Predicting Poor Outcome from Acute Upper Gastrointestinal Hemorrhage

Abstract & Commentary

By James E. McFeely, MD, Medical Director Critical Care Units, Alta Bates Summit Medical Center, Berkeley, CA, is Associate Editor for Critical Care Alert.

Dr. McFeely reports no financial relationship to this field of study.

Synopsis: In this study of male veterans with acute upper GI bleeding, those with an APACHE II score of 11 or more, esophageal varices, or stigmata of recent hemorrhage, or some combination of these, had a 6-fold higher likelihood of a poor outcome.

Source: Imperiale TF, et al. Arch Intern Med. 2007;167(12):1291-1296.

Upper gastrointestinal hemorrhage results in 250,000 hospitalizations and between 15,000 and 30,000 deaths per year in the United States. There is considerable variability in how patients with upper gastrointestinal hemorrhage are managed, as well as in resource use, both within and between various institutions. A variety of clinical prediction rules for risk stratification have been published; however, none have achieved widespread use. Reasons for poor implementation include limited clinical applicability, complexity of use and relatively weak validation sets underlying the developed rules themselves.

Investigators at three VA hospitals defined two a priori outcomes: a composite GI hemorrhage-specific variable including re-bleeding and need for surgery or an advanced technique to control hemorrhage (Outcome 1); and a more comprehensive outcome variable that includes Outcome 1 plus worsening of any additional co-morbidities (Outcome 2). A total of 391 patients were enrolled in the trial, 244 in the derivation set, 147 in the validation set. Demographic variables were typical for a VA population: 99% of the patients were men with a mean age of 63 years, and 35% of the patients were older than 70.

Eight percent of the patients experienced major re-bleeding; 22% experienced major re-bleeding or had worsening of a co-morbid condition. Three variables were identified as significant in multiple logistic regression analysis. These were stigmata of recent hemorrhage, APACHE II score greater than 11, or esophageal varices. Unstable co-morbidity at the time of hospital admission was also identified as a variable for the more broadly encompassing Outcome 2.

Only two of 138 patients (1.4%) who had none of these risk factors developed any adverse outcomes. Seven of 149 (4.6%) patients with one risk factor had significant re-bleeding.


The authors of the above study have identified a relatively straightforward clinical prediction rule that appears to work well for elderly white male VA patients. Patients who present with this typical VA profile and have none of the three identified risk factors are at relatively low risk for significant re-bleeding or development of worsening co-morbidities. This prediction rule may be useful in decision-making regarding length of stay and location of admission within the hospital. What is unclear from the study is whether this can be generalized to women or to different, non-VA-type populations.

Given that these rules seem easy to track and are relatively robust in describing the risks in this subset of patients, they can probably begin to be used when a typical VA-type patient comes to your hospital. Hopefully these prediction rules can be validated soon for other subsets of patients and implemented more broadly at that time.