Principal findings
This study investigated the relative vaccine effectiveness of a fourth dose of the BioNTech-Pfizer mRNA BNT162b2 vaccine compared with receiving three doses only, against both infection with the SARS-CoV-2 omicron variant and infection resulting in severe covid-19, assessed by hospital admissions and death, among individuals aged 60 years and older in Israel. We found that, relative to three vaccine doses, a fourth dose initially provided additional protection against both SARS-CoV-2 infection and severe disease (assessed by hospital admissions and deaths). However, relative vaccine effectiveness against infection quickly decreased over time, peaking during the third week after inoculation at 65.1% (95% confidence interval 63.0% to 67.1%) and declining to 22.0% (4.9% to 36.1%) by the end of the 10 week follow-up period. Similar results were obtained with different analytical approaches.
Comparison with other studies
The waning of vaccine effectiveness against SARS-CoV-2 infection is consistent with previous observations of the second and third doses of the BNT162b2b vaccine.1011 Nonetheless, compared with the previously demonstrated waning pattern of the relative vaccine effectiveness of three doses versus two doses in real world settings (which begins around three months after inoculation), the relative effectiveness of the fourth dose against infection appears to wane sooner, just as waning after the third dose was sooner than after the second dose.1126 This more rapid decline could be explained by a reduced effectiveness of the BNT162b2b vaccine against the omicron variant.2728 However, the effect of waning protection of sequential vaccine doses is difficult to differentiate from the real world circulation of variants.
Additionally, we need to consider that frequent stimulation with the same mRNA vaccine triggers an immunological response that is yet to be fully understood in terms of duration, effectiveness, and patterns of waning when exposed to different variants. In this respect, the interval between doses could have an impact on the duration of immunity. In Israel, the intervals between doses have been short compared with other countries. The second dose of BNT162b2b was issued between three and four weeks after the first, and the third dose was initially given to those who received the second dose at least five months before; however, a few months later, authorities shortened the required interval to three months in the light of a new increase of SARS-CoV-2 infections.29 The fourth dose, as mentioned previously, required a minimal interval of four months. Nonetheless, a growing number of studies suggest that an extended dosing interval amplifies both the humoral immune response30 and perhaps the cellular response,31 and is an effective global strategy given the discrepancies between regions with high and low vaccination coverage.32 But even recent studies have so far only investigated the second dose and have not evaluated responses to the omicron variant.3033 Our present analysis showed that receiving the third dose later (that is, after August 2021) was associated with slight protection against infection, but was not significantly associated with the odds of hospital admission or death related to covid-19. Although our findings point to a protective effect of a longer interval between the second and third doses, owing to Israeli regulations, the distribution in the population is narrow overall, and the number of people with longer intervals is small; thus, these questions should be further researched in other countries.
Unlike the relative vaccine effectiveness against SARS-CoV-2 infection, the relative effectiveness of a fourth dose against severe covid-19 was maintained at high level (>72%) throughout the 10 week follow-up period. Sustained vaccine effectiveness against severe disease has been shown for previous doses also.2034 However, severe disease was a relatively rare event, occurring in <1% of participants receiving fourth doses or third doses only. The difference between waning of protection against infection and sustained protection against severe disease could imply a different underlying immunological mechanism. A recent clinical study by Terreri et al33 suggested that breakthrough infections could be explained by a decrease in the concentration of specific antibodies, which are probably not generated by a parenteral vaccine and are slow to reach nasopharyngeal mucosal sites of viral entry. By contrast, immunological memory (memory B and T cells) does not wane, and might be important for protection against severe disease.33 However, this hypothesis does not explain why a third vaccine dose protects against severe disease compared with a second one,6 and why a fourth improves the protection of a third dose. A stabilisation effect of additional doses on immune memory should be further investigated, as well as studies on mucosal vaccines.35
Strengths and limitations of this study
Our analysis had several limitations. Firstly, to provide timely evidence of the relative vaccine effectiveness of a fourth dose of BNT162b2b vaccine, we were only able to include 10 weeks of data. Although the pattern of a short term increase in protection against SARS-CoV-2 infection followed by waning is already present, long term vaccine effectiveness needs to be evaluated and is particularly important for estimates of relative effectiveness of the fourth dose against severe covid-19. Protection from previous doses against severe disease has been shown to wane more slowly than protection against infection.2034 Nonetheless, our study suggests a more rapid waning of protection against infection from a fourth dose than from previous doses; therefore, waning of relative vaccine effectiveness against hospital admission and death related to covid-19 needs to be further examined over a longer period.
A second limitation stems from the varying dominance of different SARS-CoV-2 variants over time. The period after fourth dose vaccination in Israel has been dominated by the omicron variant, which makes assessing the relative effectiveness of the fourth dose against other covid-19 variants difficult—a well recognised limitation of real world analyses during this pandemic.63637 Furthermore, because the eligible population for a fourth dose comprised individuals aged 60 years or older, we cannot infer similar relative vaccine effectiveness and potential waning in younger people. Additionally, fourth dose recipients were more unwell overall, possibly stemming from targeted vaccination campaigns and previous rollout policies. Adjusting for comorbidities by virtue of an available comprehensive medical history, as well as adjustment and matching by other factors including timing of the third vaccine dose, residential and social factors, and previous testing renders residual confounding less likely.
