key: cord-0746017-dxt37w3g authors: Goldberg, Yair; Mandel, Micha; Woodbridge, Yonatan; Fluss, Ronen; Novikov, Ilya; Yaari, Rami; Ziv, Arnona; Freedman, Laurence; Huppert, Amit title: Protection of previous SARS-CoV-2 infection is similar to that of BNT162b2 vaccine protection: A three-month nationwide experience from Israel date: 2022-03-30 journal: Am J Epidemiol DOI: 10.1093/aje/kwac060 sha: 4c7e8fbdf2ff6f44307a881f13f52a61707bb5c2 doc_id: 746017 cord_uid: dxt37w3g Worldwide shortage of vaccination against SARS-CoV-2 infection while the pandemic is still uncontrolled leads many states to the dilemma whether or not to vaccinate previously infected persons. Understanding the level of protection of previous infection compared to that of vaccination is important for policy making. We analyze an updated individual-level database of the entire population of Israel to assess the protection of both prior infection and vaccination in preventing subsequent SARS-CoV-2 infection, hospitalization with COVID-19, severe disease, and death due to COVID-19. Outcome data were collected from December 20, 2020 up to March 20, 2021. Vaccination was highly protective with overall estimated effectiveness for documented infection of 94.5% (CI: [94.3, 94.7]); hospitalization 95.8% (CI: [95.2, 96.2]); severe illness 96.3% (CI: [95.7, 96.9]); and death 96.0% (CI: [94.9, 96.9]). Similarly, the overall estimated level of protection from prior SARS-CoV-2 infection for documented infection is 94.8% (CI: [94.4, 95.1]); hospitalization 94.1% (CI: [91.9, 95.7]); and severe illness 96.4% (CI: [92.5, 98.3]). Our results should be considered by policymakers when deciding whether or not to prioritize vaccination of previously-infected adults. Recent results based on aggregated data 4, 6 and individual level data [7] [8] [9] [10] [11] have shown that the vaccine substantially reduces the number of severe COVID-19 cases. Two studies also indicate that the viral load of vaccinated individuals is significantly reduced. 12 The database included two main tables. To account for environmental risk, we calculated a municipality daily risk index by the number of cases newly confirmed in the past seven days per 10,000 residents. We used a 7day moving average since the number of PCR tests typically drops at weekends. The index was categorized into four risk levels (up to one , one to four, four to ten, and more than ten daily cases per 10,000) to yield the municipality daily risk category, and was used as a covariate in the risk model. Recovery from SARS-CoV-2 infection is not well-defined, and individuals may continue to show traces of the virus weeks and sometimes even months after the infection. 15 We defined as a recurrent infection only cases occurring three months or more after the first diagnosis. We also considered only individuals for whom the first infection was diagnosed between June 1 and September 30, 2020, omitting the beginning of the pandemic when we suspect that many infections could have been missed due to a shortage of PCR testing facilities. Hence, individuals infected before June 1, 2020 or between October 1, 2020 and December 20, 2020 were excluded from the analysis. To estimate the real world effectiveness of the Pfizer BNT162b2 vaccine in reducing Our model assumes that for a given cohort and risk profile, the hazard was constant and did not depend on the time from entering the cohort. In other words, we assumed that the protection level did not change with time after the "completion" of the vaccination protocol or after previous infection. While protection by vaccination or previous infection is expected to decrease in the long run, our assumption is reasonable given the time frame of only three months after first vaccination, and six months after recovery, where waning immunity is not expected to play a significant role. This assumption may be too much of an approximation for cohorts 1A and 1B, in which protection may increase over time, and results for these cohorts should be regarded as averages over their respective periods. Following Skowronski and De Serres, 21 we considered, as a crude approximation, a constant hazard for each of these two sub-cohorts for every risk profile. Hazard ratios between cohorts and for each adjusting covariate were estimated via a generalized linear model with a Poisson distribution and logarithmic link function, and an offset for each risk profile. 22 The formal definition of protection adopted was as follows. Consider any particular risk Infections are confirmed only by PCR tests, and since not all individuals having SARS-CoV-2 perform testing (as many have mild or no symptoms), some individuals who actually belong to the recovered cohort are misclassified and included in the other cohorts. Probably the most notable group is of individuals who are classified as unvaccinated but should have been included in the recovered cohort. Such misclassification may have a substantial effect on the estimates of vaccine effectiveness. We therefore conducted a sensitivity analysis in which we examined the effect of different misclassification rates on the results. The technical details are deferred to Web Appendix 3, with notation given in Web Tables 1 and 2 , and results in Web Tables 3 and 4 . A person is considered fully vaccinated from days 14 after receiving the second dose. However, this 14 days lag may not be sufficient for the full immunological effect resulting in an underestimate of the true vaccine effectiveness. We therefore repeated our main analysis by re-defining Cohort 2 as individuals 21 days or more after receiving the second dose. The We first investigated the dynamics of the vaccination program, disease outcomes, PCR testing, and municipality risk as a function of calendar time. Web Figures 4 and 5 present the proportion of vaccinated over