Abstract:
|
We examine the effect of unequal censoring on six hypothesis tests for comparing two survival curves: log-rank, Gehan-Breslow generalized Wilcoxon, Tarone-Ware, Peto-Peto, modified Peto-Peto tests and a combined test using log-rank and Tarone-Wars methods. In particular, we consider noninformative right censoring in terms of occurrence period of censoring, early censoring, middle censoring and late censoring. On the basis of three types of censoring, various scenarios are modeled by the sample size (20 and 50 per group) and the proportion of censoring (no censoring, 10% and 25%) for each type of censoring and then the size and power of the six tests are evaluated through simulation studies. The early censoring appeared to have little or no influence on the size and power of all six tests. All tests except the log-rank test showed the lowest relative power in case of the middle censoring, while the log-rank test achieved the worst power when late censoring occurred.
|