Abstract Number: PB0432
Meeting: ISTH 2021 Congress
Theme: Hemophilia and Rare Bleeding Disorders » Acquired Hemorrhagic Coagulation Disorders
Background: Bleeding events are a critical outcome that must be accurately identified in observational and studies of hospitalized patients. Given major limitations in the use of diagnostic billing codes alone, manual chart review is often required for accurate identification of bleeding events. Validated algorithms are needed for accurate, non-manual identification of bleeding events in electronic health record (EHR) systems.
Aims: We developed and preliminary validated an algorithm to detect bleeding in hospitalized patients, ie, a “computable phenotype”.
Methods: We developed a “computable phenotype”, or specialized algorithm, to screen EHR medical records for bleeding events. We included all admissions to the University of Vermont (UVM) Medical Center between 2010-19. Components of the phenotype included international classification of disease (ICD)-9 and ICD-10 discharge diagnoses that were present on admission, laboratory values, current procedure terminology codes, and use of transfusion support. Clinically relevant non-major bleeding (CRNMB) and major bleeding (MB) were defined according to ISTH definitions. To improve accuracy, separate algorithms were developed by bleeding site (Table).
Results: 
Figure: 118 charts identified by the algorithm as MB were reviewed: 100 were confirmed as MB, 4 had CRNMB and in 12 hospitalizations a bleed could not be identified. 34 charts identified by the algorithm as CRNMB were reviewed: 15 of them were labeled as MB, 10 confirmed as CRNMB and in 9 hospitalizations no bleed was identified
| GI | GU | GYN | ICH | Pericardial | Retroperitoneal | Intra-ocular | Nasal | Miscellaneous | |
| Total validated | 29 | 14 | 9 | 10 | 7 | 13 | 7 | 10 | 37 |
| Accurate validation | 22 (75.9%) |
6 (42.8%) | 7 (77.8%) | 9 (90%) |
7 (100%) |
9 (69.2%) |
3 (42.8%) |
3 (30%) |
23 (62.1%) |
| No bleed identified/Unclear | 7 | 4 | 2 | 1 | 0 | 1 | 3 | 5 | 10 |
| Bleeding associated to a different site | 1 | 4 | 0 | 0 | 0 | 3 | 1 | 2 | 4 |
Table: Specific sites of bleeding as detected by Electronic Health Record compared to chart validation
Among 62,468 admissions, our algorithm identified 10,202 in-hospital bleeding events; 4,650 were CRNMB and 5,552 were MB. On manual validation of 150 charts; the algorithm positive predictive validity (PPV) was 86% (100/116) for MB and 29% (10/34) for CRNMB, with an overall PPV of 84% (126/150) for any bleeding event. The algorithm performed well for all bleeding sites (Table). Manual review determined that 41% (15/34) CRNMB events were MB events. (Figure).
Conclusions: We developed a computable phenotype for bleeding which can be applied to EHR systems. The computable phenotype was 86% specific for major bleeding and able to identify the bleeding site. This computable phenotype however, requires further validation and will serve as basis in future studies.
To cite this abstract in AMA style:
Gergi M, Wilkinson K, Koh I, Munger J, Al-Samkari H, Smith N, Roetker NS, Cushman M, Holmes C, Zakai N. Development of a Computable Phenotype for Hospital-associated Bleeding: The Medical Inpatients Thrombosis and Hemostasis (MITH) Study [abstract]. Res Pract Thromb Haemost. 2021; 5 (Suppl 2). https://abstracts.isth.org/abstract/development-of-a-computable-phenotype-for-hospital-associated-bleeding-the-medical-inpatients-thrombosis-and-hemostasis-mith-study/. Accessed August 19, 2022.« Back to ISTH 2021 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/development-of-a-computable-phenotype-for-hospital-associated-bleeding-the-medical-inpatients-thrombosis-and-hemostasis-mith-study/