key: cord-0954038-r025yr9i
authors: Xie, J.; feng, s.; Li, X.; Gea Mallorqui, E.; Prats-Uribe, A.; PRIETO-ALHAMBRA, D.
title: Comparative effectiveness of the BNT162b2 vs ChAdOx1 vaccine against Covid-19
date: 2021-12-21
journal: nan
DOI: 10.1101/2021.12.18.21268039
sha: fb3e56ad509afae81f78d0660698defd8781d49e
doc_id: 954038
cord_uid: r025yr9i

Although pivotal trials with varying populations and study methods suggest higher efficacy for mRNA than adenoviral Covid-19 vaccines, no direct evidence is available. Here, we conducted a head-to-head comparison of BNT162b2 versus ChAdOx1 against Covid-19. We analysed 235,181 UK Biobank participants aged 50 years or older and vaccinated with one or two doses of BNT162b2 or ChAdOx1. People were followed from the vaccination date until 18/10/2021. Inverse probability weighting was used to minimise confounding and the Cox models to derive hazard ratio. We found that, compared with two doses of ChAdOx1, vaccination with BNT162b2 was associated with 30% lower risks of both SARS-CoV-2 infection and related hospitalisation during the period dominated by the delta variant. Also, this comparative effectiveness was consistent across several subgroups and persisted for at least six months, suggesting no differential waning between the two vaccines. Our findings can inform evidence-based Covid-19 vaccination campaigns and booster strategies.

people receiving the second dose during the two-dose enrolment period. Vaccine uptake over 50 calendar time in our study population is depicted in Figure 1 . 51

In the one-dose vaccine cohorts, people receiving BNT162b2 were slightly older (mean (sd) age: 53 71.35 (7.21) years) than those receiving ChAdOx1 (mean (sd) age: 71.06 (6.02) years ). Sex (44.5% vs 54 44.1% male) and ethnicity (91.2% vs 92.6% White) were comparable between the two groups. Little 55 difference was seen in the prevalence of medicines or comorbidities. The main differences between 56 cohorts were vaccination dates and socio-economic factors such as income (Figure 2 and 57

Supplementary Table 2 ). Similar patterns of baseline characteristic differences were also seen in the 58 two-dose comparison cohorts (Figure 2 and Supplementary 

Kaplan-Meier curves stratified by vaccine depicted similar trends between Covid-19 infection and 79

hospitalisation. The cumulative incidence after the first dose increased rapidly in the early follow-up 80 but flattened later until 14 weeks after vaccination (Figure 3) , corresponding to the calendar period 81 from January to March 2021 (Supplementary Figure 1) . Conversely, the trend of cumulative 82 incidence was reversed for the two-dose cohorts, with a substantial increase starting 12 weeks after 83 the second dose (Figure 3 ), corresponding to the calendar period from June to October 2021 84 (Supplementary Figure 1) . Changes in community transmission over time in the general population 85

of England and among UK Biobank participants are shown in Supplementary Figure 1 Notably, incidence rates of Covid-19 infection increased substantially during the study period, from  91  3.93 and 4.87 per 1,000 person-years at the 0-4-week window to 74.74 and 111.87 per 1,000  92 person-years at the >24-week window in BNT162b2 and ChAdOx1 cohorts, respectively. However, 93

as reflected by HRs, the comparative risks were stable for at least six months (Figure 4) The main results were consistent in the propensity score-matched cohorts with slightly wider 103 confidence intervals (Supplementary Table 4 ). The rectangle background with grey color depicts the vaccination anchor windows. Only people who received study vaccines within the anchor windows 108

were included for the comparative analysis. We defined the 2 rd to 8 th (Jan 11, 2021 to Feb 28, 2021) and 12 th to 18 th (March 22, 2021 to May 9, 2021) 109 calendar weeks as the two anchor windows for the one-dose and two-dose cohorts, respectively. The decision-making for these anchor windows was based 110 on (1) there were both BNT162b2 and ChAdOx1 vaccines delivered in each epidemiological week of the window, (2) numbers of the two vaccines were 111 generally comparable, and (3) UK's policy on the gap between the first and second dose was 10 to 12 weeks for both vaccines. Notes: The different X-axis range of incidence rate for the Covid-19 infection and hospitalisation outcomes (unit, per 1000 person-years). Hazard ratios for 133

