The study start date was the earliest Emergency Use Authorization date for the BNT162b2 vaccination for each age group: for ages 5 to 11 years, October 29, 2021; ages 12 to 15 years, May 10, 2021; and ages 16 to 17 years, December 11, 2020. For the data cut dates, CVS Health data were available through May 31, 2022; HealthCore data through May 6, 2022; and Optum data through June 25, 2022.
eTable 1. Database Descriptions
eTable 2. Codes for COVID-19 Vaccine Administrations, Utilized in Claims and IIS data
eTable 3. Outcomes, Age Groups, Settings, Clean Windows, Risk Windows, and Analysis Type for the Pediatric Population (Ages 5-17)
eTable 4. Sequential Testing Results in Health Plan Members Aged 5-17 years by Outcome Following BNT162b2 All Doses (Primary Series, Dose 1, Dose 2, and Third/Booster Dose) in CVS Health, HealthCore, and Optum Databases
eTable 5. Descriptive Outcome Counts, Overall and by Data Partners
Data sharing statement
Customize your JAMA Network experience by selecting one or more topics from the list below.
Hu M, Wong HL, Feng Y, et al. Safety of the BNT162b2 COVID-19 Vaccine in Children Aged 5 to 17 Years. JAMA Pediatr. 2023;177(7):710–717. doi:10.1001/jamapediatrics.2023.1440
Does active monitoring detect potentially elevated risk of health outcomes after BNT162b2 COVID-19 vaccination in the US pediatric population aged 5 to 17 years?
In this cohort study of more than 3 million children (aged 5-17 years) who received BNT162b2 COVID-19 vaccination through mid-2022 using data from 3 US commercial claims databases, only myocarditis or pericarditis met the statistical threshold for a signal after BNT162b2 COVID-19 vaccination via near–real-time monitoring.
Results from near–real-time monitoring of health outcomes after BNT162b2 COVID-19 vaccination are consistent with current evidence and provide additional evidence of vaccine safety in the pediatric population.
Active monitoring of health outcomes after COVID-19 vaccination offers early detection of rare outcomes that may not be identified in prelicensure trials.
To conduct near–real-time monitoring of health outcomes following BNT162b2 COVID-19 vaccination in the US pediatric population aged 5 to 17 years.
Design, Setting, and Participants
This population-based study was conducted under a public health surveillance mandate from the US Food and Drug Administration. Participants aged 5 to 17 years were included if they received BNT162b2 COVID-19 vaccination through mid 2022 and had continuous enrollment in a medical health insurance plan from the start of an outcome-specific clean window until the COVID-19 vaccination. Surveillance of 20 prespecified health outcomes was conducted in near real time within a cohort of vaccinated individuals from the earliest Emergency Use Authorization date for the BNT162b2 vaccination (December 11, 2020) and was expanded as more pediatric age groups received authorization through May and June 2022. All 20 health outcomes were monitored descriptively, 13 of which additionally underwent sequential testing. For these 13 health outcomes, the increased risk of each outcome after vaccination was compared with a historical baseline with adjustments for repeated looks at the data as well as a claims processing delay. A sequential testing approach was used, which declared a safety signal when the log likelihood ratio comparing the observed rate ratio against the null hypothesis exceeded a critical value.
Exposure was defined as receipt of a BNT162b2 COVID-19 vaccine dose. The primary analysis assessed primary series doses together (dose 1 + dose 2), and dose-specific secondary analyses were conducted. Follow-up time was censored for death, disenrollment, end of the outcome-specific risk window, end of the study period, or a receipt of a subsequent vaccine dose.
Twenty prespecified health outcomes: 13 were assessed using sequential testing and 7 were monitored descriptively because of a lack of historical comparator data.
This study included 3 017 352 enrollees aged 5 to 17 years. Of the enrollees across all 3 databases, 1 510 817 (50.1%) were males, 1 506 499 (49.9%) were females, and 2 867 436 (95.0%) lived in an urban area. In the primary sequential analyses, a safety signal was observed only for myocarditis or pericarditis after primary series vaccination with BNT162b2 in the age group 12 to 17 years across all 3 databases. No safety signals were observed for the 12 other outcomes assessed using sequential testing.
