Validation of ICD-10-CM Codes for Bleeding in Patients with an Indication for Antithrombotic Therapy

Background: International Classification of Disease Clinical Modification (ICD-10-CM) codes are used to identify adverse drug events. Our goal was to assess accuracy of ICD-10-CM code assignments in billing data for identifying hospital encounters for overt bleeds. Methods : Multicenter, case-control study of patients with indications for antithrombotic therapy and who had an emergency department or hospital admission. Cases identified by ICD-10-CM codes for bleeding were matched 2:1 to controls. Medical record abstraction was the reference standard for determining bleeding occurrence and anatomical location. Diagnostic accuracy of ICD-10-CM codes for bleeding overall was calculated while adjusting for oversampling of cases. Results: Among 4803 patients (3140 cases, 1663 controls), ICD-10-CM codes for bleeds in any position had positive predictive value (PPV) of 88.5% (95% CI, 87.4%–89.7%) and adjusted sensitivity of 75.2% (70.9%–79.4%). Intracranial bleed codes had the highest PPV (93.1%; 95% CI, 91.3%–94.9%) and adjusted sensitivity (100%; 95% CI, 94.0%–100.0%). Gastrointestinal bleed codes and codes for other bleeds had PPVs (95% CI) of 83.1% (81.2%–85.0%) and 77.2% (73.9%–79.0%) and adjusted sensitivities (95% CI) of 84.8% (78.5%–91.1%) and 52.2% (46.2%–58.8%), respectively. Restricting codes to the primary diagnosis position increased PPV of any bleed to 97.2% (96.5%–97.9%), but decreased adjusted sensitivity to 44.9% (40.79%–49.6%). Codes for antithrombotic poisoning/adverse effect had PPV of 88.4% (83.9%–93.0%) but adjusted sensitivity of 5.6% (5.3%–5.9%). Conclusions: When using ICD-10-CM codes to identify overt bleeding in patients who have an indication for antithrombotic therapy, codes in the primary position have the highest PPV. Use of codes in any position maximizes sensitivity.