Comparison of parametric and non-parametric survival methods using simulated clinical data.

We derived three parametric survival models (the log-normal, log logit, and Weibull) from the clinical data of chemotherapy trials for stage II breast cancer. We then used these models to generate simulated survival data, which we analysed using both parametric (log-normal) and non-parametric (logrank, Gray-Tsiatis and Laska-Meisner) methods. With limited follow-up (5 years), the non-parametric tests had greater power than the log-normal model. This advantage diminished, however, with extended follow-up (15 years). Furthermore, only the log-normal model could distinguish reliably a survival advantage due to an increase in cured fraction from an advantage due to an increase in time to failure.