Conclusions and Relevance
Among 20 health outcomes that were monitored in near real time, a safety signal was identified for only myocarditis or pericarditis. Consistent with other published reports, these results provide additional evidence that COVID-19 vaccines are safe in children.
Three vaccines are currently available in the United States to prevent COVID-19 in children, including COVID-19 vaccines from Pfizer-BioNTech (BNT162b2) and Moderna (mRNA-1273) for ages 6 months to 17 years and Novavax (NVX-CoV2373) for ages 12 to 17 years.1-3 We present safety results from near–real-time monitoring using commercial claims databases with data for 3 017 352 children aged 5 to 17 years following administration of the COVID-19 BNT162b2 vaccine, which was the first approved vaccine for the pediatric population. This study was conducted under the US Food and Drug Administration (FDA) Biologics Effectiveness and Safety (BEST) Initiative using rapid cycle analysis, or a near–real-time monitoring framework, which enables early detection of potential safety signals. This surveillance effort in near real time is part of the US government’s larger ongoing postauthorization monitoring and collection of real-world evidence to ensure the safety of COVID-19 vaccines in children.
This surveillance activity was conducted as part of the US FDA public health surveillance mandate. This cohort study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. It was exempt from institutional review board approval because these databases contain deidentified data and did not require informed consent procedures. FDA surveillance activities under the Sentinel Initiative and BEST Initiative, which is a component of the Sentinel Initiative, are exempt from institutional review board review and approval.4
This study used administrative claims data from Optum, HealthCore (HealthCore Integrated Research Database), and CVS Health (Aetna Enterprise Data Warehouse) that contain longitudinal medical and pharmacy claims data. At the time of analysis, claims data from Optum and CVS Health were additionally supplemented with vaccination data from participating local and state immunization information systems (eTable 1 in Supplement 1).5 For BNT162b2 vaccines, claims data alone captured 19.3% of the total enrolled population, but IIS-claims linkage increased this percentage to 29.0%.
We included health plan members aged 5 to 17 years who received BNT162b2 vaccination from the earliest date of its Emergency Use Authorization by age group through June 25, 2022 (Optum), May 6, 2022 (HealthCore), and May 31, 2022 (CVS Health). The study population was divided into 3 age groups according to the vaccine authorization schedule (5-11, 12-15, and 16-17 years). We required continuous enrollment in a medical health insurance plan from the start of a clean window specific to each health outcome to the COVID-19 vaccination so that only a new incident diagnosis of an outcome during the postvaccination risk window would contribute to the rapid cycle analysis (eTable 3 in Supplement 1). (A clean window is the term for the interval used to define incident outcomes where an individual enters the study cohort only if the outcome of interest did not occur during that interval.)
Exposure was defined as the receipt of BNT162b2 COVID-19 vaccine (eTable 2 in Supplement 1). The dose number was assigned chronologically. Follow-up began on the date of eligible vaccination and was censored at subsequent vaccination, death, disenrollment, end of the outcome-specific risk window, or end of the study period. The primary analysis (dose 1 + dose 2) included all follow-up time accrued after dose 1 and after dose 2 combined. Secondary analyses included stratification by dose 1, dose 2, and dose 3 or booster, including follow-up time accrued after the individual dose up until a censoring criterion was met.
Using claims-based algorithms, we monitored 20 health outcomes selected through clinical consultation and literature review.6,7 These outcomes included myocarditis, pericarditis, or co-occurring myocarditis and pericarditis (hereafter referred to as myocarditis or pericarditis); encephalitis or encephalomyelitis; anaphylaxis; common thromboses with thrombocytopenia; seizures or convulsions; Bell palsy; deep vein thrombosis; pulmonary embolism; disseminated intravascular coagulation; immune thrombocytopenia; narcolepsy; appendicitis; nonhemorrhagic stroke; Guillain-Barré syndrome; multisystem inflammatory syndrome in children; transverse myelitis; cerebral and abdominal (unusual site) thrombosis with thrombocytopenia; Kawasaki disease; hemorrhagic stroke; and acute myocardial infarction. Among these, 13 were assessed using sequential testing and 7 were monitored descriptively because of a lack of historical comparator data.6 For each outcome for which sequential analysis was conducted, we excluded enrollees who had experienced the outcome during an outcome-specific clean window to ensure that it was not a preexisting condition before vaccination. The myocarditis or pericarditis outcome was assessed with varying outcome-specific risk windows and care settings based on evidence from prior surveillance efforts and clinician input (eTable 3 in Supplement 1).