The lack of pre-defined PCR testing protocols implemented in the study population also presented a challenge. This limitation has been discussed extensively in previous covid-19 observational studies and could lead to potential biases relating to healthcare seeking behaviour.6113839 The test negative design attempts to mitigate this potential bias. Firstly, those participants not tested are not eligible to be considered uninfected controls, thus reducing potential misclassification of SARS-CoV-2 infection status (which is more likely to occur in a cohort study), especially when the omicron variant was spreading rapidly and patients with no or mild symptoms might not be tested. Secondly, previous studies37 have shown that improved healthcare seeking behaviour might be related both to improved vaccine uptake as well as behaviours that could influence the risk of SARS-CoV-2 infection (eg, mask wearing, social distancing, and handwashing) or to more severe outcomes if infected (such as chronic disease management). Thus, healthcare seeking behaviour could confound the association between timely receipt of a fourth vaccine dose and SARS-CoV-2 related outcomes.
These potential confounders are illustrated by a directed acyclic graph (fig 3), following the conceptual scheme presented by Sullivan et al.40 When restricting the analysis only to those individuals with measurable and reasonable healthcare seeking behaviour (fig 3C)—in a similar concept yet different execution to the adjustment of other confounders (shown in fig 3B)—this potential confounding does not bias the relation between exposure and outcome. Because healthcare seeking behaviour is not a binary variable, this solution is insufficient when applied to real world data analysis. Nevertheless, a more comprehensive discussion should resolve any potential collider bias. The criterion for minimal eligibility in a test negative design is the fact that a patient was tested, so this selection bias could create a scenario where we condition on the collider (testing is a common effect of both healthcare seeking behaviour and potentially of the severity of the SARS-CoV-2 manifestation, which prompts the patient to be tested, as seen in fig 3C).41 However, controlling for both healthcare seeking behaviour and the propensity to be tested (inherent to the design) blocks the biasing path, albeit incompletely, under the limitation of measuring healthcare seeking behaviour adequately (fig 3E).42
Directed acyclic graph illustrating biases and their attempted mitigation in this test negative design study. S=sex; A=older age groups; CM=comorbidities; AL=assisted living or nursing home; SES=socioeconomic status; Ct=city of residence; TW=week of testing; Tb=time passed since third dose or first booster. (A) Known confounders by a priori knowledge of previous studies possibly confounding the association between time from the fourth dose (4V) and SARS-CoV-2 related outcomes (S-C-2). (B) For simplicity, all confounders in panel A were combined as C; healthcare seeking behaviour (HsB) possibly confounds the association between a fourth dose (4V) and S-C-2; additionally, some confounders in panel A could be causes of HsB, such as age or comorbidities. (C) In this test negative design, only study participants with a measurable and reasonable healthcare seeking behaviour (that is, HsB=1) were included, thus removing this biasing path. (D) Healthcare seeking behaviour potentially influences the propensity to be tested (PtbT); hence, when a test negative design includes only patients tested for SARS-CoV-2 (or conditioning on the possible collider, PtbT), this inclusion could create a collider bias. (E) When healthcare seeking behaviour is controlled, as attempted by the test negative design, the path is blocked
Healthcare in Israel is free and universal to all residents, as are PCR tests for the study population. Therefore, implementing this study design in Israel increases the likelihood of engagement in two groups of people who have already chosen to be at least thrice vaccinated. But alongside its advantages, generalisability of the test negative design is limited, because of the association in other populations who did not choose to be tested.42 Nonetheless, supporting this design were previous covid-19 studies implementing a test negative design that yielded comparable results to cohort studies on this population,611 while presenting diminished bias in the short term outcomes of vaccinee effectiveness.
Some studies have included rapid antigen tests in their analysis, treating them equally to PCR tests, whereas this study did not. Although our study had fewer observations by excluding rapid antigen tests, such tests are generally considered less reliable, and negative at-home tests are not reported. Furthermore, a policy was in place to use PCR tests for the entire examined age group (≥60 years, eligible for a fourth dose) during the follow-up period, making testing accessible to the study population.
Lastly, the relative metric of our main analysis warrants discussion. We compared the effectiveness of the fourth dose to that of a third one, thereby estimating the relative vaccine effectiveness rather than the absolute vaccine effectiveness comparing fourth dose recipients with unvaccinated individuals.21 Admittedly, the nature of relative vaccine effectiveness requires contextuality in its interpretation.21 Nonetheless, two overarching principles guided our choice. Firstly, owing to the rapid rollout and high compliance rates, most of the SARS-CoV-2 naive population older than 60 years had received at least two vaccine doses.1 Therefore, apart from potential scarcity of data in comparing to unvaccinated individuals, this eligible yet unvaccinated population in a massively campaigned environment such as Israel is plausibly different in terms health related behaviour, which could introduce a bias in rendering it the reference group,43 as has been pointed out in previous vaccination studies of Israeli populations measuring the effectiveness of three doses relative to two doses.61144 Secondly, the recent global spread of the omicron variant obliges policy makers worldwide to issue acute recommendations, equipped with limited information, as has been the case throughout the pandemic. The immediate question facing many countries now is whether, in the light of the rapid spread and recent studies pointing to waning of the third vaccine dose,101145 a second booster should be recommended. This question is specific to those individuals who are eligible for the fourth dose—that is, those who have already received three vaccine doses. Therefore, the relative nature between the fourth dose and three doses is inherent to this healthcare policy question. Focusing efforts on additional doses, of course, has important implications on resource allocation.
