|
Abstract:
|
The Peto analysis for 2-year carcinogenicity studies and Tarone's test for trend on censored survival data employ a linear contrast among the dose groups. Two inherent drawbacks are that (a) the choice of the contrast (i.e., of the dose group scores) is arbitrary and often the subject of debate, and (b) all commonly used contrasts assign a negative score to the low dose (in some cases also to the mid dose), so that low-dose (and perhaps mid-dose) tumors raise the p-value and hence serve as implicit evidence AGAINST the tumor-inducing capacity of the test compound. We propose an alternative method that avoids these drawbacks by employing isotonic regression instead of a contrast, so that no choice of scores is required. Results on real data will be presented, and hopefully simulation results will also be available, pending the efforts of the author to obtain access to SAS.
|