We conducted monthly sequential testing using the Poisson maximized sequential probability ratio test8 and generated incidence rate ratios (RRs) of observed outcome rates compared with database-specific historical (expected) rates. This screening method can detect a relatively large increased risk early in the surveillance period. The method does not establish a causal association and is not intended to quantify its magnitude nor its precision, so it cannot generate confidence intervals around the incidence RRs.
Historical rates before COVID-19 vaccination were adjusted for claims processing delay and standardized by age and sex where case counts permitted.9 One-tailed tests were used with a null hypothesis that the observed rate was no greater than the historical comparator rate beyond a prespecified test margin with an α of 1%, specified for each outcome-dose–age group being sequentially tested. A stringent α level was selected to increase the specificity of signals detected from testing multiple outcomes across different analyses. The log likelihood ratio was calculated comparing the likelihoods of the observed rate ratio and the null hypothesis. At each test, if the log likelihood ratio exceeded a prespecified critical value, the null hypothesis was rejected and a safety signal was declared. A statistical signal occurred if the log likelihood ratio exceeded a critical value. Surveillance continued until a signal was detected or the prespecified maximum surveillance length was reached.6
Medical record review was conducted for the myocarditis or pericarditis outcome to assess the validity of cases identified by claims-based algorithms. Clinical adjudicators used the Brighton collaboration definition to classify cases, and records meeting the confirmed or probable Brighton classifications were considered a true case.10
Across the 3 databases, a total of 5 901 825 doses of the BNT162b2 COVID-19 vaccine were administered to 3 017 352 children aged 5 to 17 years enrolled in these commercial health plans; 1 999 550 doses were observed in 1 000 895 enrollees in CVS Health, 2 033 212 doses in 1 078 712 enrollees in HealthCore, and 1 869 063 doses in 937 745 enrollees in Optum (Figure). Demographic characteristics of the vaccinated populations were largely similar across databases (Table 1). Of enrollees across all 3 databases, 1 510 817 (50.1%) were males, 1 506 499 (49.9%) were females, and 2 867 436 (95.0%) lived in an urban area. We observed a low incidence of the 7 outcomes monitored descriptively in all 3 databases (<25 events of each outcome in each database) (eTable 5 in Supplement 1).
Of the 13 outcomes sequentially tested, only myocarditis or pericarditis met the threshold for a statistical signal in any of the 3 databases. In the primary series analyses, a signal was detected in all databases for the age groups 12 to 15 years and 16 to 17 years after BNT162b2 COVID-19 vaccination for all definitions of the outcome myocarditis or pericarditis (Table 2). In the dose-specific analyses, a signal was detected after dose 2 in all definitions of the outcome myocarditis or pericarditis in the age groups 12 to 15 years and 16 to 17 years from all data partners. After dose 3, signals were detected for some of the definitions of myocarditis or pericarditis in the age group 16 to 17 years in the CVS Health database and those aged 12 to 15 years in the HealthCore database (eTable 4 in Supplement 1). No signals were observed in the age group 5 to 11 years.
Of the 153 cases of myocarditis or pericarditis among children aged 12 to 17 years, medical record review was conducted for a sample of 37 cases whose records were obtainable. Twenty-seven of these cases (73.0%) were confirmed as true cases of myocarditis or pericarditis, of which 25 patients were male, and 19 were hospitalized with a mean length of hospital stay of 2.8 days (median, 2 days). The mean time from vaccination to presentation for care for myocarditis or pericarditis was 6.8 days (median, 3 days).
