For this analysis, we included persons who, on January 1, 2022, were 60 years of age or older and had received three doses of BNT162b2 at least 4 months before the end of the study period (March 2). We excluded the following persons from the analysis: those who had died before the beginning of the study period (January 10); those for whom no information regarding their age or sex was available; those who had had a confirmed SARS-CoV-2 infection before the beginning of the study, determined with the use of either a polymerase-chain-reaction (PCR) assay or a state-regulated rapid antigen test; those who had received a third dose before its approval for all older residents (i.e., before July 30, 2021); those who had been abroad for the entire study period (January 10 to March 2; persons were considered to be abroad 10 days before traveling until 10 days after their return to Israel); and those who had received a vaccine dose of a type other than BNT162b2.
For persons who met the inclusion criteria, we extracted information on March 4, 2022, regarding SARS-CoV-2 infection (confirmed either by state-regulated rapid antigen test or by PCR) and severe Covid-19 (defined with the use of the National Institutes of Health definition2 as a resting respiratory rate of >30 breaths per minute, an oxygen saturation of <94% while breathing ambient air, or a ratio of partial pressure of arterial oxygen to fraction of inspired oxygen of <300) during the 14 days after confirmation of infection. During the study period, infections were overwhelmingly dominated by the omicron variant.3 We also extracted data regarding vaccination (dates and brands of first, second, third, and fourth doses) and demographic variables such as age, sex, and demographic group (general Jewish, Arab, or ultra-Orthodox Jewish), as determined by the person’s statistical area of residence (similar to a census block4).
The study period started on January 10, 2022, and ended on March 2, 2022, for confirmed infection and ended on February 18, 2022, for severe illness. The starting date was set to 7 days after the start of the vaccination campaign (January 3, 2022) so that at least the first four-dose group (days 8 to 14 after vaccination) would be represented throughout the study period (Fig. S1 in the Supplementary Appendix, available with the full text of this article at NEJM.org). The end dates were chosen to minimize the effects of missing outcome data due to delays in reporting PCR or antigen test results and to allow time for the development of severe illness.
The design of the study was similar to that of a previous study in which we assessed the protection conferred by the third vaccine dose as compared with the second dose.5 We calculated the total number of person-days at risk and the incidence of confirmed infection and of severe Covid-19 during the study period defined for each outcome. For persons who received the fourth dose, treatment groups were defined according to the number of weeks that had passed since receiving that dose, starting from the second week (8 to 14 days after vaccination). These four-dose groups were compared with two control groups. The first control group included persons who were eligible for a fourth dose but had not yet received it (three-dose group). Because persons who received the fourth dose might have differed from those who had not according to unmeasured confounding variables, a second control group was defined as persons who had received a fourth dose 3 to 7 days earlier (internal control group). This control group included the same persons as the four-dose groups, but during a period in which the fourth dose was not expected to affect the rate of confirmed infection or severe illness. The membership in these groups was dynamic, and participants contributed risk days to different study groups on different calendar days, depending on their vaccination status.
The study was approved by the institutional review board of the Sheba Medical Center. All the authors contributed to the conceptualization of the study, critically reviewed the results, approved the final version of the manuscript, and made the decision to submit the manuscript for publication. The authors vouch for the accuracy and completeness of the data in this report. The Israeli Ministry of Health and Pfizer have a data-sharing agreement, but only the final results of this study were shared.
Using quasi-Poisson regression, we estimated the rates of confirmed infection and severe Covid-19 per 100,000 person-days for each study group (included as factors in the model), with adjustment for the following demographic variables: age group (60 to 69 years, 70 to 79 years, or ≥80 years), sex, and demographic group (general Jewish, Arab, or ultra-Orthodox Jewish). Because incidences of both confirmed infection and severe illness increased rapidly during January 2022, the risk of exposure at the beginning of the study period was lower than at the end of the study period. Moreover, the fraction of the population in each study group changed throughout the study period (Fig. S1). Therefore, we included calendar date as an additional covariate to account for changing exposure risk.6 The end of the study period for severe Covid-19 was set to 14 days before the date of data retrieval (March 4), allowing at least 14 days of follow-up time for the development of severe illness. To ensure the same follow-up time for severe Covid-19 in all persons, we considered only cases of severe illness that developed within 14 days after confirmation of infection. The date used for counting events of severe Covid-19 was defined as the date of the test confirming the infection that subsequently led to the severe illness.
Persons who received four doses were assigned to groups according to the numbers of weeks that had passed since receipt of the fourth dose; for each outcome, we estimated the incidence rate in each of these four-dose groups and in the two control groups. We calculated two rate ratios for each treatment group and each outcome: first, the ratio of the rate in the three-dose group to that in each four-dose group; and second, the ratio of the rate in the internal control group to that in each four-dose group. Note that the higher this rate ratio is, the greater the protection conferred by the fourth dose of vaccine. In addition, adjusted rate differences per 100,000 person-days during the study period were estimated with a method similar to that used in our previous analysis.7 Confidence intervals were calculated by exponentiating the 95% confidence intervals for the regression coefficients, without adjustment for multiplicity. Thus, the confidence intervals should not be used to infer differences between study groups.
To check for possible biases, we performed several sensitivity analyses. First, we estimated the rate ratios for confirmed infection using an alternative statistical method that relied on matching (similar to that used by Dagan et al.8), as described in detail in the Supplementary Appendix; this approach could not be applied to the analysis of severe Covid-19 because of the small case numbers. Second, we examined the results of using data on infections confirmed only by PCR testing and excluding data on those confirmed by state-regulated antigen testing. Third, we repeated the analyses with data from the general Jewish population only. Fourth, we analyzed the data while accounting for the exposure risk over time in each person’s area of residence. Fifth, we analyzed the data while accounting for the time of vaccination since the third dose. Further details of the sensitivity analyses are provided in the Supplementary Appendix.