A strategy for assessment and validation of major bleeding cases in a primary health care database in Spain

• Published in: Pharmacoepidemiology and drug safety Vol. 30; no. 12; pp. 1696 - 1702
• Main Authors: Burgos‐Gonzalez, Airam, Bryant, Verónica, Maciá‐Martinez, Miguel Angel, Huerta, Consuelo
• Format: Journal Article
• Language: English
• Published: Chichester, UK John Wiley & Sons, Inc 01.12.2021; Wiley Subscription Services, Inc
• Subjects: Algorithms; Anticoagulants; Bleeding; Databases, Factual; Electronic health records; Electronic medical records; Gastrointestinal Hemorrhage; Health care; Hemostasis; Hospitalization; Humans; Localization; major bleeding; pharmacoepidemiology; Primary care; Primary Health Care; Spain - epidemiology; Thrombosis; validation studies; Spain; electronic health records; validation studies; major bleeding; pharmacoepidemiology
• ISSN: 1053-8569; 1099-1557; 1099-1557
• DOI: 10.1002/pds.5357

Purpose This study aims to validate major bleeding (MB) cases within a cohort of new users of direct oral anticoagulants (DOACs) in Electronic health records (EHRs) from primary care in Spain (BIFAP), introducing more efficient techniques and automating the process of validation in the pharmacoepidemiologic research with EHR data as much as possible. Methods Registered bleedings were identified in a cohort of new users of DOACs in BIFAP using ICPC 2 and ICD 9 codes; we ascertained these bleedings as MB through a validation strategy based on the MB definition from the International Society on Thrombosis and Hemostasis, which used hospitalization and critical localization as proxies. We assessed hospitalization with hospital discharge information (only available for some years and regions) and a free text‐based algorithm created to identify hospitalization in EHR's clinical notes. Incidence rates (IR) of MB were evaluated by bleeding type. Results The study cohort included 104 614 patients, with 274521.5 p‐y of follow up. There were 6143 registered bleedings during the study period (519 intracranial bleeding – ICB, 4606 gastrointestinal bleeding – GIB, 1018 extracranial bleeding – ECB), from which 1679 were confirmed as MB (416 ICB, 1086 GIB, and 177 ECB). The free text‐based semi‐automatic algorithm had moderate recall (0.59), but high specificity (0.99), and precision (0.94). Conclusion The combination of hospitalization and critical localization is a valid approach to validate MB in EHRs with incomplete information. The use of more automatic methods for case validation instead of manual review of clinical notes is favored.