Our near–real-time monitoring results for 20 prespecified health outcomes in the pediatric population provide reassuring real-world evidence of the safety of the BNT162b2 COVID-19 vaccine in children and adolescents. The signal detected for myocarditis or pericarditis is consistent with that reported in peer-reviewed publications demonstrating an elevated risk of myocarditis or pericarditis following mRNA vaccines, especially among younger males aged 12 to 29 years.11-13 It should be noted that either myocarditis or pericarditis is a rare event, with an average incidence of 39.4 cases per million doses administered in children aged 5 to 17 years within 7 days after BNT162b2 COVID-19 vaccination.14,15 We did not detect a signal for myocarditis or pericarditis in younger children (aged 5-11 years), which is consistent with reports from other surveillance systems.16,17
This study has several strengths. First, the study included a large, geographically diverse population from 3 commercial health insurance databases in the United States. Because of the availability of more information from claims supplemented with IIS data and a short data lag from health encounters, we were able to monitor BNT162b2 COVID-19 vaccine safety in near real time. Additionally, identified cases of myocarditis or pericarditis were validated through medical record review.
The study also has some limitations. First, this study only covers data from a commercially insured pediatric population and may not be nationally representative. We used a rapid monitoring method designed for early detection of a potentially increased risk of health outcomes with limited confounding adjustment. Therefore, results of this study do not establish a causal relationship between the vaccine and health outcomes; the signals should be further evaluated. Furthermore, the study may have limited power to detect a small increase in risk of outcomes in certain subgroups with more recent authorizations such as booster doses in younger children. Analyses were not stratified by sex, which is an important demographic characteristic for certain outcomes.
Another limitation of the study is that we could not conduct record review for all outcomes included in the study because of resource, time, and legal constraints. For the myocarditis or pericarditis signals detected across the 3 databases, we reviewed medical records of a subset of identified cases because a limited number of records were obtainable. Similarly, although clinical values may have reduced outcome misclassification for some outcomes, we were unable to obtain clinical values from administrative claims without medical record review.
This cohort study monitored 20 health outcomes in near real time and identified a safety signal only for myocarditis or pericarditis, which was consistent with other published reports. These results provide additional evidence for the safety of the COVID-19 vaccines in the pediatric population. FDA continues to monitor vaccine safety and has expanded the framework to include additional age groups and vaccine brands with updated authorizations. The FDA BEST Initiative plays a major role in the larger US federal government vaccine safety monitoring efforts and further supports decision-making concerning the safety of COVID-19 vaccines by health care professionals and the public.
Accepted for Publication: April 6, 2023.
Published Online: May 22, 2023. doi:10.1001/jamapediatrics.2023.1440
Corresponding Author: Steven A. Anderson, PhD, MPP, Office of Biostatistics and Pharmacovigilance, Center for Biologics Evaluation and Research, US Food and Drug Administration, 10903 New Hampshire Ave, Silver Spring, MD 20993 (email@example.com).
Author Contributions: Mr Hu and Ms Feng had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Hu, Lloyd, Smith, Seeger, Reich, Do, Chillarige, Cooper, Shoaibi, Forshee, Anderson.
Acquisition, analysis, or interpretation of data: Hu, Wong, Feng, Smith, Amend, Kline, Beachler, Gruber, Mitra, Harris, Secora, Obidi, Wang, Song, McMahill-Walraven, Reich, McEvoy, Do, Chillarige, Clifford, Shoaibi, Forshee, Anderson.
Drafting of the manuscript: Hu, Wong, Lloyd, Smith, Mitra, Wang, Cooper, Shoaibi, Anderson.
Critical revision of the manuscript for important intellectual content: Hu, Wong, Feng, Lloyd, Amend, Kline, Beachler, Gruber, Mitra, Seeger, Harris, Secora, Obidi, Song, McMahill-Walraven, Reich, McEvoy, Do, Chillarige, Clifford, Shoaibi, Forshee, Anderson.
Statistical analysis: Hu, Wong, Feng, Smith, Amend, Mitra, Harris, Secora, Wang, Song, McEvoy, Chillarige, Clifford, Forshee.
Obtained funding: Anderson.
Administrative, technical, or material support: Hu, Wong, Lloyd, Smith, Kline, Gruber, Mitra, Secora, Obidi, McMahill-Walraven, Do, Chillarige, Cooper, Shoaibi.
Supervision: Mitra, Seeger, McMahill-Walraven, Reich, Chillarige, Shoaibi, Forshee, Anderson.
