TL32: Randomization needs in oncology trials
*Olga M Kuznetsova, Merck & Co. 

Keywords: randomization in oncology, balance in covariates, within-center balance, minimization, modified Zelen's approach, hierarchical dynamic allocation

The purpose of this roundtable is the exchange of ideas related to complex randomization needs in oncology trials and experience implementing fixed and dynamic allocation to meet such needs. Randomized oncology trials are often of a small size and thus randomization cannot be counted upon to produce balance in known important predictors unless it enforces such balance. Examples where the treatment arms considerably differ in the distribution of strong predictors of the response are common. Such imbalance complicates the interpretation of the study results and might lessen their credibility. In cases where balance in several strong predictors is needed, dynamic randomization might be the only feasible solution for small studies. The influential role of the investigators who evaluate the response, manage concomitant treatment, are involved in the decision to discontinue the study treatment, and choose a post-discontinuation treatment is an argument for balancing the randomization within centers. In a study where a biomarker cut-off that predicts which patients are most likely to benefit from the treatment is to be established, one of the dynamic techniques that produce similar distributions of the continuous biomarker in the two treatments arms can be used. All these randomization needs might be combined in a single study. Experience using stratified and dynamic allocation (minimization, dynamic hierarchical balancing, modified Zelen’s approach with or without balance in other factors, balancing on a continuous covariate) will be discussed. Practical constraints as well as experience with regulatory interactions will be of great interest.