Empirical Calibration

Only empirically calibrated results were reported. The study used 180 negative control outcomes to detect and correct for residual systematic error in hazard ratio estimates. Effect estimates from databases passing diagnostics were pooled using Bayesian random-effects meta-analysis with non-normal likelihood approximation to reduce bias from small event counts. Cox proportional hazards models conditioned on propensity score matching strata estimated calibrated hazard ratios, which were reported with 95% confidence intervals and p values.