ICD-10 code validation for gastrointestinal bleeding in patients receiving oral anticoagulation: nationwide cohort study

ICD-10 codes are commonly used to identify gastrointestinal bleeding (GIB) events in observational studies. However, most studies have neither used validated ICD-10 coding algorithms to identify GIB events nor manually confirmed the diagnosis of GIB, which raises concerns regarding the validity of their results. The aim of the study was to estimate the accuracy of ICD-10 codes for GIB in patients receiving oral anticoagulation and to develop new coding algorithm for identifying GIB events. Using a national prescription database, all patients receiving oral anticoagulation in Iceland from 2014 to 2019 were identified. GIB events were identified using thorough search of ICD-10 codes from all five major hospitals in Iceland. To augment the code search, endoscopy results and death cause in the national death registry were reviewed. All diagnoses were manually confirmed via chart review and used as gold standard. The PPV was calculated for individual codes, and codes with PPV ≥ 75% were included in the ICD-10 coding algorithm. The ICD-10 coding algorithm had 61.3% sensitivity, 99.6% specificity, 90.0% PPV, and 98.0% negative predictive value. GIB events identified using this ICD-10 coding algorithm were more commonly major bleeding events, and more commonly required hospitalization, anticoagulation reversal, anticoagulation discontinuation, and endoscopic intervention compared to GIB events that were missed. The ICD-10 coding algorithm reported here had good accuracy in identifying true GIB events and the events that it identified were clinically more severe than those that it missed. The sensitivity of the ICD-10 coding algorithm was modest and can be improved by reviewing endoscopic results and death registries.