Going Beyond Administrative Data: Retrospective Evaluation of an Algorithm Using the Electronic Health Record to Help Identify Bleeding Events Among Hospitalized Medical Patients on Warfarin

• Published in: American journal of medical quality Vol. 32; no. 4; p. 391
• Main Authors: Moriarty, James P, Daniels, Paul R, Manning, Dennis M, O'Meara, John G, Ou, Narith N, Berg, Tamara M, Haag, Jordan D, Roellinger, Daniel L, Naessens, James M
• Format: Journal Article
• Language: English
• Published: United States 01.07.2017
• Subjects: Algorithms; Anticoagulants - adverse effects; Clinical Protocols; Electronic Health Records - organization & administration; Female; Hemorrhage - chemically induced; Hemorrhage - prevention & control; Humans; Male; Patient Safety; Quality Improvement - organization & administration; Reproducibility of Results; Retrospective Studies; Warfarin - adverse effects; algorithms; electronic health records; hemorrhage; warfarin
• ISSN: 1555-824X
• DOI: 10.1177/1062860616660757

To reliably assess quality, a standardized electronic approach is needed to identify bleeding events. The study aims were the following: (1) clinically validate an electronic health record-based algorithm for bleeding and (2) assess interrater results to determine validity and reliability. Data were analyzed before and after implementation of a pharmacist-managed warfarin protocol. Bleeding was based on ≥2 of 3 criteria: (1) diagnosis indicating bleeding, (2) lab value decrease suggesting bleeding, and (3) blood product use. All suspected bleeds (234) and a sample (58) not meeting criteria were compared with clinical review. There were 234 bleeding cases identified electronically. Reviewer agreement was 78.2% (κ = 0.565). Algorithm sensitivity was 93.9% and positive predictive value 46.2%. Algorithm identification was least accurate for those with only 2 criteria but good for those with all criteria. This study supports using multiple electronic criteria to identify bleeding events. However, cases having exactly 2 criteria may require manual review for validation.