have been >22 million reverse-transcriptase polymerase creased on average by 10.0% between 1 and 6 months af-
chain reaction (RT-PCR)-confirmed COVID-19 cases, ter full vaccination.10
>880 000 COVID-19 hospitalizations and >170 000 It is important to determine the timescale over which vac-
COVID-19 deaths.2 cines provide high levels of protection in order to inform
COVID-19 vaccines have been developed in record time. policymaking regarding dosing schedules. To our knowl-
Three vaccines have thus far been administered at scale in the edge, studies from the UK have hitherto analysed data from
UK: ChAdOx1 nCoV-19 (Oxford–AstraZeneca, hereafter a single country, e.g. Scotland9 or England,11 with no multi-
ChAdOx1), BNT162b2 (Pfizer–BioNTech, hereafter nation studies across the UK. The aim of this study was to
BNT162b2) and mRNA-1273 (Moderna). These vaccines investigate waning in VE/relative VE (rVE) against severe
have demonstrated high levels of effectiveness against a num- COVID-19 outcomes using pooled data from across the
ber of outcomes including infection, hospitalization and four nations of the UK using a target trial approach.
death, in both clinical trials and observational epidemiologi-
cal studies.3–7 Methods
However, there is evidence that vaccine protection
against SARS-CoV-2 infection and severe outcomes wanes Study design and population
notably over time. A study of the workforce of the We carried out a target trial analysis. Target trials attempt
University of California San Diego Health found that the to emulate a clinical trial by finding naturally occurring ex-
effectiveness of two doses of BNT162b2 or mRNA-1273 posure groups.12 Individuals exposed to a dose of vaccine
against symptomatic infection reduced to 65% after were statistically matched 1:1 to unexposed individuals
International Journal of Epidemiology, 2022, Vol. 00, No. 00 3

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

Figure 1 Data linkage diagram. ADDD, Annual District Death Daily; ADDE, Annual District Death Extract; CCDS, Critical Care Dataset; CDDS, COVID-19
Consolidated Deaths; ConCov, Controlling COVID-19 through enhanced population surveillance and intervention; CVVD, Covid Vaccination Dataset; EAVE,
Early Assessment of Vaccine and antiviral Effectiveness; ECOSS, Electronic Communication of Surveillance in Scotland; EDDS, Emergency Department
Dataset; HES, Hospital Episode Statistics; ICCD, Intensive Care National Research and Audit Centre, Weekly Covid-only; ICNC, Intensive Care National
Research and Audit Centre; NHAIS, National Health Application and Infrastructure Services; NHS, National Health Service; NIMS, National Immunisation
Management System; NISRA, Northern Ireland Statistics and Research Agency; NRS, National Records of Scotland; ONS, Office for National Statistics;
PATD, Pathology COVID-19 Daily; PEDW, Patient Episode Database for Wales; PHS, Public Health Scotland; RCGP RSC, Oxford-Royal College of General
Practitioners Research and Surveillance Centre; SICSAG, Scottish Intensive Care Society Audit Group; SMR01, Scottish Morbidity Records 01; SUS,
Secondary Users Service; TVMT, Turas Vaccine Management Tool; VMS, Vaccine Management System; WDSD, Welsh Demographic Service Dataset

based on a number of clinical and demographic character- graphically rather than focus on a binary definition. The data
istics. Differences in the composite outcome variable of sets consisted of linked primary care, secondary care, mortal-
COVID-19 hospitalization or death were then compared ity and virological testing data stored in secure TREs in each
between these groups. of England, Northern Ireland, Scotland and Wales (Figure 1).
We followed a pre-specified statistical analysis plan These data were deterministically linked using unique, anony-
(Supplementary File S1, Statistical analysis plan, available as mized patient identifiers—National Health Service (NHS)
Supplementary data at IJE online). Initially, we sought to number in England, Health and Care Number in Northern
study only first-dose waning. However, navigating permis- Ireland and Community Health Index number in Scotland. In
sions to undertake this analysis across national trusted re- Wales, a combination of deterministic linkage based on NHS
search environments (TREs) took far longer than anticipated number and probabilistic linkage based on personal identi-
and by the time we were able to carry out the analysis, many fiers was used. The study period was 8 December 2020 to
people had received their second dose. Therefore, we also 30 June 2021. Anyone whose most recently recorded age at
studied second dose and we looked at a longer study period the cohort start date was aged <18 years was excluded. We
than was originally specified in the statistical analysis plan. also excluded elderly care home residents because care home
We initially planned to do a secondary cohort analysis in or- residence may have been associated with strong confounding
der to make use of the full data set. However, as we navi- effects and there were very few care home residents who
gated permissions to undertake this work it became clear that were suitable to be used as controls because this population
it would not have been possible to do a pooled cohort study was a high priority group for vaccination.
as this would require sharing non-count data. Therefore, we
carried out the primary target trial analysis. We also planned Exposure
to use two separate binary definitions of vaccine waning. We defined an individual as exposed according to the type
However, after conducting the analysis, we decided that it of vaccine and dose number they received, starting from
would be most informative to present our main results the 14 days after it was administered.
4 International Journal of Epidemiology, 2022, Vol. 00, No. 00

