PB0994 - Validation of the IMPROVE Hospital-Acquired Bleeding Risk Assessment Model in The Medical Inpatients Thrombosis and Hemostasis Study (MITH) Population

Tuesday, July 12, 2022

6:30 PM – 7:30 PM

Location: Exhibition

Presentation Format: In-Person

Presenting Author(s)

KW

Katherine S. Wilkinson, MS

Statistician
The Larner College of Medicine at the University of Vermont
Burlington, Vermont, United States

Co-Author(s)

Andrew D. Sparks, MS

Statistician
The Larner College of Medicine at the University of Vermont
Burlington, Vermont, United States
Mansour Gergi, MD

Assistant Professor
Hematology/Oncology Division, Department of Medicine, The Larner College of Medicine at the University of Vermont
Burlington, Vermont, United States
AR

Allen Repp

Internist
University of Vermont
Burlington, Vermont, United States
Hanny Al-Samkari, MD

Assistant Professor of Medicine
Harvard Medical School
Boston, Massachusetts, United States
NR

Nicholas Roetker

Epidemiologist
Chronic Disease Research Group
Minneapolis, Minnesota, United States
Neil A. Zakai, MD, MSc

Professor
Hematology/Oncology Division, Department of Medicine, The Larner College of Medicine at the University of Vermont
Burlington, Vermont, United States

Background: Hospital-acquired (HA) bleeding often receives less attention than HA-venous thrombosis but may be more common. The International Medical Prevention Registry on Venous Thromboembolism (IMPROVE) risk assessment model (RAM) is the only externally validated RAM for predicting HA-bleeding in medical patients, but the validation was performed in a population designed to mimic the IMPROVE population.

Aims: To externally validate the IMPROVE bleeding RAM in unselected medical patients and by bleeding site.

Methods: Electronic health record data for medical admissions between 2010-2019 at the University of Vermont Medical Center (UVMMC) were assessed. Major and clinically-relevant non-major bleeding and IMPROVE risk factors were defined using validated computable phenotypes. Logistic regression RAMs including the IMPROVE risk factors and anticoagulation intensity at admission were fit for overall and site-specific bleeding. AUCs were calculated to assess model performance. The research received Institutional Review Board approval and was funded by the National Institutes of Health, USA.

Results: Table 1 presents the number of HA-bleeding events and the prevalence of the IMPROVE RAM risk factors in the IMPROVE and the UVMMC populations. At UVMMC there were 57,165 admissions among 29,110 people and 1,992 HA-bleeding events. Multivariable odds ratios tended to be smaller in the UVMMC population. The overall bleeding model had an AUC of 0.66 (Table 2). GI had the highest AUC (0.78) and Other Sites had the lowest AUC (0.63). When applying the points-based system proposed by IMPROVE, the observed bleeding risk for a score < 7 versus >= 7 was 3.2% and 5.5%.

Conclusion(s): The IMPROVE HA-bleeding score and the risk factors in the RAM are associated with bleeding. However, the risk score cut-off poorly discriminated risk of bleeding clinically, a potential limitation to the clinical applicability of this RAM. The different performance by bleeding site suggests further refinement of RAMs by bleeding site is warranted.