Time until first significant difference in in vivo tumor growth experiments

In in vivo tumor growth experiments it is common to report the tumor measurement time at which the volume distributions of the treatment groups become significantly different. This method of analysis, as commonly practiced, is deficient in that its type I error rate exceeds the usual nominal rate of 5%, unless one specifically corrects for multiple comparisons. A second problem is that many investigators evidently interpret the time of first significance as a statistical parameter - ie., a fixed but unknown property of the model that one can estimate by experimentation. In fact the time until first significance, like the power of the test, depends both on true model parameters (such as mean growth curves and experimental variability) and on features of the experimental design, such as the sample size and the spacing of the measurement times. We argue that investigators would do better to compare treatment groups by modeling tumor growth curves or estimating volume doubling times.