Conflict of Interest Disclosures: Dr Amend reported equity or stocks from Optum outside the submitted work. Ms Kline reported working on grants, subcontracts, or contracts from Harvard Pilgrim Health Care Institute, Brown University (National Institute on Aging, IMPACT Collaboratory), Reagan Udall Foundation for the FDA, Academy of Managed Care Pharmacy’s Biologics and Biosimilar Collective Intelligence Consortium (BBCIC), TherapeuticsMD, REACHnet (Louisiana Public Health Institute), IQVIA, Pfizer, HealthCore, and Patient-Centered Outcomes Research Institute as an employee of CVS Health and reported stock or stock options from CVS Health. Dr Beachler is an employee of HealthCore who has previously contracted with Pfizer for separate projects. Dr Seeger reported stock or stock options in UnitedHealth Group. Dr Harris reported other from Harvard Pilgrim Health Care Institute, Reagan Udall Foundation for the FDA, and Academy of Managed Care Pharmacy’s BBCIC and stock or stock options in CVS Health outside the submitted work. Ms Song reported equity or stocks from Optum outside the submitted work. Dr McMahill-Walraven reported working on grants, subcontracts, or contracts from Harvard Pilgrim Health Care Institute, Brown University (National Institute on Aging, IMPACT Collaboratory), Reagan Udall Foundation for the FDA, Academy of Managed Care Pharmacy’s BBCIC, TherapeuticsMD, REACHnet (Louisiana Public Health Institute), IQVIA, Pfizer, Janssen, HealthCore, Patient-Centered Outcomes Research Institute, National Evaluation System for Health Technology Coordinating Center, and Harvard Pilgrim HealthCare Institute as an employee of CVS Health and reported stock or stock options from CVS Health. Ms Cooper reported other from Harvard Pilgrim Health Care Institute, Brown University (National Institute on Aging, IMPACT Collaboratory), Reagan Udall Foundation for the FDA, Academy of Managed Care Pharmacy’s BBCIC, TherapeuticsMD, REACHnet (Louisiana Public Health Institute), IQVIA, Pfizer, HealthCore, and Westat as an employee of CVS Health outside the submitted work. Ms Clifford reported stock or stock options in UnitedHealth Group. No other disclosures were reported.
Funder/Support: The US Food and Drug Administration provided funding for this study.
Role of the Funder/Sponsor: The US Food and Drug Administration contributed as follows: led the design and conduct of the study; contributed to the coordination of collection, management, and analysis of the data; led the interpretation of the data; reviewed and approved the manuscript; and submitted the manuscript for publication.
Data Sharing Statement: See Supplement 2.
Additional Contributions: We thank Tainya C. Clarke, PhD (project management), and Kristin A. Sepúlveda, MBA (project management), of the US Food and Drug Administration; Anchi Lo, MS, MPH (data analysis), Ruobing Lyu, MPP (data analysis), Derick Ambarsoomzadeh, BA (data analysis), Gyanada Acharya, MSc (writing), Laurie Feinberg, MD, MPH, MS (editing), Sandia Akhtar, BS (editing), Shruti Parulekar, MPH (data analysis), Yiheng Zhu, MPP (data analysis), William (Trey) Minter, BA (writing), Yeerae Kim, MPH (data analysis), Yixin Jiao, MPP (editing), and Zhiruo (Cassie) Wan, MS (data analysis), of Acumen; Ana M. Martinez-Baquero, MA (data analysis), Carla Brennan, Nancy B. Sheffield-Shaik, BS (data collection and analysis), Smita Bhatia, MCA (data analysis), and Vaibhav Sharma, MS (data collection), of CVS Health; Shiva Vojjala, MS (data collection), Ramya Avula, MS (data collection), Shiva Chaudhary, MCA (data collection), Shanthi P. Sagare, MS (data collection), Ramin Riahi, BS (data collection), Navyatha Namburu, MS (data collection), and Grace Stockbower, MPH (data collection), of HealthCore; Michael Goodman, PhD (data analysis), Michael Bruhn, MBA (editing), and Ruth Weed, MS (editing), of IQVIA; Grace Yang, MPA, MA (data collection), Karen Schneider, PhD (data collection), Rebecca Braun, BA (data collection), Megan Ketchell, BS (data collection), and Kate Federici, MSW (data collection), of Optum. No compensation was received by any of these individuals.