Effectiveness of BNT162b2 mRNA vaccine third doses and previous infection in protecting against SARS-CoV-2 infections during the Delta and Omicron variant waves; the UK SIREN cohort study September 2021 to February 2022

Study Design and Oversight

The SIREN study is an ongoing, multicentre, prospective cohort study involving health care workers (≥18 years of age) in the United Kingdom (ISRCTN 11041050)¹⁶. All the authors vouch for the accuracy and completeness of the data and for the fidelity of the study to the protocol.

Participant Involvement

At its core the SIREN study is a collaboration between participants and researchers. We are committed to ensuring participant feedback is heard, respected, and influences our research. The SIREN study runs a comprehensive engagement programme providing participants with regular opportunities to engage with, and feedback on, our research. The SIREN Participant Involvement Panel (PIP) is central to this, providing timely ongoing participant input into cohort recruitment and retention, study developments and communications. PIP members represent the diversity of the SIREN cohort and are supported by experts in participant engagement research.

Study Participants and Data

Participants underwent PCR testing for SARS-CoV-2 every two weeks, supplemented by widespread lateral-flow testing, and regular (monthly or quarterly) antibody testing. Data sources have been described previously.⁶ Participants were included in this analysis if they had received at least two doses of COVID-19 vaccine and their previous infection status at the point of entry to the analysis period was known. Participants were excluded if the outcome could not be determined (e.g., if they did not undergo PCR testing during the follow-up period), or if the date of onset of the primary infection, based on either a positive PCR test or COVID-19 symptoms, was not available. We included participants whose primary vaccine course was either BNT162b2 vaccine (Pfizer–BioNTech) or ChAdOx1 nCoV-19 vaccine (AstraZeneca), and whose booster vaccine (if received) was BNT162b2 vaccine (Pfizer–BioNTech). Participants with other vaccine combinations were excluded because of the small numbers involved.

Primary Outcome

The primary outcome was a PCR-confirmed SARS-CoV-2 infection, irrespective of the participant’s symptom status. This outcome was defined as a primary infection in the previously uninfected cohort or a reinfection in the previously infected cohort (two PCR-positive samples ≥90 days apart or, if the primary infection PCR is not available, a new PCR-positive sample ≥28 days after an antibody-positive result consistent with previous infection).

Participants could move from the previously uninfected cohort to the previously infected cohort 90 days after their PCR-positive test. For individual participants, the end of follow-up was the date of primary infection (in the previously uninfected cohort), the date of reinfection (in the previously infected cohort), or the date of the last PCR-negative test.

Statistical Analysis

In our Cox proportional-hazards model, the outcome was time to PCR-positive SARS-CoV-2 infection, and main predictors — vaccination status and previous infection status — were categorical and time-varying. We stratified according to age group, geographic region, and frequency of exposure to persons with Covid-19. We chose stratification based on these categorical predictors because they were statistically significant when controlled for but did not satisfy the proportional-hazards assumption (Schoenfeld test, according to predictor and global fit). We also controlled for sex, ethnic group and workplace setting (see Supplementary tables A.1 through A.6), because we observed that these predictors were statistically significant, led to a significant increase in the likelihood value (likelihood ratio test) and satisfied the proportional-hazards assumptions In Tables 1-3, we also report estimates from an unadjusted model, without stratifying or controlling for any predictor other than the time since vaccination and infection. The model accounted for calendar time, which is crucial given the varying infection rate, through the baseline hazard that can take any functional form. In this model, the hazard is assumed to be

with a time-varying baseline hazard for each stratum. We estimated the parameter β, report the hazard ratio HR=exp(β), and report vaccine effectiveness and protection from primary infection calculated as 1 minus the hazard ratio, along with Wald statistic confidence intervals. The estimates of the hazard ratios are independent of the baseline hazard, on which no assumption was made.