Outcomes prior to the matching period. In England, Scotland and

The primary outcome was a composite of incident Wales, smoking status, BMI and number of QCovid risk
COVID-19 hospitalization or death. In England, groups were also included in the propensity model. QCovid
Scotland and Wales, COVID-19 hospitalization was an is a tool for predicting the risk of COVID-19 hospitalization
RT-PCR-confirmed positive test for SARS-CoV-2 in the and death that takes into account a range of demographic
and clinical variables and has been used to inform policy
28 days prior to admission or hospitalization with an
deliberations on shielding and vaccine prioritization in
International Classification of Diseases (ICD-10) code for
England.20 In Northern Ireland, the number of chapters of
COVID-19 in any diagnostic position. ICD-10 codes for
the British National Formulary (BNF) from which individuals

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

COVID-19 were U07.1 and U07.2. In Northern Ireland,
received repeat prescriptions prior to the vaccination pro-
COVID-19 hospitalization was an RT-PCR-confirmed
gramme was included as a proxy for co-morbidity. To be in-
positive test for SARS-CoV-2 from 14 days prior to ad-
cluded in the BNF prescription count, a medicine had to be
mission to 7 days after admission or hospitalization with
prescribed in both of two 3-month periods in the 6 months
an ICD-10 code for COVID-19 in any diagnostic posi-
before the start of the vaccination programme in the UK,
tion. COVID-19 death was defined as death within
8 December 2020. Medications related to contraceptives
28 days of a positive RT-PCR test for SARS-CoV-2 infec-
(BNF chapter 7, section 3) were removed as these do not indi-
tion or death with COVID-19 as the underlying ICD-10
cate an illness. This method was adapted from one validated
cause of death recorded on the death certificate. We se-
in other multimorbidity studies using administrative data.21
lected the first event for each person in the study period.
Within each nation and for each dose, individuals were
The event date was whichever came first out of the date
matched on their propensity to get vaccinated, area of resi-
of hospital admission or the date of death.
dence and age. Matching on vaccination propensity was by
single percentile bands of the propensity score and match-
Statistical analysis ing on age was by individual years up to 79 years, then 2-
year bands from ages 80 to 89 years, 5-year bands for those
In Scotland and Wales, only the body mass index (BMI)
aged 90–99 years and one final band for all those aged
variable had missing values and these were imputed using
100 years. Due to rapid uptake, an exception was made
ordinary least squares regression. In Northern Ireland and
for second-dose analysis in Wales, where matching was
England, a complete case analysis was done.
based on two-percentile propensity bands, by year of age
We used time-varying matching in approximately 1-
up to 59 years and 5-year bands for 60–99 years and all
month-long intervals to construct the exposure groups of
those aged 100 years. Whenever a case had more than
the target trial. Administration of vaccines in the UK
one candidate control, one was selected at random, with
started on 8 December 2020 for BNT162b2 and the restriction that an individual could only be used as a
4 January 2021 for ChAdOx1. Therefore, we took the in- control once. Matching was assessed using covariate bal-
terval between these two dates to be the first matching pe- ance plots.
riod for the propensity model and then full months Individuals were censored if they had a non-COVID-19
thereafter. Within each time period, a propensity score for death. Matched pairs were jointly censored whenever ei-
receiving the next vaccine dose was calculated. For the ther individual received the next dose of the vaccine.
first-dose analysis, individuals who received the first dose We fitted Poisson models for each exposure group to es-
of a given vaccine type were matched with individuals who timate the rate of incident events. The Poisson model in-
were unvaccinated on the date the vaccine was adminis- cluded a spline in days since start of follow-up, an offset
tered. For the second-dose analysis, individuals who re- for total person-days of follow-up and a vaccination expo-
ceived the second dose of a given vaccine type were sure group as a stratification variable. We used this model
matched with individuals who had received only one dose to predict event rates for each exposure group by day of
of the same vaccine type at that time. follow-up. We then divided these rates to obtain rate ra-
Analysts in each nation had full access to that nation’s tios. Under the model assumptions, the logarithm of this
data. In all nations, the following predictors were included rate ratio is asymptotically normally distributed with vari-
in the propensity model: sex, age, local authority area, ance that can be calculated from the covariance matrix of
urban/rural classification,13–15 quantiles of multiple depriva- the parameter estimates. We used this to construct 95%
tion index,16–19 number of previous SARS-CoV-2 tests CIs. VE was calculated as 1 – (Rate Ratio). We also fit a
pre-vaccination, mean household age, number of people in second model with a quadratic in time instead of a spline
household, presence in hospital for any reason 4 weeks prior and we tested whether the coefficient on the squared term
to the matching period and positive RT-PCR test at any time was equal to zero for the exposed group in order to assess
International Journal of Epidemiology, 2022, Vol. 00, No. 00 5

