Booster vaccination protection against SARS-CoV-2 infections in young adults during an Omicron BA.1-predominant period: a retrospective cohort study

Introduction

COVID-19 vaccines reduce SARS-CoV-2 infection and symptom severity [1], yet breakthrough infections occur, especially with the Omicron (B.1.1.529) variant [2, 3]. The effectiveness of Food and Drug Administration (FDA)-authorized or approved vaccines BNT162b2, mRNA-1273, and Ad26.COV2.S in preventing SARS-CoV-2 infection has dropped dramatically due to immune evasion and waning of vaccine-induced immunity over time [4-7]. The Omicron variant exhibits immune-system escape as the result of several mutations [8, 9]; this, and the high transmissibility of the Omicron variant, are leading to higher infection rates, strain on healthcare systems, and increased mortality [10, 11]. With the emergence of new variants combined with waning immunity, the CDC recommends a booster six months after an initial mRNA vaccine series or two months after Ad26.COV2.S vaccination to prevent symptomatic and severe outcomes of COVID-19 [7, 12]. The booster dose elicits an increase in antibody neutralization titers against the Omicron variant, and causes affinity maturation leading to a better antibody response, maintaining long-term protection against severe COVID-19 outcomes [12-15].

While boosters are understood to be effective against severe disease, hospitalization [16] and symptomatic infections [12, 17] resulting from the Omicron variant, limited information is available about their effectiveness against asymptomatic and mild symptomatic infections that may go unreported in the absence of asymptomatic surveillance testing. Asymptomatic and mild symptomatic infections play an important role in spreading SARS-CoV-2 [18]. Moreover, the clinical course of COVID-19 is understood to vary by age [19]; current estimates of booster effectiveness based on the general population may not apply to cohorts whose age distribution differs substantially from that of the general population.

Leveraging PCR testing data derived from mandatory SARS-CoV-2 surveillance program through December 14, 2021 for the study population and required departure tests before students leaving the Ithaca campus, we sought to estimate the effectiveness of COVID-19 boosters in reducing SARS-CoV-2 infections in a vaccinated population of students at Cornell University’s Ithaca campus during December 5 to December 31, 2021, when there was an explosive increase in Omicron cases to the point where the Omicron variant (BA.1) became predominant [20] and 1,926 SARS-CoV-2 infections were identified.

Methods

We used a retrospective cohort study to estimate the effectiveness of COVID-19 boosters in reducing SARS-CoV-2 infections. Specific to our context was a highly vaccinated study population, and a time period where Omicron BA.1 was the dominant viral strain. We did not have an a priori protocol or statistical analysis plan. As detailed in Supplementary Section S1, our initial analysis [21] came as part of Cornell University’s institutional response to COVID-19 during an Omicron-driven outbreak among a highly vaccinated population. This plan was made after observing aggregate infection rates that compared those who had uploaded early proof of booster vaccination (less than 17% of those in the study population who would eventually provide documentation) against those who had not. We subsequently developed a study plan that modified two of the covariates in this analysis (adding sex and changing how primary vaccine series completion date was coded) as we hypothesized that they are confounders. We then made additional modifications in response to reviewer requests. As described in Supplementary Section S1, estimates of vaccine booster effectiveness were not sensitive to these changes in the analysis.

We utilized de-identified student data from Cornell University’s COVID-19 surveillance database. We included polymerase chain reaction (PCR) positive cases identified at Cornell surveillance testing sites, the campus student health service, and a sampling site operated by the local hospital system during the period of December 5, 2021 to December 31, 2021 (referred to as the “Omicron outbreak” or the study period), in which the Omicron variant (BA.1) was the predominant strain. Samples collected at Cornell surveillance sites and the local hospital’s sampling site were tested using EZ-SARS-CoV-2 Real-Time RT-PCR [22] within a two-stage Dorfman procedure [23] using pools of size 5. The assay targets two highly-conserved regions of the SARS-CoV-2 N gene and the Omicron variant is not expected to significantly affect test performance. Samples collected at the campus student health service were tested using the Cepheid Xpert Xpress SARS-CoV-2 Assay [24]. The FDA reports that the introduction of the Omicron variant does not appear to have resulted in a significant change in the test performance of this assay [25]. This work was completed as a part of Cornell’s institutional planning and preparedness, designated as exempt from IRB review by the Cornell IRB.

