Comparing survival of responders and nonresponders after treatment: a potential source of confusion in interpreting cancer clinical trials.

The comparison of survival distributions between patients who respond to therapy and those who do not can present methodologic and interpretational difficulties. Since assignment of patients to the responder or nonresponder groups is not random, statistical procedures that test the equality of survival distributions only demonstrate association between response and survival, not cause and effect. This association may have no relevance to the efficacy of treatment. The assignment of patients to response categories also represents a methodologic problem. Variability in the definition of a nonresponder and the handling of early deaths can both lead to varying conclusions concerning survival. In spite of these problems, statistical comparisons of survival distributions of responders and nonresponders are reported in approximately 20% of phase II and phase III clinical trials. Descriptive statistics may be more useful than inferential statistics in this situation.