Lambda-OR

Initial Release. Attribution in large-scale observational studies is undermined by outcome misclassification; label noise, magnified by large sample sizes, can inflate weak associations into significant risk factors or spurious “protective” effects. We introduce the lambda-Odds Ratio (Λ-OR), a misclassification-corrected odds ratio that incorporates sensitivity–specificity variance and carries formal guarantees: it collapses under non-informative predictors and resists amplification of vanishing effects. Applied to Alzheimer’s disease and related dementias (ADRD), idiopathic pulmonary fibrosis (IPF), and autism spectrum disorder (ASD), Λ-OR suppresses the inflated significance seen with classical odds ratios and SHAP while preserving biologically credible modulators. These results establish Λ-OR as a robust
framework for attribution in biobank- and EHR-scale studies.