waning. This analysis was repeated in each nation of the available as Supplementary data at IJE online) give event
UK. counts by vaccination status in each nation.
The data governance procedure of the TREs in each Table 1 shows the number and proportion of individu-
country did not allow individual-level data to be shared. als who were exposed to each vaccine dose that were suc-
However, we obtained permission from data controllers in cessfully matched with a control by nation. The proportion
each nation to confidentially share count data with the of exposed individuals that were matched ranged from
Scottish TRE on the condition that these would be com- 45–58% with the exception of second-dose matching in
bined in a pooled count. In each nation’s TRE, counts of Wales (17%). Descriptive tables of matched exposure
outcome events and person-days of follow-up were col- groups for each country are given in Tables S5a–h in

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

lated by vaccine type, target trial arm, age group (18–64, Supplementary File S5 (Exposure group summary tables,
65–79, 80þ years) and day of follow-up. These were then available as Supplementary data at IJE online). The lower
gathered in the Scottish TRE and summed across the four matching rate for the second-dose analysis in Wales was
nations. A similar analysis was then carried out on these likely due to especially rapid vaccine uptake and the re-
pooled data. The target trial design meant that confound- striction that an individual could only be used as a control
ers could be controlled for by matching without each na- once. Covariate balance plots by vaccine type, dose and
tion having to share individual-level data and the Poisson country are provided in Supplementary File S6 (Covariate
modelling strategy meant that only counts of events and balance, available as Supplementary data at IJE online).
person-years were required, thus allowing a pooled analy- Plots of VE/rVE over time by vaccine type, dose and
sis to be carried out sharing only count data. country are provided in Supplementary File S7 (VE/rVE by
country, available as Supplementary data at IJE online).
Table 2 and Figures 2–5 show VE/rVE in the pooled analy-
Reporting sis, as well as P-values for testing that the coefficient on the
squared term in time is zero for the exposed group in the
This study is reported in accordance with the REporting of
model that included a quadratic in time. VE/rVE reached
studies Conducted using Observational Routinely-
zero by approximately Days 60–80 for Doses 1 and 2 of
collected Data (RECORD) guidelines (Supplementary File
ChAdOx1 and Dose 1 of BNT162b2. rVE for Dose 2
S2, Checklist, available as Supplementary data at IJE
of BNT162b2 remained above zero throughout 98 days of
online).22,23
follow-up. However, in the models that included quadratic
terms in time, the P-value for the coefficient on the squared
terms for the exposed group were quite high for both vac-
Results
cines and doses, ranging from 0.47 to 0.95. Figure S8a–l in
Supplementary Tables S3a–d in Supplementary File S3 the Supplementary File S8 (Pooled VE/rVE by age group,
(Cohort summary tables, available as Supplementary data available as Supplementary data at IJE online) show
at IJE online) show the marginal distributions of a number pooled VE/rVE stratified by age group (18–64, 65–79,
of characteristics in each country’s cohort. Clinical risk 80þ years). These results were qualitatively similar, but
groups in these tables were derived from the QCovid algo- estimates were less precise than the combined analysis for
rithm.20 There were no notable differences in the marginal all age groups.
distributions of these characteristics by country with the
exception of BMI, with Scotland tending to have a greater
proportion of the cohort in the overweight category and Discussion
fewer in the obese category compared with England and We carried out a pooled epidemiological analysis of linked,
Wales (BMI data were not available for Northern Ireland). pseudonymized national-level vaccination data across the
Tables S4a–d in Supplementary File S4 (Event count tables, four nations of the UK. We employed a novel methodology