Study Population

Study inclusion criteria were: being an active student at Cornell’s Ithaca campus enrolled in mandatory surveillance testing, being vaccinated with a vaccine approved by the World Health Organization (WHO) for emergency use at the time of the study (two doses of BNT162b2, two doses of mRNA-1273, one dose of Ad26.COV2.S, or two doses of other WHO-approved vaccines), and having no positive SARS-CoV-2 PCR test within 90 days before the start of the study period (Figure 1). Students were excluded from the study population if they were not subject to mandatory surveillance testing (i.e., graduate students) (n = 6,372), tested PCR-positive within 90 days before the start of the study period (n = 214), did not have at least one PCR surveillance test between November 5, 2021 and December 4, 2021 (n = 130), had not completed a WHO-approved initial vaccine series (n = 701), had an unspecified sex (n = 9), or had invalid vaccination records (n = 163). Our study included 15,800 Cornell students. Students with no PCR test records during the study period (n = 285) were considered lost to follow up. Students were required to upload a photo of a vaccine card documenting their booster status via an online form or request an exemption by February 1, 2022 [26]. According to vaccination records that were uploaded and validated by February 8, 2022, 11.9% of the study population (n = 1,876) received a booster before the study period, with the earliest booster dose administered on August 3, 2021. Our analysis does not distinguish the booster vaccine manufacturer, although this information is available in Supplementary Table S2.

• Download figure
• Open in new tab

Figure 1:

Enrollment of the study population. Students were included if they were enrolled in an full-time academic program based at Cornell’s Ithaca campus in Fall 2021, were subject to mandatory surveillance testing, had no positive SARS-CoV-2 PCR test within 90 days before the start of the study period, had at least one PCR test between November 5, 2021 and December 4, 2021, had a valid vaccination record indicating completion of WHO-approved initial vaccination series at least 14 days before the study period, and specified their sex in university records.

Results

Boosted students were more likely to be female, professional students, and have an earlier initial vaccine series than unboosted individuals (Table 1). During the study period, a total of 1,926 PCR-positive cases were identified and reported out of 15,800 students included in this analysis (overall infection risk of 12.2%). None of the cases required hospitalization. The overall infection risk among students that received a booster dose before December 5, 2021 is 6.2% (117 PCR-positives out of 1,876 boosted students), whereas the infection risk among students that did not receive a booster before December 5, 2021 is 13.0% (1,809 PCR-positives out of 13,924 students). The cumulative incidence among students who received a booster before December 5, 2021 was approximately half the cumulative incidence among students who were unboosted at that time (Figure 2). Similarly, the incidence rate was 0.6 per 100 person-days (127 PCR-positive cases/19,842 person-days) among students who had received the booster dose ≥ 7 days earlier, compared to 1.3 per 100 person-days (1,799 PCR-positive cases/135,214 person-days) among those who had not received the booster dose or had received the booster dose < 7 days earlier.

• Download figure
• Open in new tab

Figure 2:

SARS-CoV-2 infection cumulative incidence rate (number of infections per person) and its 95% confidence interval during the study period, broken out by booster dose status.

The booster effectiveness estimate from the main GEE Poisson model is 56% (95% CI: [42%, 67%]). In addition to boosters, the incidence rate in our fitted GEE Poisson model depends on the student group and the date of completing the initial vaccine series (Table 2). The incidence rate was significantly lower among students who completed the initial vaccine series after May 1, 2021 compared with those completing the initial vaccine series earlier. During the study period, the incidence rate was significantly higher among undergraduate students participating in fraternity and sorority activities (aIRR 2.16 [1.84, 2.54]) or belonging to athletic teams (aIRR 2.02 [1.57, 2.60]), and among professional students at the business school (aIRR 1.64 [1.15, 2.32]), when compared with other undergraduates. Students vaccinated with Ad26.COV2.S had a higher risk of infection relative to other vaccines, but difference was not statistically significant in our model (Table 1, Supplementary Table S2).

To study robustness to population size variation as students left campus at the end of the semester, we performed a GEE Poisson regression (Supplementary Table S3) for a shorter period (December 5, 2021 to December 18, 2021). This yields a similar booster effectiveness of 57% (95% CI: [42%, 68%], suggesting that our estimate of booster effectiveness from the full regression model is robust to the effect of students leaving campus at the end of the semester and the accordingly paused mandatory surveillance. As an additional robustness check, we performed a GEE Poisson regression (Supplementary Table S4) with multiple classes for the booster status of a person-day (unboosted, 0-6 days after booster administration, ≥ 7 days after booster dose administration). As expected, a booster status corresponding to 0-6 days after booster dose administration was not found to have a statistically significant effect on PCR-diagnosed infection (P > 0.99) relative to being unboosted.

