A parametric model for time dependent forecasting of post COVID-19 vaccination cardiovascular risk

This study develops and validates a mathematical survival framework to estimate the time-dependent cumulative risk of post-vaccination cardiovascular outcomes following COVID-19 vaccination, accounting for vaccine platform, dose number, and time since vaccination. Existing observational studies often report hazard ratios (HRs) only for discrete follow-up intervals, which limits continuous risk forecasting. We formulated a parametric survival model with a Weibull baseline hazard and a log-linear predictor in which vaccine type is represented by categorical (dummy) variables, alongside dose number and a vaccine–dose interaction term. To align with interval-reported evidence, model parameters were calibrated by minimizing the discrepancy between model-implied HR profiles and published HRs across reported time bins and vaccine–dose strata over a 0–6 week post-vaccination window, while producing continuous cumulative risk trajectories. External validation used Swedish cohort evidence (Xu et al., 2025) for extrasystoles and myopericarditis. The model generated monotone, nonlinear increases in cumulative risk and clear heterogeneity across vaccine–dose strata, with patterns consistent with a non-additive vaccine–dose effect. In the Swedish calibration/validation setting, mRNA vaccines showed higher short-term risk accumulation than AstraZeneca, in agreement with the reported HR ordering. Quantitative agreement with the published HR time profiles (dose 2, weeks 1–6; normalized trajectory comparison) yielded pooled RMSE values of approximately 0.61 for extrasystoles and 1.21 for myopericarditis. For interpretability, the model reports absolute cumulative risk at fixed horizons; at week 6 for dose 2, predicted cumulative risk was about 0.160% (Moderna), 0.138% (Pfizer), and 0.097% (AstraZeneca). We introduce an interpretable parametric framework that converts interval-reported HR evidence into time-continuous cumulative risk forecasts and captures vaccine–dose interaction effects for defined outcomes (here, extrasystoles and myopericarditis). Although demonstrated using Swedish cohort evidence, the approach is transferable and can support decision-making in African settings once locally calibrated to regional surveillance data or scenario-based inputs.