[Clinical prediction of endoscopic signs in active or recent upper gastrointestinal bleeding].

Background And Objective: The aim of this study was to construct and validate a mathematical model, based on clinical and laboratory data, that could be useful in the emergency department (ED) to predict which patients with upper gastrointestinal bleeding (UGB) have an active or recent bleeding.

Patients And Method: During a period of 12 months, we included all consecutive cases of UGB that came to the ED of an urban hospital. These patients made up the primary or model obtaining series. During the 12 following months, we selected a sample of UGB patients who made up the secondary series. The mathematical predictive model was built using logistic regression analysis.

Results: 623 patients made up the primary series and 251 the secondary series. A score equal or higher than 4 indicated the best diagnosis accuracy (63.6%) with positive and negative predictive values of 65.1 and 40.0%, respectively. The discriminative power of the model was significant (p < 0.001) but it displayed little accuracy.

Conclusions: Shock, positive tilt test, increased uremia, previous hematemesis, to be an usual alcohol drinker, presence of sweating, and no previous treatment with antiulcer drugs were independent predictors for active or recent bleeding. The diagnostic accuracy of the model does not not allow its systematic use in clinical practice. However, it may be helpful for the triage of upper gastrointestinal bleeding in the ED.