Table 1 Number of exposed individuals who were successfully matched

First-dose vaccinated First-dose matched Second-dose vaccinated Second-dose matched

England 4 223 375 2 283 348 (54.1%) 3 632 725 1 658 061 (45.6%)
Northern Ireland 1 105 511 640 964 (58.0%) 844 018 393 820 (46.7%)
Scotland 3 481 808 1 650 088 (47.4%) 2 582 105 1 358 286 (52.6%)
Wales 1 629 997 912 704 (56.0%) 1 239 608 216 608 (17.5%)
6 International Journal of Epidemiology, 2022, Vol. 00, No. 00

Table 2 Vaccine effectiveness and relative vaccine effectiveness in pooled study

Day First dose ChAdOx1a Second dose ChAdOx1a First dose BNT162b2b Second dose BNT162b2b

14 47 (50 to 52) 38 (22 to 50) 64 (60 to 67) 71 (63 to 78)

28 50 (38 to 60) 36 (15 to 52) 69 (61 to 75) 70 (58 to 79)
42 36 (15 to 51) 33 (8 to 51) 62 (49 to 71) 60 (42 to 72)
56 50 (29 to 65) 26 (–4 to 47) 55 (37 to 68) 54 (30 to 69)
70 –25 (–80 to 14) 10 (–32 to 39) 42 (9 to 64) 53 (26 to 70)
84 –185 (–73 to –369) –19 (–94 to 27) –121 (–14 to –326) 58 (30 to 75)

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

98 –89 (–15 to –212) –59 (–247 to 27) –85 (1 to 46) 46 (13 to 67)

a
Oxford–AstraZeneca.
b
Pfizer–BioNTech.

Figure 2 Pooled rate ratios for COVID-19 hospitalization or death, first-dose ChAdOx1 (Oxford–AstraZeneca), all ages

Figure 3 Pooled rate ratios for COVID-19 hospitalization or death, first-dose BNT162b2 (Pfizer–BioNTech), all ages

Figure 4 Pooled rate ratios for COVID-19 hospitalization or death, second-dose ChAdOx1 (Oxford–AstraZeneca), all ages

to allow a pooled study to be done with only count data becoming negative thereafter. Our rVE estimates for Dose
being shared between each country’s TREs. We found evi- 2 of BNT162b2 remained above zero throughout 98 days
dence of waning in VE/rVE for Doses 1 and 2 of ChAdOx1 of follow-up.
and Dose 1 of BNT162b2, with VE/rVE dropping to zero We believe that the most likely explanation for negative
60–80 days after the date of administration and VE/rVE is that vaccination caused recipients to believe
International Journal of Epidemiology, 2022, Vol. 00, No. 00 7

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

Figure 5 Pooled rate ratios for COVID-19 hospitalization or death, second-dose BNT162b2 (Pfizer–BioNTech), all ages

