Detecting Contradictory Evidence on Drug-Disease Associations at Scale

DrugCon detects contradictions in drug–disease efficacy claims across 224,756 RCT abstracts with an F1 of 0.744, and explains disagreements using study-level factors.

• Why it matters: Contradictory findings across randomized controlled trials hinder evidence-based medicine, and prior automation often labels sentence pairs without clarifying why results differ.
• What they did: DrugCon extracts RCT conclusions, links claims for the same drug–disease association, uses counterfactual reasoning to flag contradictions, and attributes them to a taxonomy of ten study-level factors. It was evaluated on DrugCon-Bench with 542 contradictory claim pairs from 819 abstracts.
• The result: Most contradictions reflect multiple study-level factors (75.5%), especially outcome-measure differences (64.0%), population differences (51.3%), and dosage or exposure-duration differences (35.6%); it also identifies explanatory factors with an F1 of 0.482.

AI summary of the abstract · may contain errors · not medical advice · read the paper.