Objective: Progression-free survival is an often-used endpoint in clinical trials comparing preoperative therapy and surgery-first therapy. Because the surgery date is always later in the preoperative arm than in the surgery-first arm, it is difficult to define progression-free survival optimally. We evaluated three progression-free survival definitions that used different methods to handle incomplete resection.
Methods: The three definitions specify the event date of incomplete resection (IR) as follows: 'IR = event' method, date of surgery; 'IR not event' method, date of radiological or clinical progression after incomplete resection; landmark method, landmark time. According to these definitions, the theoretical strengths and weaknesses of the three definitions are investigated. Three patterns of progression-free survival and overall survival were estimated using the data of the Japan Clinical Oncology Group studies.
Results: Theoretically, 'IR = event' inflates alpha error while 'IR not event' method and landmark method reduce the statistical power under the alternative hypothesis. In JCOG9907, hazard ratios for the three definitions were: 'IR = event', 0.80 (95% confidence interval, 0.59-1.07; P = 0.13); 'IR not event', 0.81 (95% confidence interval, 0.60-1.09; P = 0.16); landmark, 0.80 (95% confidence interval, 0.59-1.07; P = 0.15). No P value of any methods corresponded with the positive result for overall survival (P = 0.03). In the preoperative arms of the four studies, maximum differences in median and percentage of 1 year progression-free survival among the three definitions were 0-6.4 months and 1.2-5.2%.
Conclusions: Progression-free survival sometimes fails as a surrogate of overall survival, and differences among results obtained with various progression-free survival definitions can be large. Overall survival should be used as primary endpoint in studies evaluating preoperative therapy.