time among different age and municipality risk groups, respectively. As can be seen from Web Figure 4 , the Israeli vaccination policy was initially to immunize the older population, and as time progressed, younger age groups. Web Figure 5 shows the association between environmental risk and vaccination. Web Figure 6 shows the rates over time of the different age groups among those tested, infected, hospitalized, having severe disease, and dying. Misclassification has a small effect on the crude rates, leading to a slight underestimation of vaccine effectiveness. The largest effect is observed in the 16-39 age group where vaccine effectiveness is estimated to be 93.6% if 3 out of 4 infected individuals are misclassified, and 92.8% if 1 out of 2 is misclassified; the crude effectiveness assuming no misclassification is 92.5%. Similar, but smaller, effects are observed in the other age groups. As described above, we assigned the recovered individuals to the middle PCR risk group, so that the estimated protection of a prior infection is compared to unvaccinated individuals having a single PCR cluster in the past. The protection levels afforded by a prior infection compared to unvaccinated persons who had no or 2+ past PCR tests are given in a sensitivity analysis shown in Web Table 5 . In addition, Web Table 5 presents results of a model without PCR, which can be interpreted as the overall protection of a prior infection. As expected, the protection of a prior infection compared to unvaccinated persons who did not have past PCR tests is estimated to be smaller and compared to those who had 2+ tests is larger. The results when omitting the PCR variable are very similar to the figures in There are some limitations to this observational study. One major source of confounding is related to possible population differences between individuals who recovered, were vaccinated, or neither. This confounding is partially addressed by controlling for risk factors. Specifically, for each individual we adjusted for sex, age group, number of past PCR tests and the time-dependent environmental exposure. Another major source of potential bias is related to detection of SARS-CoV-2 infection. As apparent from the PCR test counts in Table 3 , individuals who are fully vaccinated or were previously infected get tested less often than the unvaccinated cohort. Our results for the outcomes of hospitalization, severe disease, and death do not suffer from this bias and thus are more reliable. The vaccine protection against infection might be biased upward as explained above, nevertheless the remarkable curtailing of the outbreak in Israel, which followed the high vaccine uptake by the Israeli population, further suggests that the vaccine is efficient in blocking transmission, see Figure 1 . infected individuals into the Unvaccinated cohort may potentially lead to a substantial systematic bias. However, the sensitivity analysis presented in Web Appendix 3 suggests that this is not the case in our analysis, and the effect of detection bias is rather small (Web Table 4 ). This is a consequence of the Recovered cohort being much smaller than the Unvaccinated cohort, leading to a low sensitivity of misclassification on the effectiveness measure. follow-up and a larger number of events than in the previous individual-level data reports. For example, the analysis of severe cases in the randomized clinical trial is based on only 10 cases, and that of Israel's largest HMO on 229. 7 In comparison, the analysis in our study is based on 8,463 cases, with 172among them being from Cohort 2. On the other hand, the other two studies 1,7 have the respective advantages of randomization and a detailed matching process which help in bias reduction. The estimated protection against reinfection in this study is similar to that of the BNT162b2 vaccine. For documented SARS-CoV-2 reinfection, these results are similar to the results obtained in a large study from Qatar of 95% protection, 14 Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine Covid-19: Pfizer-BioNTech vaccine is rolled out in US Signals of hope: gauging the impact of a rapid national vaccination campaign COVID-19 dynamics after a national immunization program in Israel COVID-19 vaccination in Israel Impact and effectiveness of mRNA BNT162b2 vaccine against SARS-CoV-2 infections and COVID-19 cases, hospitalisations, and deaths following a nationwide vaccination campaign in Israel: an observational study using national surveillance data. The Lancet BNT162b2 mRNA Covid-19 Vaccine in a Nationwide Mass Vaccination Setting Associations of the BNT162b2 COVID-19 vaccine effectiveness with patient age and comorbidities Assessment of Effectiveness of 1 Dose of BNT162b2 Vaccine for SARS-CoV-2 Infection 13 to 24 Days After Immunization Early rate reductions of SARS-CoV-2 infection and COVID-19 in BNT162b2 vaccine recipients. The Lancet BNT162b2 mRNA Covid-19 Vaccine Effectiveness among Health Care Workers Initial report of decreased SARS-CoV-2 viral load after inoculation with the BNT162b2 vaccine Initial real world evidence for lower viral load of individuals who have been vaccinated by BNT162b2 SARS-CoV-2 antibody-positivity protects against reinfection for at least seven months with 95% efficacy 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 Antibody Status and Incidence of SARS-CoV-2 Infection in Health Care Workers SARS-CoV-2 infection rates of antibodypositive compared with antibody-negative health-care workers in England: a large, multicentre, prospective cohort study (SIREN) Clinical Spectrum of SARS CoV-2 Infection Analysis of Multivariate Survival Data Efficacy and Safety of the mRNA-1273 SARS-CoV-2 Vaccine Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine Generalized Linear Models: With Applications in Engineering and the Sciences