We conducted analyses separately for the periods of Delta-variant and Omicron-variant dominance, splitting follow-up time: Delta 07/09/2021 to 30/11/2021; Omicron 13/12/2021 to 28/02/2022. We defined three post-V3 time intervals for the Omicron period: 0 - 2 months, 2 - 4 months (upper bound included) and >4 months. For the Delta period we have a single interval approximately corresponding to the first two months after the third dose, before Omicron’s dominance. The baseline was either 8+ months after a second dose of BNT162b2, or, alternatively, 6+ months after a second dose of ChAdOx1. We first fitted the model in the previously uninfected cohort with vaccination status as a predictor with postvaccination (post-booster) intervals further categorized according to manufacturer of the vaccine administered in the primary course. The VE estimates represent the additional protection associated with a booster, relative to a waned primary course

We then performed analyses on both previously uninfected and previously infected participants, introducing previous infection status. First, we estimated the protection associated with primary infection. We grouped previous infection status into 6 categories: no primary infection, 3-6 months, 6<9, 9<12, 12-15 and >15 months after primary infection. We fitted a model where time since primary infection and time since vaccination are added as marginal predictors, without interactions. For the Delta period, we considered additional vaccine status categories (see table 2) in order to have enough follow-up time to obtain estimates. With this model we estimate protection associated with a previous infection, assuming that it is independent from vaccine status. In order to test this assumption, we fitted a model with interaction between the two predictors and verified that interaction terms were not significant. For this interaction model we split time since primary infection in only two categories, 3-12 and more than 12 months ago, to ensure convergence of hazard ratios on all combinations. We assumed that if interactions were not significant on a coarse scale, they would not appear in a more granular one. In Table 2 we only report the estimates for the previous infection categories, but not for vaccine status (see supplementary tables A.3 and A.4 for all estimates)

Next, we focussed on protection associated with third-dose vaccination and previous infection (hybrid immunity). We selected the recipients of the BNT162b2 vaccine and fitted a model with combination of the time since the primary infection (3-12 and more than 12 months ago) and the time since vaccination (table 3). The baseline here was a waned primary course of vaccination without any previous infection. The VE estimates from this model represent the combined effect associated with booster and previous infection. Recipients of the ChAdOx1 nCoV-19 vaccine for their primary course were excluded because of small numbers in the previously infected cohort.

Goodness of fit was assessed with the use of the likelihood ratio test (against the null model) and Akaike information values. The widths of the confidence intervals have not been adjusted for multiplicity and cannot be used to infer effects. We used robust variance estimates to guard against the potential for unmeasured confounders at the study site level.

We performed sensitivity analyses to assess the extent of depletion-of-susceptibles and the effect of excluding participants who enrolled in the extension for the Delta period, or that did not enrol for the Omicron period. All the sensitivity analyses provided results that were similar to those presented here, but the estimates were more uncertain (see Tables B.1 through C.6). All the analyses were conducted with the use of Stata software, version 15.1 (StataCorp). The results were independently replicated with the use of R software, version 4.1.1, survival package v.3.2-13 (R Foundation for Statistical Computing). Our annotated code is available at https://github.com/SIREN-study/SARS-CoV-2-Immunity.

Study population

We included a total of 19,614 individuals who were participating in regular SIREN testing between September 2021 and February 2022. The cohort were predominantly female (84%), of white ethnicity (89%), with a median age of 47 (IQR 16), 74% reported no comorbidities, and 69% were employed in clinical roles (Supplementary Table i).

We analysed protection over two different time periods: Delta (07 September 2021 to 30 November 2021; n= 18,429) and Omicron (13 December 2021 to 28 February 2022; n= 16,458) (Table 1). Third-dose vaccinations were introduced into our cohort on 16 September 2021 and coverage was 95% by the end of the analysis period (Supplementary Figure1). On 15 September 2021, 29% had a history of SARS-CoV- 2 infection, of which 1,901 (65%) were presumed infected in the Wild-Type period, 688 (24%) in the Alpha variant period and 8 (0.3%) in the Delta variant period (prior to the start of this analysis).