Covid-19 hospitalisation were truncated at 0.2 (lowest) and 2 (highest likely to be exposed to SARS-CoV-2 than our cohort: a community-based "healthy volunteer" 157 population, 15 which was corroborated by the higher infection rate in their study (58 per 1,000 158 person-years) compared to ours (~13 per 1,000 person-years) during the first few weeks following 159 the first dose. Second, both our and Hulme's studies found that, on average, vaccination with 160 BNT162B2 happened earlier than ChAdOx1. Given the variation over time in community 161 transmission in England (Supplementary Figure 1) , insufficient control for vaccination date can result 162

in an inflated risk of COVID-19 amongst BNT162B2 recipients. Informed by this, we observed a 163 substantial change of hazard ratio after the complete alignment of the first vaccination date by 164 weighting in our study. Thirdly, Hulme's risk assessment period started from the first dose (January 165

to February 2021) and ended on 13 June 2021, reflecting a mixed effect of one and two doses of 166 vaccines. 167

Finally, our data for the first time demonstrated that the observed differences did not attenuate 168 over time, verifying the hypothesis from a few recent preprint studies. 16,17 169

The leading challenges in estimating vaccine effectiveness with observational data lies in 171 confounding by indication and potentially differential testing rates between exposed vaccinated and 172 unvaccinated populations. 18,19 However, our study minimized the impact of such differences by 173 comparing two vaccines and restricting the analysis to a period when both vaccines were available 174 and had a similar national delivery. UK data are ideal for comparative effectiveness research into 175

Covid-19 vaccines, as both BNT162b2 and ChAdOx1 were rolled out simultaneously for the target 176 population (adults ≥ 50 years) included in our analyses. 20 21,22 . 177 However, additional limitations remain. Information on a few participants' characteristics was 178 collected ten years ago and may have changed since then. However, given that all people at the 179 cohort recruitment were already middle-aged or older adults (40-69 years old), we expected any 180 changes in those features such as socio-economic status and education are likely to be minor or 181 unrelated to the choice of vaccine types. Admittedly, misclassification of covariates could bias the 182 share, reuse, remix, or adapt this material for any purpose without crediting the original authors. placed this preprint (which was not certified by peer review) in the Public Domain. It is no longer restricted by copyright. Anyone can legally share, reuse, remix, or adapt this material for any purpose without crediting the original authors. placed this preprint (which was not certified by peer review) in the Public Domain. It is no longer restricted by copyright. Anyone can legally

The copyright holder has this version posted December 21, 2021. ; https://doi.org/10.1101/2021.12.18.21268039 doi: medRxiv preprint

week of receipt of the Covid-19 vaccine was also included as it affected the probability of receiving 245 different vaccines and infection risk (through changes in community transmission level). 246

The outcome risk assessment window for the one-dose cohort went from receiving the first dose to 248 the earliest of outcome occurrence, receiving the second dose, or 14 weeks after the vaccination. 249

For the two-dose cohort, follow-up was from receiving the second dose to outcome occurrence or 250 end of the study (18/10/2021). 251

We used the propensity score-based inverse probability of treatment weighting (IPW) to minimise 252

confounding. 25 The specification of propensity score modelling was described in Supplementary  253 Methods. We generated Kaplan-Meier plots to depict the cumulative incidence of outcome over 254 time in each cohort. We applied Cox proportional hazards regression with robust variance 255 estimators to derive average hazard ratio (HR) and calculated incidence rates using weighted counts 256 and follow-up time. We assessed the proportionality of hazards in the Cox models by visually 257

inspecting scaled Schoenfeld residuals. 258

To evaluate for potential heterogeneity of the comparative effectiveness among specific 259 demographic subgroups and overtime after the second dose, we performed several secondary 260 analyses by including multiplicative interaction terms between the vaccine types and the following 261 categories separately: age (50-75 years or > 75 years), sex (male or female), ethnicity (white or other 262 ethnic groups), BMI (< 25 vs ≥ 25), and four weeks' consecutive time intervals. 263

We conducted the negative control experiment to assess potential residual (unobserved) 264

confounding. Three clinical outcomes (limb pain, fracture, and peptic ulcer events) were pre-265 specified and should not be associated with vaccination status (BNT162b2 vs ChAdOx1) if potential 266 confounders have been adequately controlled for. 267

Finally, we performed a sensitivity analysis using propensity score 1:1 matching without 268

replacement. Specifically, we set a caliper of width equal to 0.2 of the standard deviation of the logit 269 of the propensity score. 26 We used R version 3.5.1 for all analyses. 270 share, reuse, remix, or adapt this material for any purpose without crediting the original authors. placed this preprint (which was not certified by peer review) in the Public Domain. It is no longer restricted by copyright. Anyone can legally

The copyright holder has this version posted December 21, 2021. ; https://doi.org/10.1101/2021.12.18.21268039 doi: medRxiv preprint

Supplementary Figure 1 

Notes: The data for weekly Covid-19 cases and SARS-CoV-2 variants of concern in England were from the Public Health England 273

