Fuller CC, Kawai AT, Daniels K, Iyer GS, Koram N, Avula R, Brown JS, Cai B, Clasen SC, Hawrusik R, Layton JB, Love SAM, Martinez K, Messenger-Jones E, Paullin M, Rabbani R, Reynolds JS, Wideman K, Uribe W, Platt R, Gilsenan AW. Validation of a claims-based algorithm for Myocarditis/pericarditis and assessment of the impact of outcome misclassification in a US postapproval study of COVID-19 vaccine safety. Poster presented at the ISPE 42nd Annual Meeting; September 1, 2026. Milan, Italy.

BACKGROUND: Myocarditis/Pericarditis (m/p) is a potential safety concern after mRNA COVID-19 vaccination. Data are lacking on the validity of claims-based algorithms for m/p. As part of a US postapproval study of the original monovalent Pfizer BioNTech COVID-19 vaccine (BNT162b2), we validated an ICD-10 code algorithm for m/p (EUPAS43468).

OBJECTIVES: To estimate the positive predictive value (PPV) of a claims-based m/p algorithm in a population of BNT162b2 vaccinees and matched unvaccinated individuals, and to conduct quantitative bias analyses (QBA) examining the potential for outcome misclassification in the final study results.

METHODS: Potential cases of m/p were identified in individuals vaccinated with BNT162b2 from Dec 2020-Apr 2023 and matched unvaccinated comparators within a 21-day risk window using data from 5 insurers participating in the Sentinel System. Validation was conducted in a sample of individuals aged ≥ 12 years with a recorded m/p diagnosis at one insurer. Two clinicians blinded to exposure independently reviewed medical records and classified potential cases using Brighton-adapted definitions (primary: definitive/probable; secondary: definitive/probable or clinician diagnosis). PPVs were calculated as confirmed cases among algorithm-identified cases, overall, by care setting, and exposure status using two denominators: all reviewed cases and those with sufficient information to confirm status. QBA was performed as a sensitivity analysis, using estimated exposure-specific PPVs to calculate corrected incidence rate ratios (IRR). Sensitivity was assumed to be 5-20% higher in the exposed than the unexposed, and the sensitivity in the exposed was assumed to be 80% or 100%.

RESULTS: Of the 277 cases for whom medical records could be requested, 163 (59%) had at least 1 medical record available, and 128 (46%) had sufficient information available in medical records to determine case status. Overall PPVs were 36%/46% for the primary and 41%/49% for the secondary definition (all cases vs. only those with sufficient information). PPVs were highest in inpatient (63%/71%) and emergency settings (53%/75%), and lowest in ambulatory settings (16%/27%). PPVs were lower among exposed (28%/39%) than unexposed individuals (42%/58%). The observed primary series IRR in the final analysis was 1.14, whereas the corrected IRRs for m/p ranged from 0.55- 0.70.

CONCLUSIONS: The PPV of the m/p algorithm was < 50% but increased when restricted to inpatient events. In QBA, the corrected IRR was attenuated to below the null, suggesting that the IRR observed in main study was not underestimated relative to the true IRR due to outcome misclassification.