Examples of approaches to interim analyses

During the course of a trial suppose the plan is to perform four analyses, three interim analyses and one final analysis. Suppose also that these analyses will be performed at approximately equal intervals, then three ways of spending this 0.05 level over the four analyses are presented in Table 9.14.

In Table 9.14, it can be seen that the more of the a that is spent in analyses one, two and three the less remains for the final (and primary) analysis. It is for this reason the most widely used approaches are O'Brien and Fleming and the Peto-Haybittle procedures, as they retain the large portion of a for the final analysis. Despite these proposed solutions for spending the a it should be stressed that there are many difficulties and assumptions associated in monitoring the accumulating results of a trial and, in particular, in stopping a trial as a consequence of the results at an interim analysis. Some of these issues are discussed below.

0 0

Post a comment