Confidence intervals for the difference of median failure times applied to censored tumor growth delay data.

In tumor xenograft experiments, tumor response to the anti-cancer agents is often assessed by tumor growth delay, the difference in median tumor quadrupling times between treatment and control. Tumor quadrupling time is often subjected to right censoring because death of the experimental mice or limitations of the follow-up period. In the literature, tumor growth delay is simply reported without a standard error or confidence interval being given, which ignores the noise in the experimental data. Here, we present seven confidence intervals for the difference of medians in a general right-censored failure time framework. Simulation studies are conducted to compare the coverage probability of these intervals for small samples. The proposed intervals are applied to tumor growth delay data from an actual tumor xenograft experiment.