Development and Validation of Risk Models for Hospital-Acquired Bleeding in Medical Inpatients: The Medical Inpatients Thrombosis and Hemostasis (MITH) Study.

Abstract

Background: Hospital-acquired (HA) bleeding in medical inpatients is a serious complication with limited tools to predict risk.

Objectives: To develop and validate risk assessment models (RAMs) for anatomic location-specific HA bleeding in medical inpatients using objective and routinely available risk factors.

Patients/methods: Patients aged ≥18 years admitted to medical hospital services from 6 health systems and 15 hospitals in the United States between 2016-2020 for at least 1 midnight and without bleeding at admission were eligible for inclusion. Two health systems (10 hospitals) formed the development cohort, and 4 systems (5 hospitals) constituted the validation cohort. Bayesian Least absolute shrinkage and selection operator was used to develop anatomic site-specific risk assessment models (RAMS).

Results: Among the development (153,707 admissions) and validation (137,460 admissions) cohorts, 5,999 (3.9%) and 3,282 (2.4%) HA bleeds occurred. RAMs had varied risk factors by anatomic bleeding location; for instance, older age was associated with genitourinary bleeding in men but not women. Different cancer types were associated with different anatomic bleeding locations such as genitourinary cancers associated with genitourinary bleeding and gynecologic bleeding. The RAMs demonstrated good predictive performance for most anatomical bleeding locations; the C-statistic was ≥0.67 for all anatomic location bleeding in the development and validation cohorts, and generally higher in the anatomic-location specific RAMs.

Conclusions: These RAMs provide anatomic location- and sex-specific HA bleeding risk stratification, addressing a critical gap in individualized risk assessment. Integration into clinical workflows could enhance patient safety and outcomes. Implementation studies are essential to maximize their clinical utility.