Current issues of ACP Journal Club are published in Annals of Internal Medicine


Risk factors predicted death after upper gastrointestinal hemorrhage

ACP J Club. 1996 Sept-Oct;125:49. doi:10.7326/ACPJC-1996-125-2-049

Source Citation

Rockall TA, Logan RF, Devlin HB, Northfield TC, and the steering committee and members of the National Audit of Acute Gastrointestinal Haemorrhage. Risk assessment after acute upper gastrointestinal haemorrhage. Gut. 1996 Mar;38:316-21. [PubMed ID: 8675081]



To determine the risk factors for mortality after acute upper gastrointestinal (GI) hemorrhage and to develop a scoring system using these risk factors to categorize patients.


Cohort analytic study.


4 UK health regions.


Consecutive adults who presented with acute upper GI hemorrhage were identified from hospital records. Data from 4185 patients were used to develop a prediction model; data from 1625 patients were used to test it.

Assessment of risk factors

Hospital records were used to obtain patient characteristics (age and sex) and history and physical data (shock, use of nonsteroidal anti-inflammatory agents and anticoagulants, diagnosis, comorbid conditions, hemoglobin levels, presence of stigmata of previous hemorrhage, and further hemorrhage).

Main outcome measures

Death and recurrent hemorrhage.

Main results

14% of the patients died, and 18% had recurrent hemorrhage. Multiple logistic regression showed that mortality was independently associated with age, shock, comorbid conditions, diagnosis, stigmata of previous hemorrhage, and rebleeding (P < 0.001). These data were combined, and an 11-point scale was developed. The scoring system was validated using the second group of patients. Scores for age were 0 for < 60 years, 1 for 60 to 79 years, and 2 for ≥ 80 years. Scores for shock were 0 for no shock, 1 for tachycardia, and 2 for hypotension. Scores for comorbid conditions were 0 for no major comorbid conditions; 2 for cardiac failure, ischemic heart disease, or other major comorbid conditions; or 3 for renal failure, liver failure, or disseminated malignancy. Scores for diagnosis were 0 for Mallory-Weiss tear or no lesion identified and no stigmata of recent hemorrhage; 1 for all other diagnoses; and 2 for malignancy of the upper GI tract. Scores for major stigmata of recent hemorrhage were 0 for none or dark spot only and 2 for blood in upper GI tract, adherent clot, or visible or spurting vessel. Patients with a score of 0, 1, or 2 (26%) had negligible mortality (0.03%) and a low incidence of recurrent hemorrhage. The model predicted mortality in each risk category with none of the differences between predicted and actual rates reaching statistical significance.


A risk score using the variables age, shock, comorbidity, diagnosis, stigmata of recent hemorrhage, and rebleeding predicted the risk for mortality in patients presenting with acute upper gastrointestinal hemorrhage.

Source of funding: Lederle Pharmaceuticals.

For article reprint: Mr. T.A. Rockall, The Surgical Epidemiology and Audit Unit, The Royal College of Surgeons of England, 35-43 Lincoln's Inn Fields, London WC2A 3PN, UK.


The vigorous diagnostic approach and the aggressive application of endoscopic hemostatic therapies have not substantially changed the overall outcome of upper GI hemorrhage, and mortality rates have stubbornly remained at 8% to 15%. This intransigence suggests that the problem has not been properly analyzed.

Rockall and colleagues provide a fresh approach by using problem-modeling techniques and specialized statistical tools to identify 6 interactive risk factors that can predict mortality from a larger group of upper GI hemorrhage risk factors that are selected on the basis of information availability and predictive success in single-variable studies. A weighted scoring system was developed from the 6 risk factors, and its predictive accuracy was validated in a separate study population.

Unfortunately, the inclusion of endoscopically identified stigmata of recent hemorrhage and endoscopic diagnosis in the problem model makes the scoring system unusable for patients for whom urgent endoscopy is unavailable or not done. Urgent upper GI endoscopy is expensive, and its availability is limited. Because the problem model was developed for purposes of risk standardization across a group of hospitals in the United Kingdom that were part of a national audit of upper GI hemorrhage, it necessarily embraced what was being done to evaluate upper GI hemorrhage rather than what might be done.

Refining the problem model is necessary to convert the scoring system from an audit tool to a clinical tool that can be widely implemented. Specifically, substitutes will need to be discovered for risk factors whose identification is dependent on the availability of urgent upper GI endoscopy. A potential substitute might be a quantitative gastric lavage, which appears to have prognostic value in single-variable analysis (1). Clinical implementation of the scoring system is unlikely to show much improvement over the less-formal endoscopy-based decision process that is currently in use.

James S. Barthel, MD
University of Missouri School of MedicineColumbia, Missouri, USA


1. McLaughlin WD, Kolts BE, Achem SR. Nasogastric lavage compared with outcome in 101 patients seen in an emergency room for upper gastrointestinal hemorrhage [Abstract]. Gastroenterology. 1987;92:1529.