they were protected, leading them to change their behav- exposure groups in the population. This design together
iour in ways that increase their chance of contracting the with the Poisson regression modelling strategy meant that
infection. These changes in behaviours should initially the only data that were required in the analysis were
have been outweighed by the protection offered by the im- count-level data from the exposure groups in each nation.
mune response stimulated by the vaccine, but as time pro- We obtained permission to confidentially share counts of
gressed the protection is likely to have diminished such individuals stratified by categories between TREs in each
that the impact of behavioural changes may have become nation on the understanding that these would be combined
dominant. It is also possible that naturally acquired immu- in a pooled count. This enabled us to conduct a pooled
nity provides more robust protection than vaccination.24 analysis across the four nations of the UK. Typically, the
Our VE/rVE estimates were lower than estimates of vac- only way to combine results from separate analyses on
cine efficacy seen in clinical trials. Clinical trials for both individual-level data that cannot be shared is to do a meta-
ChAdOx1 and BNT162b2 reported 100% efficacy against analysis. Our use of splines in time allowed us to model
severe COVID-19 outcomes after a two-dose regimen.5,6 complex trajectories for VE/rVE. A final strength of this
There are several possible explanations for this. These clin- paper is that we used time-varying, incident density sam-
ical trials were carried out during different time periods pling in our propensity score matching procedure to ac-
and in different populations compared with our study. In count for changes in vaccine prioritization over time.
particular, the dominant variants in circulation were not A limitation in our analysis is that there was limited
the same. The clinical trials included a blinding procedure, follow-up in populations that were prioritized for vaccina-
where the control group was given a placebo injection. tion, particularly the elderly. Matched pairs in older age
Our study examined real-world effectiveness, where a pla- groups tended to be censored relatively quickly, contribut-
cebo was not administered. This could have led to behav- ing to imprecision in our estimates of VE/rVE. This was
ioural confounding that might be expected to make VE particularly true in the second-dose analysis in Wales, for
estimates lower than vaccine efficacy. Our first-dose VE which the matching percentage was only 17% due to an es-
estimates tended to be lower compared with other esti- pecially rapid rollout there. However, Scotland and
mates in the literature.25 One major difference is that we England contributed most of the data that were used in the
used a target trial methodology as opposed to other com- pooled analysis and data from Wales likely had a relatively
mon observational epidemiological designs such as cohort minor effect on the results. We considered allowing indi-
and case–control studies. We have previously carried out a viduals to be used multiple times in the matching, but we
cohort study of vaccine waning for two doses of ChAdOx1 believe this would have led to unacceptable distortion of
on the same EAVE II data set, in tandem with a separate CIs on our estimates. Target trials tend to offer better con-
data set from Brazil.9 Our rVE estimates are consistent trol over confounders at the cost of lower statistical power
with results from this previous study, as well as a cohort and it is not uncommon for a significant proportion of the
study of second-dose waning of BNT162b2 in Israel.26 study population to be lost when creating exposure groups
Our results may also be broadly consistent with a target in target trials. There was also a potential selection bias in
trial study of VE of BNT162b2 in Israel, although a direct that those who are matched are not representative of those
comparison cannot be made for second dose because we who are vaccinated. Comparison of Table S3 with S5
estimated VE relative to those who had received a first (available as Supplementary data at IJE online) provides
dose, as opposed to the unvaccinated.27 A key strength of information on this and generally there is reasonable agree-
this paper is the use of the target trial method, which seeks ment. It is possible that younger people were more likely to
to emulate a clinical trial by finding naturally occurring be matched and this may have been due to the elderly being
8 International Journal of Epidemiology, 2022, Vol. 00, No. 00

a smaller group that was prioritized for vaccination. It is vaccine. We have also used a novel methodology that
not clear in which direction this bias may have worked. allowed us to carry out a pooled study—for the first
Whereas England, Scotland and Wales had access to time—across all UK nations without having to share
QCovid risk group variables for use in the propensity individual-level data. This will we hope serve as a proof of
matching, Northern Ireland did not and BNF chapters pre- concept for further pooled multicentre/country studies in
scribed was used as a proxy. Although pooling data across the future.
the nations noticeably improved the precision in our esti-
mates, there was still significant uncertainty, particularly
in the age-stratified analyses and at large times elapsed Ethics approval

