The common practice of comparing the survival of responders and nonresponders when reporting the results of cancer chemotherapy treatment is investigated. The usual method of comparing responders and nonresponders is biased in favor of responders, and these results are frequently misinterpreted as providing evidence that response prolongs survival, or that the treatment under study is effective. Two valid

T IS a common practice for investigators to present comparisons of survival for responders versus nonresponders when reporting the results of chemotherapy regimens for cancer patients with advanced measurable disease. In many papers, these results represent the major" statistically significant" finding of the study. Investigators often conclude that if responders survive significantly longer than nonresponders, then the effect of response is to prolong survival. They often further conclude that, when response" prolongs" survival, increasing the response rate will be a way to increase survival. Many of the papers that compare survival for responders versus nonresponders contain two serious errors. The first error is in the statistical method used for detecting a significant difference between the survival of the two groups. The standard methods are biased and can produce a" significant" difference when none in fact exists. The reasons for this are discussed later, and two correct methods are described. The second common error is that the demonstration of a significantly longer survival for responders is frequently misinterpreted to imply that response causes longer survival. This con-