Jaba Tkemaladze
Background: Post‑authorisation vaccine safety studies are prone to confounding by indication, healthy vaccinee bias, and differential healthcare seeking. Emulating a pragmatic randomised trial with observational data can reduce some biases, but residual confounding remains a concern for rare adverse events following COVID‑19 vaccination. Objective: To estimate the 12‑month risk of Guillain‑Barré syndrome (GBS), myocarditis, and a composite autoimmunity outcome (immune thrombocytopenia, autoimmune haemolytic anaemia, inflammatory arthritis) after mRNA [4,8] versus adenoviral vector (ChAdOx1 nCoV‑19, Ad26.COV2.S) vaccination, using a target trial emulation framework with advanced bias detection and quantitative calibration methods. Design: Emulation of a hypothetical pragmatic randomised trial with sequential eligibility periods, using weighted difference‑in‑differences (DID), negative control outcome calibration (Control Outcome Calibration Approach — COCA), and sensitivity analyses for unmeasured confounding including E‑values on the risk difference scale. Setting: Population‑based electronic health records and claims data from the United States (Optum de‑identified EHR) and United Kingdom (CPRD Aurum linked to Hospital Episode Statistics), December 2020 to June 2024. Participants: Adults aged 18–64 years without prior SARS‑CoV‑2 infection or COVID‑19 vaccination, initiating first‑dose vaccination with either an mRNA (n=2,431,890) or adenoviral vector (n=452,180) vaccine. Intervention: Assignment to mRNA vaccine versus adenoviral vector vaccine at baseline, followed according to the licensed schedule with 7‑day grace period. Main outcome measures: Cumulative incidence at 12 months estimated via inverse probability weighting (IPW) to adjust for 42 baseline covariates, combined with a weighted DID approach using pre‑exposure historical trends (12 months before vaccination). Six negative control outcomes (e.g., cholelithiasis, appendicitis, fall‑related fractures) were used for both qualitative and quantitative bias detection via COCA. E‑values were calculated on the risk difference scale for primary findings. A mechanistic literature synthesis was performed to validate biological plausibility. Results: After IPW, all baseline covariates were balanced (absolute standardised difference 3.1‑fold to explain away the association. A benchmarking analysis comparing nine adjustment methods showed that our Weighted DID + COCA approach produced the smallest mean bias across negative controls (mean bias –0.4 per 100,000) compared with unweighted DID (mean bias –4.1 per 100,000). A booster‑censor analysis (censoring follow‑up at receipt of any additional COVID‑19 vaccine dose) yielded consistent results (RD +8.1 per 100,000, 95% CI 3.8 to 12.4). Sensitivity analyses (per‑protocol, 6‑month follow‑up, restriction to Omicron period, negative exposure control) all confirmed the primary findings. Mechanistic literature synthesis identified plausible biological pathways for the observed myocarditis signal: mitochondrial vulnerability with RIPK3‑mediated necroptosis in young male cardiomyocytes, protective effects of estrogen signalling, and a CXCL10/IFN‑γ cytokine cascade following mRNA vaccination. Conclusions: In this target trial emulation with weighted DID, negative control calibration (COCA), benchmarking, and mechanistic validation, mRNA vaccines were associated with a small absolute increase in 12‑month risk of myocarditis in young males compared with adenoviral vector vaccines (approximately 8.3–8.7 additional cases per 100,000). No significant differences were observed for GBS or composite autoimmunity outcomes. The absence of effects on negative control outcomes — including in the high‑risk subgroup — and the sensitivity analyses support minimal residual confounding. These findings provide comparative causal evidence for vaccine safety and illustrate a rigorous methodological template for post‑market surveillance.