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

since exposure. The Alpha variant was dominant in the UK In England, approvals were obtained from the Health Research
until May 2021, switching to the Delta variant thereafter. Authority, London Central (reference number 21/HRA/2786). In
Scotland, data approvals were obtained from the National Research
This change in the dominant variant part-way through our
Ethics Service Committee, Southeast Scotland 02 (reference number:
study period meant it was difficult to determine the extent
12/SS/0201) and the Public Benefit and Privacy Panel for Health and
to which waning in VE was due solely to the passage of Social Care (reference number: 1920–0279). In Northern Ireland,
time vs the emergence of new variants and potential vac- approval was by the Honest Broker Service Governance Board
cine escape. Our analysis of first-dose waning indicated (Project 064). In Wales, approval was provided by Secure
that VE changes non-trivially with time. In the second-dose Anonymised Information Linkage (SAIL) independent Information
Governance Review Panel (IGRP) (Project 0911).
analysis, we considered matching exposure group pairs by
date of first-dose vaccination to control for VE of the first
dose changing with time. However, this resulted in very Data availability
few matches and very little follow-up time. We believe it is
All code, metadata and documentation for this project is publicly
likely that matching by propensity score for vaccination
available at https://github.com/EAVE-II/Covid-vaccine-waning-
controlled adequately for this confounder because the dos- pooled. Most of the data that were used in this study are highly sen-
age schedule was similar for most people who were vacci- sitive and will not be made available publicly.
nated. Our estimates of VE/rVE against COVID-19
hospitalization or death tended to be notably high immedi-
ately after vaccination, on a timescale that was too short to Supplementary data
be plausibly explained by the immune response generated Supplementary data are available at IJE online.
by the vaccine. This may have been due to behavioural
changes associated with the vaccination programme.
People who contracted COVID-19 prior to a scheduled
Author contributions
vaccination may have postponed vaccination, causing early A.S. and C.R. conceived of this study. The analysis in England was
carried out by M.J. The analysis in Northern Ireland was carried out
VE/rVE to be biased upwards. In addition, some recipients
by D.T.B. The analysis in Scotland was carried out by C.R., S.K.
may have suffered mild illness following vaccination, caus-
and R.M. The analysis in Wales was carried out by S.B. and E.L.
ing them to reduce their social contact, reducing the chance The pooled analysis was carried out by S.K. S.K., S.B. and E.L.
of contracting COVID-19. Other behavioural explanations wrote an initial draft of the manuscript. All authors edited later ver-
have also been proposed for this phenomenon.28 On the sions of the manuscript.
other hand, it is possible that there were people who had
an event shortly after vaccination but were infected prior
to vaccination. This effect would work in the opposite di-
Funding
rection, lowering VE/rVE estimates in a period immedi- This research is part of the Data and Connectivity National Core
Study, led by Health Data Research UK in partnership with the
ately following vaccination. Finally, it is possible that there
Office for National Statistics and funded by UK Research and
was residual confounding that was not fully accounted for
Innovation (grant ref MC_PC_20058). The Honest Broker Service
by the target trial study design. (HBS) is funded by the Business Services Organisation (BSO) and
This paper contributes additional robust evidence on the Department of Health for Northern Ireland. This work was sup-
the waning of VE/rVE for first and second doses of ported by the Con-COV team funded by the Medical Research
ChAdOx1 and BNT162b2 vaccines. In particular, if pro- Council (grant number: MR/V028367/1). This work was supported
tection against severe COVID-19 wanes over the course of by Health Data Research UK, which receives its funding from HDR
UK Ltd (HDR-9006) funded by the UK Medical Research Council,
months, then COVID-19 vaccination may have greatest
Engineering and Physical Sciences Research Council, Economic and
utility as a tool for ‘flattening the curve’. Although we Social Research Council, Department of Health and Social Care
studied first- and second-dose vaccination only, the evi- (England), Chief Scientist Office of the Scottish Government Health
dence here may still be informative for future doses of and Social Care Directorates, Health and Social Care Research and
International Journal of Epidemiology, 2022, Vol. 00, No. 00 9

Development Division (Welsh Government), Public Health Agency a member of advisory boards for AstraZeneca, Seqirus and Sanofi.
(Northern Ireland), British Heart Foundation and the Wellcome I.R. is a member of the Scientific Advisory Committee of the
Trust. This work was supported by the ADR Wales programme of Government of the Republic of Croatia and co-Editor-in-Chief of
work. The ADR Wales programme of work is aligned to the priority the Journal of Global Health. R.A.L. is a member of the Welsh
themes as identified in the Welsh Government’s national strategy: Government COVID-19 Technical Advisory Group. D.T.B. was on
Prosperity for All. ADR Wales brings together data science experts secondment to the Department of Health (Northern Ireland) and
at Swansea University Medical School, staff from the Wales Institute was a member or observer on the Scientific Advisory Group for
of Social and Economic Research, Data and Methods (WISERD) at Emergencies and several of its subgroups while conducting this
Cardiff University and specialist teams within the Welsh work.
Government to develop new evidence that supports Prosperity for

