Validity of ICD-10 codes for identifying gastrointestinal bleeding in anticoagulated patients: an encounter-level...

International Classification of Diseases, Tenth Revision (ICD-10) encounter diagnosis codes are commonly used to identify gastrointestinal bleeding (GIB) in electronic health record (EHR)-based research and clinical decision-support (CDS) tools. However, variability in coding accuracy may introduce misclassification bias, particularly among patients receiving oral anticoagulants (OACs). Contemporary, institution-specific validation of ICD-10 GIB codes is needed in the post-ICD-10 implementation era. We conducted a diagnostic accuracy study of ICD-10 encounter diagnosis codes for GIB among adults receiving warfarin or direct oral anticoagulants (DOACs) within a large academic healthcare system. A random sample of inpatient and emergency department encounters between June and December 2024 was identified. Manual chart review served as the reference standard. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic odds ratio (DOR) were calculated overall and stratified by anatomic bleeding site, anticoagulant class, and admission status. Among 656 anticoagulated encounters, 26 (4.0%) had chart-confirmed GIB. Overall sensitivity was 81% (95% CI, 60%-93%), specificity 98% (95% CI, 96%-99%), PPV 64% (95% CI, 45%-79%), and NPV 99% (95% CI, 98%-100%). The DOR was 215.6 (95% CI, 69.6-667.6). Sensitivity was higher for upper versus lower GIB (86% vs. 50%), while specificity remained high across subgroups. Diagnostic performance was similar between DOAC- and warfarin-treated encounters, though estimates for warfarin were imprecise due to small event counts. PPV varied across individual ICD-10 codes, with symptom-specific codes demonstrating higher predictive accuracy. ICD-10 encounter diagnosis codes demonstrated high specificity and NPV for identifying GIB among patients receiving OACs, supporting their use for ruling out bleeding events in EHR-based research and CDS. However, moderate PPV and heterogeneity across codes and clinical contexts indicate that ICD-10 codes alone should be used cautiously when defining GIB outcomes. Local validation remains essential to ensure reliable bleeding identification in anticoagulated populations.