Cantu and Karlstadt question our conclusions because we excluded patients "at increased risk for gastrointestinal bleeding." Our aim was to determine whether prophylaxis with sucralfate or cimetidine was more effective than placebo in the prevention of stress-related hemorrhage for patients admitted to a medical intensive care unit [1]. Thirty-six percent of our 877 patients were not eligible because gastrointestinal bleeding was detected before admission. An additional 12% of patients had a recent history of hemorrhage, which is a frequent cause for exclusion from prophylaxis studies [2, 3]. Patients at risk for gastrointestinal hemorrhage from causes other than stress-related erosions (for example, users of nonsteroidal anti-inflammatory agents) were excluded. Patients receiving sucralfate or inhibitors of acid secretion at the time of admission or those who had esophageal or gastric surgery in the previous year were excluded because they might have a lower risk for stress gastritis. The demographic and clinical characteristics of the 143 patients who were excluded because they were receiving one of the study medications at the time of admission or because informed consent was not provided were similar to those of patients entering the study. The exception was that the randomly assigned patients had significantly higher Acute Physiology and Chronic Health Evaluation (APACHE) II scores. Thus, we believe that we did not exclude those patients at increased risk for bleeding. Our results are not comparable to those of Karlstadt and colleagues [2] or Martin and colleagues [3] because different patient populations were studied. Most of the patients entered into these two studies had recently had either major surgery or multiple trauma. We excluded these patient categories.

We disagree that failure to show an effect of cimetidine was caused by methodologic confounders. Administration of an inappropriately low dosage of cimetidine is unlikely. We achieved gastric pH values of 4.0 or more with the initial dose of cimetidine in 79% of patients, and 88% of pH readings for those receiving cimetidine were 4.0 or more. These percentages are similar to those reported by Martin and colleagues [3]. Although fewer patients in the cimetidine group received enteral feeding, the use of enteral and parenteral nutrition among the three groups was not different. Ruiz-Santana and colleagues [4] showed that stress-induced gastroduodenal lesions were prevented by parenteral nutrition, suggesting that the protective effect of nutrition is independent of the route of administration. The incidence of sepsis during the entire hospital stay did not differ among the three groups. Because few patients entering the study had received omeprazole, the 24-hour washout period had no substantial effect.

The issues raised by Dr. Martin are problems for many clinical trials. When we designed the study, our review of the medical literature indicated that a 12% incidence of stress-related hemorrhage was a reasonable assumption. As in many randomized controlled trials, the initial assumptions were not realized because of advances in treatment or changes in the risk profile of the study cohort. Consequently, the study had reduced power to detect the hypothesized reduction in hemorrhage incidence. The criticism that we may have missed small differences in efficacy is valid. To exclude a 50% reduction in bleeding rates would require randomly assigning 814 patients per study group; this would require multiple centers and substantial financial support. Moreover, it is debatable whether smaller differences in efficacy are clinically important. In our control group, stress-related hemorrhage was not a major cause of disease and was not a cause of death. For this reason, we do not believe that a study to detect smaller differences in efficacy among competing therapies is warranted.