We further performed a sensitivity analysis to assess the impact of the assumed delay between booster administration and booster effectiveness. Although 5,993 students received their booster dose during the study period (Supplementary Figure S4), 5,210 of them received it after leaving the active surveillance program at the end of the semester. These students did not contribute any person-days to the study after their last surveillance test, so they contributed only unboosted person-days. For the 783 students who received a booster dose while in active surveillance, we varied this delay from 1 day to 14 days and calculated the booster effectiveness against infection using the full model (Supplementary Figures S2 and S5). The estimated booster effectiveness is consistently above 48% and is relatively flat after 7 days suggesting that our estimate of booster effectiveness is robust to variation in the time required to mount an effective immune response after booster vaccination. Similarly, the incidence in the boosted population does not vary substantially as we change the assumed delay between booster administration and effectiveness (Supplementary Table S5).

In a separate sensitivity analysis using logistic regression (with GEE) (Supplementary Table S6), the estimated booster effectiveness is 57% (95% CI: [43%, 68%]). The estimate from the logistic regression (with GEE) is close to the estimate from the full GEE Poisson regression model.

In a subset analysis (Supplementary Section S2) on the undergraduate population living in on-campus housing, we did not observe a significant change in the estimated booster effectiveness on the subpopulation whether or not an additional covariate for housing building was included in the regression. This subset analysis justifies the use of our regression model without controlling for housing type on the full study population, given that housing information is only available for the on-campus undergraduates.

Discussion

This study is one of the first community studies to quantify booster vaccine effectiveness from an actively surveilled population of young adults. Our study provides evidence that booster vaccinations significantly reduced infections in university settings during periods when the Omicron BA.1 variant was predominant, supporting the implementation of booster vaccination requirements to minimize community transmission.

In this retrospective analysis of SARS-CoV-2 tests performed at Cornell University’s Ithaca campus during a 27-day period when Omicron BA.1 was the predominant variant, the incidence of COVID-19 infections was approximately halved among participants vaccinated with a booster dose of a WHO-approved COVID-19 vaccine when compared with vaccinated participants without a booster dose. The calculated booster effectiveness was 56% (95% CI: [42%, 67%]), which is slightly lower than a previously reported effectiveness against symptomatic COVID-like illness in adults (66%; 95% CI: [64%, 68%]) in the same time period [17]. Our estimate is similar to the reduction in cumulative incidence of reported infections associated with booster vaccination during an Omicron wave in Los Angeles County [33] and consistent with a pooled estimate of booster effectiveness (47%; 95% CI: [19%, 65%]) from a meta-analysis of studies completed during periods of Omicron predominance across the globe [34]. Importantly, our analysis includes both symptomatic and asymptomatic infections because the student population was under active surveillance, whereas none of the studies included in the meta-analysis examined a population under active surveillance [7, 35-40]. Interestingly, between two studies on the effectiveness of a second booster dose (i.e., fourth vaccine dose) [41, 42], one that used data collected during active surveillance [42] also found lower effectiveness than one that did not [41]. Students vaccinated with Ad26.COV2.S had a higher risk of infection relative to other vaccines, similar to other studies [43], though the difference was not statistically significant, likely because a small number of students had Ad26.COV2.S initial doses.

We observed lower incidence rates among students with more recent initial vaccinations, consistent with waning protection against SARS-CoV-2 infection observed within a few months of completing the initial vaccination series [44]. Months in which fewer individuals completed their initial vaccine series did not have a statistically significant effect, consistent with fewer data points resulting in less statistical power. Varsity athletes and students in fraternities/sororities at institutions of higher education may have more social contact than other undergraduates; thus, consistent with previous studies, we found that this population may be at higher risk for SARS-CoV-2 infection [31, 45, 46]. As in other higher-education institutions [46], contact tracing at Cornell identified fraternity/sorority gatherings as significant spreading events. Local case investigation efforts also pointed to additional events, including post-Thanksgiving break travel and a series of end-of-semester gatherings, as contributing to Omicron spread. With the emergence of highly transmissible variants, travel and social gatherings may put students at increased risk of SARS-CoV-2 infection. We did not find a booster status corresponding to 0-6 days after booster dose administration to have a statistically significant effect on PCR-diagnosed infection relative to being unboosted. This could be because of an insufficient immune response to the booster within 6 days of administration, consistent with previous findings [22], or because of a lack of statistical power resulting from a small number of person days in the 0-6 day window after booster administration.

Our Poisson regression analysis is similar to Bar-On et al. [47], a previous study of vaccine booster effectiveness against COVID-19. See also [41] for another similar approach to estimating vaccine booster effectiveness against COVID-19. There are two common alternative statistical approaches to estimating vaccine effectiveness. The first approach is the emulated trial design (e.g., [48]), which matches subjects with the same covariates in an observational study to emulate a randomized trial. This design offers more robustness against misspecification of the functional form by which booster status and infection risk depend on covariates, but requires a larger dataset to enable matching simultaneously on all covariates.

