Tarazjani AD, de Luise C, Manders E, Ryan O, Hoxhaj V, Aguado J, Weinrib R, Marsal JR, Desalegn A, Lupattelli A, Zidan M, Mohammadi S, Roy D, Lane S, Aurelius T, Morris D, Pala E, Pazos I, Yefimenko N, Brutau A, Martinez FV, Overbeek J, van den Berg JA, Gaspersz J, Garcia RP, Sanz MP, Poblador-Plou B, Santos-Mejias A, Gutierrez AG, Gimeno-Miguel A, Moreno-Juste A, Stona L, Barbieri E, Gini R, Arana A, Sturkenboom M. Validation of algorithms for the identification of myocarditis and pericarditis in real-world data to evaluate the safety of the Pfizer-BioNTech COVID-19 vaccine (COMIRNATY): a post-authorisation safety study in the VAC4EU network. Presented at the ISPE 42nd Annual Meeting; August 31, 2026. Milan, Italy.

BACKGROUND: Adverse events of special interest (AESI) are assessed in post-authorization safety studies (PASS) using real-world data, but the accuracy of codes and algorithms to identify these AESIs requires validation studies in line with regulatory guidance.

OBJECTIVES: To estimate the positive predictive value (PPV) of diagnostic algorithms for myocarditis and pericarditis in a retrospective PASS of COMIRNATY in Europe.

METHODS: We validated a sample of myocarditis and pericarditis cases in six data sources (CPRD Aurum, Pedianet, EpiChron, NHR, PHARMO, SIDIAP) using the VAC4EU pipeline. Trained abstractors, blinded to COVID-19 vaccination status, reviewed clinical documentation using standardized REDCap forms and assigned levels of certainty (LOC 1= confirmed, LOC 4 = non-assessable, LOC 5= not a case) based on an adapted Brighton Collaboration (BC) case definition. Validation data access differed by data sources: NHR used full hospital chart review; EpiChron had linked hospital discharge reports and diagnostic results; SIDIAP and PHARMO were hospital‑linked but had limited confirmatory hospital documentation; CPRD was based on patient profile datasets generated using CPRD Aurum primary care GP data. PPVs (95% CIs) were calculated by dividing the number of confirmed cases by all cases.

RESULTS: The highest PPVs were observed in hospital-linked systems. In NHR and EpiChron, myocarditis:73.6% (67.7-78.8%) and 74.7% (64.2-83.3%); pericarditis: 61.1% (57.4-69.5%) and 69.1% (63.7-74.2%), respectively. In contrast, PPVs were lower when confirmatory hospital documentation was limited. In PHARMO and CPRD, myocarditis:22.2% (13.2-32.7%) and 6.3% (3.9–9.4%); pericarditis:14.7% (8.6-33.0%) and 5.3% (3.2–8.3%) respectively. PPVs in SIDIAP were: 29.3% (24.4–34.7%) for myocarditis and 29.5% (24.5–34.9%) for pericarditis. Pedianet results were limited by very small sample size. Differences in PPV were mirrored by the LOC4 frequency: LOC4 was highest in CPRD/PHARMO/SIDIAP and lowest in EpiChron (myocarditis: 93.8% in CPRD, 2.5% in EpiChron; pericarditis: 94.7% in CPRD, and 22.1% in EpiChron). Availability of key BC criteria was highest in NHR and EpiChron (symptoms and ECG >92%; biomarkers >80%; echocardiography ≥74%) and more limited in SIDIAP (ECG < 33%, biomarkers < 27%, echocardiography < 25%) and GP-profile–based sources.

CONCLUSIONS: The PPVs for myocarditis and pericarditis varied by data source. Hospital-based systems achieved the highest PPVs, while some primary care data sources without full hospital linkage demonstrated lower PPVs. These results underscore the importance of availability of clinical evidence (e.g. imaging and labs) when planning validation studies of algorithms used in vaccine safety studies.