This paper examines the amount of variation among centres and estimates the overall effect of therapy in a multi-centre cancer clinical trial with censored failure time data. To investigate the centre effects, the variation in the treatment effect must be taken into consideration in addition to the variation in the baseline risk. We treat centre effects as random ones and extend the penalized partial likelihood approach proposed by McGilchrist to estimate the treatment-by-centre interaction as well as the baseline risk. This method is applied to data from a superficial bladder cancer clinical trial investigating the efficacy of intravesical chemotherapy after transurethral resection. In this trial, although there exists some degree of centre variation, especially in the baseline risk, the treatment is effective in preventing recurrence among the participating centres. This result indicates that the treatment effect is generalizable to the target population.
|Number of pages||11|
|Journal||Statistics in Medicine|
|Publication status||Published - 1999 Aug 15|