Improving the selection of patients for upper gastrointestinal endoscopy.

A retrospective study was undertaken to investigate how endoscopies yielding positive findings differ a priori from those yielding negative findings: and how those judged 'helpful' (in the sense of influencing management) differ a priori from those judged 'unhelpful'. A total of 483 patients undergoing endoscopy was sampled and a wide range of data abstracted, including 48 patient characteristics available to the gastroenterologist at the time of the decision to perform endoscopy. Sixty nine per cent of endoscopies were positive. Multivariate statistical analysis identified four variables which taken together were strongly predictive of a positive endoscopy. The resulting mathematical formula correctly predicted the outcome of 76% of endoscopies. Eighty two per cent of the endoscopies were retrospectively classified by the gastroenterologists as helpful. Six variables were strongly predictive of a helpful endoscopy. The corresponding formula correctly predicted the finding of 84% of endoscopies. Comparison of the two analyses shows that the two sets of predictions differ substantially. Thus it is important that decision tools should be based not on the crude distinction between positive and negative, but on the more useful distinction between helpful and unhelpful in influencing management.