Das A et al. (2008) Artificial neural network as a predictive instrument in patients with acute nonvariceal upper gastrointestinal hemorrhage. Gastroenterology 134: 65–74

Most triage techniques used in patients with acute, nonvariceal, upper-gastrointestinal hemorrhage (UGIH) require urgent endoscopic evaluation, which might not always be possible. Das and colleagues evaluated the ability of an artificial neural network (ANN) to evaluate risk in patients with UGIH.

Data from 387 patients with acute, nonvariceal UGIH between January 1998 and June 1999 were prospectively collected. Data from half of these patients were used to train an ANN model of nonendoscopic triage, and the other half was used for internal validation of the model; external validation was then performed for 200 patients from a different center.

The best ANN model, which incorporated 27 input variables, had a sensitivity and specificity of 89% and 89%, respectively, for predicting the presence of major stigmata of recent hemorrhage, and 81% and 82%, respectively, for predicting the need for endoscopic therapy. In the external validation group, the sensitivity of the ANN model was >90% for predicting both main outcomes, but the specificity was decreased. Negative predictive values of the ANN model were superior to those of the clinical Rockall score in both validation groups, and were comparable to that of the complete Rockall score in the external validation group.

The authors conclude that nonendoscopic triage using the ANN model performed just as well as the endoscopic procedure, and should be considered for identifying patients with UGIH at low risk of adverse outcomes.