Accuracy of ICD-10 codes for identifying hospitalizations for acute anticoagulation therapy-related bleeding events

Introduction: Administratively coded data are frequently used in observational research to identify outcome events. With the transition to the new International Classification of Diseases coding system's 10th version (ICD-10), information is needed about the coding accuracy for bleeding events in anticoagulated patients. We aimed to determine ICD-10 code accuracy for bleeding events in anticoagulated patients admitted to the hospital. Methods: This cross-sectional study retrospectively examined charts of anticoagulated patients who were admitted to the University of Utah Hospital between October 1, 2017 and December 31, 2017. Two trained chart abstractors blinded to ICD-10 code status independently reviewed medical charts to determine the presence or absence of bleeding events. ICD-10 code status in any diagnosis position was unblinded and code accuracy was assessed using sensitivity, specificity, positive predictive value (PPV), and negative predictive values (NPV) along with 95% confidence intervals (CI). Results: Out of 661 admissions, 487 unique patients and 71 bleeding events were identified. Gastrointestinal tract bleeding and intracranial hemorrhage comprised 32.4% and 19.7% of bleeding events respectively. ICD-10 code sensitivity was 91.4% (95% CI, 82.3-96.8), specificity was 90.2% (87.5-92.5), PPV was 52.5% (43.2-61.6) and NPV 98.9% (97.6-99.6). Individual codes for intracranial hemorrhages and gastrointestinal tract bleeding had similar accuracy as the overall set of bleeding codes. Conclusions: Our results demonstrate that ICD-10 codes can reliably rule-out hospitalizations for bleeding events in patients receiving anticoagulation therapy. Due to unacceptable false positive rates ICD-10 codes should not be used for identifying bleeding complications without confirmatory chart review.