Effectiveness of BNT162b2 Vaccine against Critical Covid-19 in Adolescents | NEJM – nejm.org

Study Design

We used a case–control, test-negative design to assess the effectiveness of vaccination against Covid-19 resulting in hospitalization, ICU admission, or life-supporting interventions by comparing the odds of antecedent vaccination among laboratory-confirmed case patients and hospitalized controls without Covid-19.2,6,7 Evaluations of vaccine effectiveness have commonly used test-negative controls to reduce bias from health care–seeking behavior and to improve logistics.8-11 Estimates of vaccine effectiveness that are generated by the case–control or test-negative design are expressed as percentages and can be interpreted as the fraction of the specified outcome prevented in association with vaccination.7,8,12 The surveillance protocol and the statistical analysis plan (in the Supplementary Appendix, both available with the full text of this article at NEJM.org) were reviewed by the Centers for Disease Control and Prevention (CDC) and by the other participating institutions as public health surveillance; this review was conducted in accordance with applicable federal laws and CDC policy.13 CDC technical staff members served as coinvestigators and were involved in the study design, participated in the data collection and analysis and in the preparation of the manuscript, and were involved in the decision to submit the manuscript for publication.

Enrollment of Case Patients and Controls

To identify case patients and controls, we conducted active surveillance of adolescents between 12 and 18 years of age who had been admitted to 31 hospitals in 23 states in the CDC-funded Overcoming Covid-19 Network.14,15 The network was funded to evaluate vaccine effectiveness against severe Covid-19 and multisystem inflammatory syndrome in children (MIS-C) in vaccine-eligible participants. After the CDC contract had been awarded, 39 referral health centers for pediatric patients were approached on the basis of their previous experience in the enrollment of patients with Covid-19 or in conducting evaluations of vaccine effectiveness against influenza.15,16 Representatives at 31 centers agreed to participate during this period.

During the surveillance period at each study site, investigators attempted to capture all cases that met the inclusion criteria. All case patients and controls were enrolled regardless of the availability of information regarding their vaccination status. During the period from May 30 through October 25, 2021, investigators began screening for potentially eligible patients through a review of hospital admission logs and electronic medical records. For this report, the hospitalization date of the first enrolled case patient was July 1, when the percentage of fully vaccinated adolescents surpassed 20% in the United States and thus was sufficient for an evaluation of vaccine effectiveness.10,17 The onset of enrollment varied depending on local incidence and ethics approval at the site.

Case patients were selected among adolescents who were hospitalized with Covid-19 as the primary reason for admission or who had a clinical syndrome consistent with acute Covid-19 (one or more symptoms of fever, cough, shortness of breath, loss of taste, loss of smell, gastrointestinal symptoms, respiratory support, or new pulmonary findings on chest imaging). All case patients had positive results for SARS-CoV-2 on reverse transcriptase–polymerase-chain-reaction (RT-PCR) assay or on antigen testing within 10 days after symptom onset or within 72 hours after hospitalization. Results of documented positive tests before admission were accepted in 28 case patients. We excluded 23 adolescents who had received a diagnosis of MIS-C during their current hospitalization (Table S1 in the Supplementary Appendix).

Because of potential biases related to the selection of controls,18-20 we included two groups of hospitalized patients as controls: those who had negative results for SARS-CoV-2 on RT-PCR assay or antigen testing (test-negative) but who had Covid-19–like symptoms; and those without Covid-19–like symptoms who may or may not have undergone SARS-CoV-2 testing (syndrome-negative). At each site, investigators targeted a case-to-control ratio of approximately 1:1 for each of the two control groups. Eligible controls were selected from among patients in closest proximity to the ward where the case patients were hospitalized within 3 weeks after the case patient’s hospitalization date.

Data Collection

The parent or guardian of each participant was approached by trained study personnel or electronic medical records on all case patients and controls were reviewed to collect data regarding demographic characteristics, clinical information about the current illness, and SARS-CoV-2 testing history. Parents or guardians were asked about the patient’s Covid-19 vaccination history, including the number of doses and whether the most recent administration had occurred during the previous 14 days, the location where vaccination had occurred, the vaccine manufacturer, and the availability of a Covid-19 vaccination card. Study personnel searched sources, including state vaccination registries, electronic medical records, or other sources (including documentation from pediatricians), to verify reported or unknown vaccination status.

Vaccination Status

Patients were considered to have received Covid-19 vaccination based on source documentation or by plausible self-report if vaccination dates and location were provided by a parent or guardian at the time of the interview. Because the mRNA-1273 vaccine (Moderna) and Ad26.COV2.S vaccine (Johnson & Johnson–Janssen) had not been authorized for use in adolescents at the time of study initiation, patients who had received those vaccines were excluded. Patients were categorized as being unvaccinated (no receipt of the BNT162b2 vaccine before illness onset) or vaccinated if the most recent dose (first or second dose of the BNT162b2 vaccine) had been administered at least 14 days before illness onset. Adolescents who had received only one dose of vaccine or who had received a second dose less than 14 days before illness onset were considered to have been partially vaccinated; those who had received two doses at least 14 days before illness onset were considered to have been fully vaccinated. Patients who had received only one dose less than 14 days before illness onset were excluded from the analysis.2


The prespecified primary outcomes were Covid-19 resulting in hospitalization, ICU admission, the receipt of life-supporting interventions, or death. Life support was defined as the receipt of noninvasive or invasive mechanical ventilation, vasoactive infusions, or extracorporeal membrane oxygenation.

Statistical Analysis

We first conducted bivariate analyses to assess for between-group differences in characteristics on the basis of case status (case patients vs. controls) and vaccination status (fully vaccinated vs. unvaccinated). We then constructed logistic-regression models for the prespecified primary outcomes to calculate odds ratios of antecedent vaccination (fully or partially vaccinated vs. unvaccinated) in case patients as compared with controls, with associated 95% confidence intervals. A priori, we adjusted models for the U.S. Census region, calendar date of admission, age, sex, and race or ethnic group.6,10 To evaluate clustering according to hospital, we also included the hospital as a random effect in mixed-effects regression models, an analysis that did not substantially alter the results. Using a change-in-estimate approach, we assessed other potential confounding factors (the presence of underlying health conditions, specific underlying conditions, and the score on the Social Vulnerability Index) that were not included in the final models because these factors did not change the odds ratio for vaccination by more than 5%.6,21

We calculated vaccine effectiveness against the primary outcomes by comparing the odds of full vaccination against Covid-19 among case patients and controls using the equation for vaccine effectiveness of (1– adjusted odds ratio)×100, as determined from logistic-regression models. We used Firth logistic regression (a penalized likelihood–based method) for models with fewer than five vaccinated case patients.22 Preplanned subgroup analyses included effectiveness against Covid-19 hospitalization according to age group (12 to 15 years vs. 16 to 18 years) and protection of partial vaccination with the BNT162b2 vaccine against Covid-19 hospitalization. We computed effectiveness separately with each control group and overall with the two control groups combined. The widths of the confidence intervals have not been adjusted for multiplicity, so the intervals should not be used to infer vaccine effectiveness for the subgroup analyses. All statistical analyses were performed with the use of SAS software, version 9.4 (SAS Institute).