(https://coronavirus.data.gov.uk/). 274 

The construction of the "origin" field is based on information provided on the specimen request form. If the specimen was marked as being from an acute 293 (emergency) care provider, an A&E department, an inpatient location, or resulted from health care associated infection, it is recorded by PHE as an 294 inpatient sample. Tests marked as being from "Healthcare Worker Testing" are never recorded as inpatient samples, though some may also carry an acute 295 flag. 296

The aim of designating inpatient status for the SARS- We calculated propensity scores for a vaccination with BNT162b2 against ChAdOx1 using logistic regression. The variables included in the model were 322

previous Covid-19 infection status (binary), age on the vaccination date (continuous linear), sex (binary), ethnicity (categorial, collected at the UK Biobank 323 recruitment), multiple socio-economic deprivation scores (Townsend deprivation index, income score, employment score, health score, education score, 324

housing score, and crime score; continuous linear, collected at the UK Biobank recruitment), education levels (categorial, collected at the UK Biobank 325 recruitment), body mass index (continuous linear, collected at the UK Biobank recruitment), a list of pre-specified medications (binary, obtained from 326 primary care prescription records), the number of hospital admissions (continuous linear, obtained from HES), and comorbidities (binary, obtained from 327 primary care diagnosis records). The weights for each participant were then computed based on the Rosenbaum formula:

= +

(1− )

(1− )

. 27 To reduce 328 unstable estimate due to extreme weights in the tails of the propensity score distribution, we used asymmetric trimming to exclude people whose 329

propensity score was below the 1 th percentile of the propensity score of the BNT162b2cohort and above the 99 th percentile of the propensity score of the share, reuse, remix, or adapt this material for any purpose without crediting the original authors. placed this preprint (which was not certified by peer review) in the Public Domain. It is no longer restricted by copyright. Anyone can legally

The copyright holder has this version posted December 21, 2021. ; https://doi.org/10.1101/2021.12.18.21268039 doi: medRxiv preprint

Safety and Efficacy of the BNT162b2 mRNA Covid-19

Efficacy and Safety of the mRNA-1273 SARS-CoV-2 355

Safety and efficacy of the ChAdOx1 nCoV-19 vaccine 357 (AZD1222) against SARS-CoV-2: an interim analysis of four randomised controlled trials in Brazil

South Africa, and the UK. The Lancet

Safety and Efficacy of Single-Dose Ad26.COV2.S Vaccine 361 against Covid-19

BNT162b2 mRNA Covid-19 Vaccine in a Nationwide Mass 363 Vaccination Setting

Interim findings from first-dose mass COVID-19 vaccination 365 roll-out and COVID-19 hospital admissions in Scotland: a national prospective cohort study. The 366 Lancet

Demographic risk factors for COVID-19 infection, severity, 368 ICU admission and death: a meta-analysis of 59 studies

Male sex identified by global COVID-19 meta-371 analysis as a risk factor for death and ITU admission

Ethnic differences in SARS-CoV-2 infection and COVID-374 19-related hospitalisation, intensive care unit admission, and death in 17 million adults in 375 England: an observational cohort study using the OpenSAFELY platform. The Lancet

Hospitalization, and Death: A Cohort Study of 2 524 926 Catalans

Safety and immunogenicity of heterologous versus 381 homologous prime-boost schedules with an adenoviral vectored and mRNA COVID-19 vaccine 382 (Com-COV): a single-blind, randomised, non-inferiority trial

Heterologous vaccine regimens against COVID-19. The 385 Lancet

Immunogenicity and reactogenicity of 387

BNT162b2 booster in ChAdOx1-S-primed participants (CombiVacS): a multicentre, open-label, 388 randomised, controlled, phase 2 trial. The Lancet

Comparative effectiveness of ChAdOx1 versus 391

BNT162b2 COVID-19 vaccines in Health and Social Care workers in England: a cohort study 392 using OpenSAFELY

Comparison of Sociodemographic and Health-Related 394 Characteristics of UK Biobank Participants With Those of the General Population

SARS-CoV-2 variants: levels of neutralisation required for 397 protective immunity

Effectiveness of COVID-19 vaccines against the B.1.617.2 399 variant

Theoretical Framework for Retrospective Studies of the 401 Effectiveness of SARS-CoV-2 Vaccines

Methodological challenges in measuring vaccine 404 effectiveness using population cohorts in low resource settings

UK COVID-19 vaccines delivery plan

409 21 Regulatory approval of Pfizer/BioNTech vaccine for COVID-19

Regulatory approval of Vaxzevria (previously COVID-19 Vaccine AstraZeneca). GOV.UK. Updated 413

The UK Biobank resource with deep phenotyping and 416 genomic data

Dynamic linkage of COVID-19 test results between Public 418

Health England's Second Generation Surveillance System and UK

An Introduction to Propensity Score Methods for Reducing the Effects of 421 Confounding in Observational Studies