In their article on the prophylaxis of stress-related gastric hemorrhage, Ben-Menachem and coworkers [1] discuss several potential explanations for their finding of lack of efficacy of either cimetidine or sucralfate. Although alluded to in the discussion, the study's low statistical power deserves further elaboration.

The determination that 160 patients would be required in each treatment arm to provide 80% power to detect a 75% reduction in bleeding from a baseline rate of 12% is statistically correct [2]. A 75% reduction, however, may be overly optimistic for sample size calculations. Using the concepts of Goodman and Berlin [3] and basing calculations on an expected 75% reduction, it can be shown before the trial that any observed reduction in bleeding would have a predicted precision of about ±52%. In other words, even if the trial had found no (0%) reduction in bleeding, the study would not have statistically excluded up to a 52% reduction in bleeding. The use of a 75% reduction as the projected effect in sample size calculations severely diminished the study's ability to detect smaller effects that were still clinically relevant. Most clinicians would agree that if an agent reduced bleeding by 50%, or even by 30% to 40%, it would still be useful.

The unforeseen diminished rate of bleeding (6%) in the control group (compared with a pretrial expectation of 12%) also contributed to the low statistical power. Given 160 patients in each group, but only 6% bleeding in the control group, the study had only a 45% chance (power) of detecting the original assumption of a 75% reduction in bleeding. Stated more simply, the study's very wide CI (relative risk, 0.26 to 2.64) suggests that a wide range of clinically important reductions in bleeding was not excluded in the trial. Accordingly, it is difficult to draw any firm conclusion from this study.