In the previously uninfected cohort, the PCR testing frequency before and after V3 was every 15.7 days and 11.2 days respectively in the Delta period and 11.0 days and 8.6 days in the Omicron period. In the previously infected cohort, in the Delta period it was every 14.2 and 11.7 days before and after V3; in the Omicron period it was 11.3 and 8.9 days.

Primary outcome

There were 2467 primary infections and 881 reinfections during the Omicron period, 278 primary infections and 85 reinfections during the Delta period. Cumulative incidence rates were higher against Omicron (primary: 43/10,000 person-days, reinfections: 32/10,000 person-days) than Delta (primary: 4/10,000 person-days, reinfections: 1/10,000 person-days).

Relative vaccine effectiveness following third dose in previously uninfected participants

In the previously uninfected cohort, third doses increased protection against both variants relative to waned dose-2 but were less effective against Omicron (Vaccine Effectiveness (VE) 0-2 months post-V3, 3-doses BNT162b2: 35% (95% Confidence Interval (CI) 21%-47%) vs. 63% (CI 40%-77%) than Delta (table 1). Heterologous third doses (BNT162b2, 2-dose ChAdOX1) were especially effective; with VE 0-2 months post-V3 over ≥68% for both variants. Dose-3 protection waned rapidly against Omicron, with no additional benefit >4-months.

Protection from previous infection

When included as a marginal predictor, previous infection continued to provide additional protection against Omicron (67% (CI 56%-75%) 3-6 months post-infection), but this waned after 9 months to around 25%, markedly lower than against Delta where it remained at 85% (CI 81% - 89%) after more than 15 months (table 2). We found no evidence of interaction between vaccine-acquired and infection-acquired protection.

Relative protection following infection and vaccination (hybrid protection)

In the previously infected cohort, with participants with two waned BNT162b2 doses and no previous infections as the reference group, we saw additional benefit associated with previous infection (table 3). Against Delta, a third dose and previous infection were associated with around 90% additional protection, even more than one year after primary infection. Against Delta, the combined additional protection remained around 70% for previous infection less than one year ago and around 50% for previous infection more than one year ago. Despite wide confidence intervals in the 4-month post-V3 category (reflecting small numbers of participants), there is an indication that the additional protective effect associated with previous infection wanes over time against Omicron, as in Table 2.

Data sharing

The metadata will be available through the Health Data Research UK Co-Connect platform. Anonymised data will be made available for secondary analysis to trusted researchers upon reasonable request.

Declaration of interests

We declare no competing interests.

Funding statement

The SIREN study is funded by UK Health Security Agency and the Department of Health, with contributions from the Northern Irish, Scottish and Welsh governments. The study receives support from the National Institute for Health Research (NIHR) as an Urgent Priority Study. SH and VJH are supported by the NIHR Health Protection Research Unit in Healthcare Associated Infections and Antimicrobial Resistance at the University of Oxford in partnership with UKHSA (NIHR200915). FI was supported by Imperial College’s President’s PhD Scholarship. The study has received grant funding from the Medical Research Council (Investigation of proven vaccine breakthrough by SARS-CoV-2 variants in established UK healthcare worker cohorts: SIREN consortium & PITCH Plus Pathway, MR/W02067X/1).

Statement of contributions

VH, FI and SF are joint first authors. VH, FI, SF, AC and SH designed the analysis plan. VH, FI and SF wrote the first draft. SF led on data management, cleaning and descriptive analysis. FI designed and undertook the statistical analysis, including sensitivity analyses, with statistical oversight provided by AC. PK independently replicated the statistical analysis in an alternative programme (R). SF, FI, PK and VH had full access to and analysed the data for this analysis. DS undertook and wrote the literature review. SF, AA, MC, JI, LP, DC, EL, CSB, are responsible for study delivery, oversight and ongoing data collection. SH is the Chief Investigator for the SIREN study and VH is the study lead, and guarantors of the study. All authors reviewed the draft manuscript and approved the final version for submission. All authors had full access to all the data in the study and accept responsibility to submit for publication.