Statistical considerations in drug trials of peptic ulcer.

Drug trials of peptic ulcer healing and recurrence must be carefully designed, conducted, and reported so that the conclusions drawn from them will be valid. Randomization is best achieved in multicenter studies by double-blind coding of drugs in blocks of six to eight for each participating center. The first step in determining sample size is to assess the magnitude of effects which are of clinical importance. For 4-week studies of healing comparing an active drug to a placebo, trials should consist of at least 40 subjects per group. Comparisons between two active drugs must be based on even larger samples to draw a conclusion that they are equivalent with any degree of certainty. During the study, preliminary analyses of the data should be avoided unless a statistical stopping rule with safeguards for maintaining the double-blind status is employed. When the results are reported, dropouts must be carefully accounted for and a confidence interval for differences in percentage healed or percentage recurred should be given. Predictors of outcome should be investigated and reported in detail.