Replication Data for: Lambda-OR

Lambda-OR is a statistical method and accompanying software package for estimating odds ratios in the presence of outcome misclassification when the outcome is defined using a predictive score or proxy label. The method was developed to address a common problem in large observational datasets such as electronic health records, claims data, and large-scale screening pipelines: researchers often rely on high-performing predictive models to define case–control cohorts, but these proxy labels are not perfectly accurate. Even small amounts of misclassification can bias effect-size estimates and distort statistical significance, particularly when analyses are conducted on very large cohorts. The Lambda-OR framework provides a principled correction for this setting. It uses the predictive model’s operating characteristics to construct a misclassification operator linking the observed proxy outcome to the latent true outcome. The observed 2×2 contingency table between exposure and proxy outcome is then corrected by inverting this operator. Because the operator can be ill-conditioned in realistic settings, the method introduces a ridge-stabilized inversion that ensures the reconstructed counts remain non-negative and numerically stable. The corrected table is then used to compute a misclassification-adjusted log odds ratio, together with a variance estimate that accounts for both sampling variability in the contingency table and uncertainty in the proxy model’s sensitivity and specificity as estimated from a validation cohort.