2010 Volume 13 Issue 1 Pages 88-104
Non-parametric tests, such as the log-rank test and Cox regression based on the proportional hazards model, have prevailed in cancer survival analysis for more than 30 years. With improvements in cancer therapy and longer-term follow-up, the limitations of the Cox model and the merits of the Boag parametric model have been recognized by some investigators. The Boag model supposes that a proportion of the patients in a group are cured of the cancer, and the failure time of the remaining incurable patients follows a log-normal distribution. We contrast the strengths of the Boag and Cox models from the clinician's perspective. The advantage of the Boag model is that its three parameters allow us to disentangle and evaluate two effects of treatment: cure, and delaying of death from cancer. The Cox model fails to distinguish between these two effects and it is liable to overestimate the effects of palliative treatment when follow-up is immature. Also, the diagnostics to check the proportional hazards assumption are unreliable. Furthermore, non-parametric tests which inherently lack information on survival length are inadequate for measuring survival benefit, since survival benefit generally increases with survival length. However, the Boag model suffers from the disadvantages that the parameter estimation does not always converge (depending on the initial estimate), that the cause of death needs to be determined, and that longer follow-up is required. To confirm the validity of a model, clinicians need life-long follow-up of their patients.