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

All by using the SAIL Databank at Swansea University, to link and References
analyse anonymized data. ADR Wales is part of the Economic and
Social Research Council (part of UK Research and Innovation) 1. Coronavirus disease (COVID-19) pandemic. World Health
funded ADR UK (grant ES/S007393/1). This work was supported Organization. https://www.who.int/emergencies/diseases/novel-
by the Wales COVID-19 Evidence Centre, funded by Health and coronavirus-2019 (28 June 2022, date last accessed).
Care Research Wales. 2. GOV.UK. COVID-19 in the UK. https://coronavirus.data.gov.
uk (28 June 2022, date last accessed).
Acknowledgements 3. Polack FP, Thomas SJ, Kitchin N et al.; C4591001 Clinical Trial
We thank Dave Kelly from Albasoft for his support with making pri- Group. Safety and efficacy of the BNT162b2 mRNA COVID-19
mary care data available and James Pickett, Wendy Inglis- vaccine. N Engl J Med 2020;383:2603–15.
Humphrey, Vicky Hammersley, Maria Georgiou and Laura 4. Voysey M, Clemens SAC, Madhi SA et al. Safety and efficacy
Gonzalez Rienda for their support with project management and ad- of the ChAdOx1 nCoV-19 vaccine (AZD1222) against SARS-
ministration. This work uses data provided by patients and collected CoV-2: an interim analysis of four randomised controlled tri-
by the NHS as part of their care and support. We would also like to als in Brazil, South Africa, and the UK. Lancet 2021;397:
acknowledge all data providers who make anonymized data avail- 99–111.
able for research. Patients and practices who are members of the 5. Falsey AR, Sobieszczyk ME, Hirsch I et al. Phase 3 safety and ef-
Oxford-Royal College of General Practitioners Research and ficacy of AZD1222 (ChAdOx1 nCoV-19) COVID-19 vaccine. N
Surveillance Centre who allow data sharing; EMIS, TPP, INPS and Engl J Med 2021;385:2348–60.
Wellbeing for facilitating pseudonymized data extraction. We wish 6. Baden LR, El Sahly HM, Essink B et al. Efficacy and safety of the
to acknowledge the collaborative partnership that enabled acquisi- mRNA-1273 SARS-CoV-2 vaccine. N Engl J Med 2021;384:
tion and access to the de-identified data in Wales, which led to this 403–16.
output. The collaboration was led by the Swansea University Health 7. Vasileiou E, Simpson CR, Shi T et al. Interim findings from first-
Data Research UK team under the direction of the Welsh dose mass COVID-19 vaccination roll-out and COVID-19 hos-
Government Technical Advisory Cell and includes the following
pital admissions in Scotland: a national prospective cohort study.
groups and organizations: the SAIL Databank, Administrative Data
Lancet 2021;397:1646–57.
Research Wales, NHS Wales Informatics Service, Public Health
8. Keehner J, Horton LE, Binkin NJ et al.; SEARCH
Wales, NHS Shared Services Partnership and the Welsh Ambulance
Alliance. Resurgence of SARS-CoV-2 infection in a highly
Service Trust. All research conducted has been completed under the
vaccinated health system workforce. N Engl J Med 2021;
permission and approval of the SAIL independent IGRP (project
385:1330–32.
number: 0911). We acknowledge the help provided by the staff of
the HBS in the Business Services Organisation Northern Ireland 9. Katikireddi SV, Cerqueira-Silva T, Vasileiou E et al. Two-dose
(BSO). The authors alone are responsible for the interpretation of ChAdOx1 nCoV-19 vaccine protection against COVID-19 hos-
the data and any views or opinions presented are solely those of the pital admissions and deaths over time: a retrospective,
author and do not necessarily represent those of the BSO. This study population-based cohort study in Scotland and Brazil. Lancet
was supported by the CO-CONNECT Patient Usher Group, the 2022;399:25–35.
EAVE II Patient Advisory Group and the Ahmadiyya Muslim 10. Feikin DR, Higdon MM, Abu-Raddad LJ et al. Duration of ef-
Women’s Association UK who have advised in developing this re- fectiveness of vaccines against SARS-CoV-2 infection and
search and the authors thank them for their support. COVID-19 disease: results of a systematic review and meta-re-
gression. Lancet 2022;399:924–44.
11. Andrews N, Tessier E, Stowe J et al. Duration of protection
Conflict of interest against mild and severe disease by covid-19 vaccines. N Engl J
A.S. is a member of the Scottish Government’s Standing Committee Med 2022;386:340–50.
on Pandemic Preparedness and the UK Government’s New and 12. Hernán MA, Robins JM. Using big data to emulate a target trial
Emerging Respiratory Virus Threats Risk Advisory Group’s Risk when a randomized trial is not available. Am J Epidemiol 2016;
Stratification Subgroup. A.S. previously served on the Scottish 183:758–64.
Government’s CMO COVID-19 Advisory Group and AstraZeneca’s 13. Rural Urban Classification. (2011) of Output Areas in England
Thrombotic Thrombocytopenic Task Force. All A.S.’s roles are and Wales, Office for National Statistics. https://geoportal.statis
unremunerated. S.de.L. has received vaccine-related research fund- tics.gov.uk/datasets/3ce248e9651f4dc094f84a4c5de18655/about
ing from AstraZeneca, GSK, Seqirus, Sanofi and Takenda; and been (7 June 2022, date last accessed).
10 International Journal of Epidemiology, 2022, Vol. 00, No. 00

