When to stop the search for an elusive source of gastrointestinal bleeding

This analysis investigates the clinical parameters that should drive decisions about when to continue or stop the search for an elusive source of gastrointestinal bleeding.

The number of endoscopies necessary to find a source of bleeding was estimated using the geometric distribution. A threshold analysis was used to develop a stop rule for the search for a site of bleeding. Bayes' formula served to estimate changes in the probability of achieving a diagnosis associated with a series of consecutive endoscopic tests.

With decreasing probability of diagnostic success associated with an individual endoscopic procedure, such as P = 50%, 33%, or 25%, the mean (standard deviation [SD]) number of procedures needed to find the source of bleeding increases to 2 (1.41), 3 (2.45), or 4 (3.46), respectively. The threshold analysis suggests that work-up should be discontinued if the expected rise in diagnostic probability does not exceed the ratio of work-up cost to bleeding cost, that is, Δ P < work-up cost/bleeding cost. For instance, a 10-fold higher cost of bleeding than endoscopy would justify continued work-up if it can improve diagnostic probability by 10%. Bayesian analysis shows that after three negative tests the diagnostic probability drops below such a threshold.

The analysis suggests the following basic rules. The search for a site of gastrointestinal bleeding will take on average 2 procedures with a range of 1 - 4. The search should be continued as long as the diagnostic probability is expected to rise by more than 10 %, which is unlikely after three consecutive negative tests.