Twenty-Twenty Hindsight: An Adjuvant Breast Cancer Trial Through the Retrospectoscope

Successful conduct of phase III clinical trials can be challenging, particularly in the adjuvant breast cancer setting, where relatively good-prognosis disease can result in low outcome event rates, thus requiring large sample sizes with prolonged follow-up, even with the simplest study design. The report by Shulman et al that accompanies this editorial highlights some of the trials and tribulations of conducting a complex, large-scale, adjuvant breast cancer trial, the ultimate aim of which is to provide results that are relevant to usual clinical practice. The Cancer and Leukemia Group B (CALGB) 40101–Alliance phase III trial was initiated in 2002. The study used a factorial design and asked two questions: first, whether six cycles of adjuvant chemotherapy was superior to four cycles of adjuvant chemotherapy; and second, whether single-agent paclitaxel (T) was equivalent (noninferior) to doxorubicin-cyclophosphamide (AC) chemotherapy. Relapse-free survival (RFS) was the outcome measure for both questions. The original study design called for a sample size of 4,646 patients, allowing for a superiority analysis with an of .05 (twosided) and 91% power to detect a hazard ratio (HR) of 0.77 for superiority of six cycles of chemotherapy relative to four cycles (HR of 1.30 for four cycles relative to six cycles). The sample size also allowed for a noninferiority analysis with an of .05 (one-sided) and a noninferiority HR margin of 1.30 for the comparison of T versus AC (ie, if the upper 95% CI for HR contained only values 1.30, this was evidence that T was noninferior to AC). The results of the first of these two questions were reported in Journal of Clinical Oncology in 2012: six cycles of T or AC were not superior to four cycles of T or AC (HR, 1.12; 95% CI, 0.84 to 1.49; P .44). The second question concerning the noninferiority of single-agent T relative to AC chemotherapy is addressed in the report that accompanies this editorial. The authors concluded that T was not noninferior to AC (ie, T could likely be worse than AC) on the basis of an RFS HR of 1.26 with a one-sided upper 95% CI limit of 1.48. Although we agree with the interpretation of this result, several important methodologic issues emerged during the conduct of this trial that warrant further discussion. First, this was an extremely complex trial. The use of a factorial design to ask two questions can be efficient by allowing both questions to be addressed using a single data set. The CALGB had embraced factorial designs for their adjuvant breast cancer trials in the late 1990s (eg, in CALGB 9344, three dose levels of doxorubicin and the addition of T v no T were tested using a 3 2 factorial design). However, in a factorial design, there must be no interaction between the two treatment questions if the statistical efficiency of this design is to be realized. If there is a significant interaction (ie, if the relative effectiveness of T and AC vary according to whether four or six cycles of treatment are administered), the four arms of the study must be analyzed separately, with resultant reduction in power for each question. In this trial, the assumption was that there was no interaction between duration of treatment (four v six cycles) and type of treatment (AC v single-agent T). From a clinical perspective, one might have postulated that six cycles of T compared with four cycles of T results in a greater improvement in RFS than six cycles of AC compared with four cycles of AC, because the increased toxicity of AC might have resulted in a greater proportion of patients unable to tolerate six cycles of AC compared with six cycles of T. This would be an interaction between the two study questions and would necessitate an analysis of the separate arms; in other words, the comparison of the two AC arms combined, with the two T arms combined, to test the hypothesis that T is noninferior to AC, would not be valid. The issue of interaction was not addressed in either of the publications that arose from this study. If present, an interaction would have required modification of the analytic approach to take the interaction into account, and would have led to an increased sample size requirement to address the two specific questions. It is possible that the investigators could not test for an interaction because of changes that occurred to the protocol during the course of the trial, and even if they had tested for interaction, such a test would have been underpowered, so a nonsignificant result could not be interpreted. The use of a noninferiority design to address the question of whether T and AC adjuvant chemotherapy yielded similar results was a potentially appropriate choice in CALGB 40101. Noninferiority clinical trials are trials that aim to “demonstrate that the experimental treatment is not clinically worse than the standard of care by more than a prespecified boundary, termed as noninferiority margin (NIFm).” Such an approach is useful if the new therapy is of lower cost, easier to administer, or safer than the standard therapy. JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L VOLUME 32 NUMBER 22 AUGUST 1 2014

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.