Multiple comparisons in carcinogenesis study with right-censored survival data.

This paper considers the practical problem in animal carcinogenesis experiments where several treatment groups are compared with a control group in a one-way layout and the observed survival data are subject to random right-censorship. Proposed herein are multiple testing procedures based on two-sample weighted logrank statistics, each comparing an individual treatment with the control, for determining which treatments are more effective than the control. The associated p-value of claiming a certain treatment is more effective than the control is also discussed. A test-based confidence set for the scale changes between each treatment and the control is then obtained. The comparative results of a Monte Carlo error rate and power study for small sample sizes are presented. Finally, a numerical example involving renal carcinoma in mice demonstrates the feasibility of the proposed multiple testing procedures and test-based confidence set. Copyright 2000 John Wiley & Sons, Ltd.