Adjusting for Crossover in Analyses of Letrozole vs. Tamoxifen

San Antonio, TX— Findings from the BIG 1- 98 trial have been presented at prior conferences, and at the San Antonio Breast Cancer Symposium, Meredith Regan, ScD, assistant professor, Department of Medicine, Harvard Medical School, spoke on new data that show even after taking into account the effect of patients who decided to crossover, 5 years of treatment with letrozole (Femara) was still associated with significantly better overall survival (OS) than 5 years with tamoxifen (P <.05). Explaining the rationale for conducting the analysis, Dr Regan, who is group statistician for the International Breast Cancer Study Group, which is conducting the trial, said, “We have an obligation to present updated result accounting for women who crossed over to the other arm following report of positive results. In trials with selective crossover, an intent-to-treat analysis is no longer relevant for patient care.”

BIG 1-98, a randomized phase III double-blind, controlled trial, included 4922 postmenopausal women with hormone receptor—positive early breast cancer. They were randomly assigned to treatment with either the aromatase inhibitor letrozole or tamoxifen. When BIG 1-98 results were first presented in 2005, they demonstrated superior disease-free survival (DFS) and reduced disease recurrence risk for women who received letrozole for an average of 18 months. At that time, the tamoxifen-only treatment arm was unblinded, and 25% of women in that group selectively crossed-over to the letrozole arm.

After a median follow-up of 76 months, an intention-to-treat (ITT) analysis revealed significant benefits in DFS (HR, 0.88; P = .03) and time to distance recurrence (TDR; HR, 0.85; P = .05) for letrozole, but the OS benefit did not achieve statistical significance (HR, 0.87; P = .08).

While selective crossover is in the best interest of patients, it does complicate further trial analyses, Dr Regan said. “The trial becomes a hybrid of a randomized trial and an observational study,” she commented, adding that accurate assessment of outcomes requiring longer follow-up, such as OS, becomes difficult. With the ITT analysis, the effects of letrozole were attenuated by the crossovers, which moved the randomized trial more toward an observational study. As a result, Dr Regan said, statistical methods needed to be modified.

The solution lies in alternative modeling strategies, Dr Regan explained. In the case of BIG 1-98, using inverse probability of censoring weighted analysis (IPCW), an established modeling method for addressing bias due to treatment selection, weights the follow-up for the women who stay on tamoxifen so that they account not only for themselves but also for the censored follow-up of matched patients who cross over. The weighting adjusts for factors associated with OS and selective crossover, including baseline factors like age, nodal status, tumor grade, and time-varying performance status. It estimates outcomes, she said, that would have been observed had there been no selective crossover.

With IPCW and censoring follow-up at selective crossover, the OS HR for letrozole was 0.83 (P <.05), a statistically significant benefit. “The ITT analysis on updated data from BIG 1-98 is inaccurate and no longer relevant for patient care.” Dr Regan concluded, “Our best estimate is that 5 years of letrozole is significantly better than tamoxifen for DFS, OS and TDR.”

In an interview, Dr Regan emphasized, “We owe it to clinicians and to patients to learn as much as we can from trials. This method deals with selective crossover, and gives an estimate of the important longer-term outcome of overall survival. It gives clinicians good information for making decisions for patients.” SABCS Abstract 16.