Diagnostic emergency endoscopy in upper gastrointestinal bleeding--do we have any decision aids for patient selection?

The benefit of emergency endoscopy and therapeutic policies based on certain stigmata of bleeding has recently been demonstrated in patients with peptic ulcer hemorrhage. Applying a simple method of computer-aided diagnosis to a set of prospective data (n = 571) we investigated the question as to whether information on the history (28 variables) and clinical examination (8 variables) could be used to predict ulcer bleeding or certain stigmata of bleeding, with a view to deciding when to perform an emergency endoscopy in patients with upper gastrointestinal bleeding. The patients were assigned to either a high-risk group (probability greater than 0.50 for ulcer bleeding, arterial bleeding, etc.) or a low-risk group (p less than 0.50), and the prediction was compared with the actual findings at endoscopy. The results were disappointing, with an overall accuracy of 71% for the prediction of bleeding peptic ulcer and 71% for the prediction of a bleeding or non-bleeding visible vessel. Despite a relative risk of 2.8 for "bleeding ulcer" and 2.5 for "visible vessel" in the high-risk group, only 72% of all "bleeding ulcer" patients, and 69% of the "visible vessel" patients could be identified by the model. These results indicate that neither a bleeding ulcer nor stigmata of bleeding can be reliably predicted by the patient's history and clinical examination. Emergency endoscopy should therefore be performed in all patients with gastrointestinal bleeding.