Risk models fail to predict lower-GI bleeding outcomes
In cases of lower gastrointestinal bleeding (LGIB), albumin and hemoglobin levels are the best independent predictors of severe bleeding, according to investigators.
These findings came from a sobering look at LGIB risk-prediction models. While some models could predict specific outcomes with reasonable accuracy, none of the models demonstrated broad predictive power, reported Natalie Tapaskar, MD, of the department of medicine at the University of Chicago, and her colleagues.
LGIB requires intensive resource utilization and proves fatal in 5%-15% of patients, which means timely and appropriate interventions are essential, especially for those with severe bleeding.
“There are limited data on accurately predicting the risk of adverse outcomes for hospitalized patients with LGIB,” the investigators wrote in Gastrointestinal Endoscopy, “especially in comparison to patients with upper gastrointestinal bleeding (UGIB), where tools such as the Glasgow-Blatchford Bleeding Score have been validated to accurately predict important clinical outcomes.”
To assess existing risk models for LGIB, the investigators performed a prospective observational study involving 170 patients with LGIB who underwent colonoscopy during April 2016–September 2017 at the University of Chicago Medical Center. Data were collected through comprehensive medical record review.
The primary outcome was severe bleeding. This was defined by acute bleeding during the first 24 hours of admission that required a transfusion of 2 or more units of packed red blood cells, and/or caused a 20% or greater decrease in hematocrit; and/or recurrent bleeding 24 hours after clinical stability, involving rectal bleeding with an additional drop in hematocrit of 20% or more, and/or readmission for LGIB within 1 week of discharge. Secondary outcomes included blood transfusion requirements, in-hospital recurrent bleeding, length of stay, ICU admission, intervention (surgery, interventional radiology, endoscopy), and the comparative predictive ability of seven clinical risk stratification models: AIMS65, Charlson Comorbidity Index, Glasgow-Blatchford, NOBLADS, Oakland, Sengupta, and Strate. Area under the receiver operating characteristic curve (AUC) was used to compare model predictive power. Risk of adverse outcomes was calculated by univariable and multivariable logistic regression.
Results showed that median patient age was 70 years. Most of the patients (80%) were African American and slightly more than half were female (58%). These demographic factors were not predictive of severe bleeding, which occurred in about half of the cases (52%). Upon admission, patients with severe bleeding were more likely to have chronic renal failure (30% vs. 17%; P = .05), lower albumin (3.6 g/dL vs. 3.95 g/dL; P less than .0001), lower hemoglobin (8.6 g/dL vs. 11.1 g/dL; P = .0001), lower systolic blood pressure (118 mm Hg vs. 132 mm Hg; P = .01), and higher creatinine (1.3 mg/dL vs. 1 mg/dL; P = .04). After adjustment for confounding variables, the strongest independent predictors of severe bleeding were low albumin (odds ratio, 2.56 per 1-g/dL decrease; P = .02) and low hemoglobin (OR, 1.28 per 1-g/dL decrease; P = .0015).
