Repeated assessment of results in clinical trials of cancer treatment

The problem of early assessment of results of clinical trials of cancer treatment is discussed, and the results of a computer simulation experiment presented. These show the extent to which repeated assessment of results before the end of a trial may invalidate the use of the usual significance tests. In typical practical situations, the use of yearly assessments and the acceptance as “significant” of a difference more than 1·96 times its standard error can increase two or threefold the probability of finding a “significant” difference where in fact no real difference exists between the treatments. This may be avoided if a more stringent criterion for significance is used in the early assessments. The results of a further set of simulation experiments are presented, where a difference in an early assessment is only considered significant if it is more than three times its standard error. This procedure enables results to be watched as a trial proceeds, but does not invalidate the use of a standard significance test to compare the final results. At the same time a small saving is achieved in the number of patients required.