We did not have a sufficiently large dataset to match on all the covariates included in our model. A second alternative approach is the test-negative case-control design [17], which controls bias by including those individuals who sought testing, and improvements that model the propensity to test [49]. Since testing was mandatory and compliance was high, it is not necessary to model the propensity to test in this study population.

Limitations and Conclusion

We report on limitations of our approach that should be considered in interpreting our results. First, as we did not have an a priori protocol or statistical analysis plan (because the analysis was initially done for institutional research purposes), and because some analysis choices were made after observing the data (see Supplementary Section S1), confidence intervals and hypothesis tests would be invalidated by correlation that could have been unintentionally introduced between the data and the choice of analysis plan. Second, a limitation of the Poisson regression is information bias due to the fact that individuals were not tested every day. Delays in reporting a positive test would cause the period of the delay to have a misclassified infection status, whether this period is within one exposure period or spans multiple exposure periods. Third, there may be additional confounding bias because, even after controlling for covariates, people who avoid risk could also be more likely to get a booster vaccination, which could result in the boosted group having decreased exposure to SARS-CoV-2. This would cause our regression analyses to overestimate vaccine effectiveness. Conversely, if people engage in more risky behavior in response to perceived protection offered by booster vaccination, our analyses would underestimate vaccine effectiveness. Fourth, our data do not allow distinguishing between booster doses and additional doses for immunocompromised individuals, or distinguishing between symptomatic and asymptomatic infections for confirmed PCR-positive cases. As mentioned earlier, not all samples were tested for S-gene dropout and sequencing of every PCR-positive case was not performed. If boosters are more effective against the Delta variant than the Omicron variant, then Delta infections during the study period would lead our analysis to overestimate booster effectiveness against Omicron. Fifth, enough COVID-19 cases were observed to provide sufficient statistical power for moderately precise booster effectiveness estimates, but not to estimate how booster effectiveness varies with the manufacturer of the booster dose or the original vaccine. Most students were boosted with BNT162b2 in this study (Supplementary Table S2). Sixth, we did not include information on previous COVID-19 infections and had no information on the students’ underlying medical conditions or use of non-pharmaceutical interventions, although students were required to wear a mask (cloth, medical, or respirator) while inside campus buildings. This may lead to confounding bias if medical history differentially affected booster choice or if booster choice affected the use of non-pharmaceutical interventions. Seventh, gaps in time between a student’s last test and their departure from Ithaca (or the end of the study period, whichever came first) could lead to attrition bias. Eighth, although nearly 90% of positive samples tested for S-gene dropout during the study period exhibited this marker of the Omicron BA.1 variant [20], and sequencing confirmed the presence of the Omicron variant in all samples sequenced during this period, not all samples were tested for S-gene dropout and sequencing of each PCR-positive case was not performed. Finally, not every individual who was boosted before or during the study period may have uploaded their vaccination record, which may have led to information bias towards the null.

In conclusion, our findings are consistent with the notion that booster vaccine doses, relative to being vaccinated with the initial series, are effective in reducing infections in young adults during Omicron-predominant periods, thereby helping universities and other institutions to remain open safely. In the two months subsequent to the period covered in this study, almost the entire Cornell student body, if eligible, received at least one booster dose.

Supplementary Appendix

Section S3: Supplementary Tables and Figures

• Download figure
• Open in new tab

Supplementary Figure S1:

Distribution of the number of person-days contributed by each student in the study population (n = 15,800). Mean = 9.8 person-days; Median = 10 person-days; Interquartile range = 4 person-days.

• Download figure
• Open in new tab

Supplementary Figure S2.

Allocation of person-days to the control and booster cohorts. The black timeline shows the main analysis, with students having a “boosted” status 7 days after SARS-CoV-2 booster vaccine administration. The red timeline shows one example of our sensitivity analysis varying the time to “boosted” status, with students in this example achieving a “boosted” status 2 days after booster vaccine administration.

• Download figure
• Open in new tab

Supplementary Figure S3:

Age distribution of students in the study population (n =15,800). Age was estimated from birth year. Mean age = 20.6 years; Median age = 20 years; Interquartile range = 2 years.

• Download figure
• Open in new tab

Supplementary Figure S4:

Cumulative number of students receiving COVID-19 booster dose, over time. The shaded region represents the Omicron predominance period.

• Download figure
• Open in new tab

Supplementary Figure S5:

Mean and 95% confidence interval for the booster effectiveness against infection during the Omicron predominance period, as we vary the assumed delay for the booster dose to become effective after booster administration. Note that this analysis differs from our other sensitivity analysis considering the period shortly after booster administration: the analysis here combines all person days after the assumed delay into a single group, diluting the booster’s effect when the assumed delay is too short; the other sensitivity analysis separates people 0-6 days after booster administration from those that are 7+ days after.