Two-by-Two Factorial Cancer Treatment Trials: Is Sufficient Attention Being Paid to Possible Interactions?

Abstract

Factorial 2 × 2 designs can be used to combine evaluation of two treatments in a single study. The standard analysis approach is based on a factorial analysis that evaluates each treatment by pooling data over the other treatment. This approach relies on the assumption that the effect of each treatment is not substantially affected by the other treatment. In many oncology settings, this no-interaction assumption cannot be adequately supported at the time the trial is designed. In this Commentary, we consider current practices for the design and analysis of factorial trials by performing a survey of factorial treatment trials published in the Journal of the National Cancer Institute, Journal of Clinical Oncology, and the New England Journal of Medicine (2007-2016). The protocol-specified sample size was derived based on the factorial (pooled) analysis in 96.7% of the 30 identified trials, and the factorial analysis was specified as the primary analysis in 90.0% of these identified trials. An interaction complicating study interpretation was reported in 16.7% of the trials. We provide recommendations for matching the trial analysis and design to the study goals to account for possible interaction and illustrate the recommendations on the data from several published trials.

Figure 1.

CALGB 40603 (15): 2 × 2 factorial design: 1:1:1:1 randomization between the four arms. *Standard neoadjuvant chemotherapy backbone: paclitaxel weekly for 12 weeks followed by dose-dense doxorubicin plus cyclophosphamide for four cycles.