14. Northern Ireland Statistics and Research Agency. Urban-Rural sectional study of multimorbidity and social care use in
Classification. https://www.nisra.gov.uk/support/geography/ur people aged over 65 years in Scotland. Age Ageing 2021;50:
ban-rural-classification (7 June 2022, date last accessed). 176–82.
15. GOV.SCOT. Scottish Government Urban Rural Classification 22. Strengthening the reporting of observational studies in epidemiol-
2016. https://www.gov.scot/publications/scottish-government- ogy (STROBE). https://www.strobe-statement.org (15 April 2022,
urban-rural-classification-2016/pages/2/ (7 June 2022, date last date last accessed).
accessed). 23. Reporting of studies using observational routinely-collected data
16. GOV.UK. English Indices of Deprivation. https://www.gov.uk/gov (RECORD). https://www.record-statement.org/ (15 April 2022,
ernment/collections/english-indices-of-deprivation (7 June 2022, date date last accessed).
last accessed). 24. Gazit S, Shlezinger R, Perez G et al. Severe acute respiratory syn-

Downloaded from https://academic.oup.com/ije/advance-article/doi/10.1093/ije/dyac199/6770060 by guest on 04 November 2022

17. Northern Ireland Statistics and Research Agency. Northern drome coronavirus 2 (SARS-CoV-2) naturally acquired immu-
Ireland multiple deprivation measure 2017. https://www.nisra. nity versus vaccine-induced immunity, reinfections versus
gov.uk/statistics/deprivation/northern-ireland-multiple-deprivation- breakthrough infections: a retrospective cohort study. Clin
measure-2017-nimdm2017 (7 June 2022, date last accessed). Infect Dis 2022;75:545–51.
18. GOV.SCOT. Scottish Index of Multiple Deprivation 2020. 25. Zheng C, Shao W, Chen X, Zhang B, Wang G, Zhang W. Real-
https://www.gov.scot/publications/scottish-index-multiple-depri world effectiveness of COVID-19 vaccines: a literature review
vation-2020/ (7 June 2022, date last accessed). and meta-analysis. Int J Infect Dis 2022;114:252–60.
19. GOV.WALES. Welsh Index of Multiple Deprivation 2019. 26. Goldberg Y, Mandel M, Bar-On YM, Bodenheimer O et al.
https://gov.wales/welsh-index-multiple-deprivation (7 June 2022, Waning immunity after the BNT162b2 Vaccine in Israel. N Engl
date last accessed). J Med 2021;385:e85.
20. Clift AK, Coupland CAC, Keogh RH et al. Living risk prediction 27. Dagan N, Barda N, Kepten E et al. BNT162b2 mRNA COVID-19
algorithm for risk of hospital admission and mortality from vaccine in a nationwide mass vaccination setting. N Engl J Med
coronavirus 19 in adults: national derivation and validation 2021;384:1412–23.
cohort study. BMJ 2020;371: m3731. 28. Bar-On YM, Goldberg Y, Mandel M, Bodenheimer O et al.
21. Henderson DAG, Atherton I, McCowan C, Mercer SW, Bailey Protection of BNT162b2 vaccine booster against COVID-19 in
N. Linkage of national health and social care data: a cross- Israel. N Engl J Med 2021;385:1393–400.