Adaptive Randomization and the I-SPY 2 Trial Platform
Researchers interpreting results from studies using adaptive randomization must be aware of assumptions made during study.
I-SPY 2 is an adaptive clinical trial platform that supports testing of multiple neoadjuvant treatment regimens in different subtypes of breast cancer.1,2
The study enrolls women who are newly diagnosed with high-risk stage II or stage III breast cancer and divides them into 8 biomarker groups defined by hormone receptor status, HER2 status, and recurrence risk (high vs low) determined by the MammaPrint 70-gene assay.
The primary outcome of the study is having at least an 85% predicted Bayesian probability of success in a hypothetical, phase 3, randomized clinical trial of 300 patients with similar biomarker subtypes and pathologic complete response (pCR) as the primary outcome.
Within each biomarker group, patients are randomized 4:1 so that 80% will receive an experimental therapy and 20% will be assigned to a control therapy regimen that is tailored to their biomarker subtype. The experimental therapy the new patient will receive is influenced by the current results of the study.
While I-SPY 2 enrolls new patients and others finish chemotherapy and undergo surgery, the primary outcome–pCR–is routinely being calculated for all biomarker signatures to determine which investigational arms will remain open to accrual, and which meet criteria for success or futility.
These predicted probabilities of success are also incorporated into the adaptive randomization scheme so that new patients have a higher chance of being randomized to the experimental arm with the highest predicted probability (at time of measurement) of meeting the primary outcome in patients with the same biomarker signature. This determination is based, in part, on observed trial outcomes like pCR and other predictive data such as biomarker signature and change in tumor size by serial magnetic resonance imaging (MRI).
In an interview with Cancer Therapy Advisor, Giovanni Parmigiani, PhD, chair of the department of biostatistics and computational biology at Dana-Farber Cancer Institute in Boston, Massachusetts, discussed some of the potential benefits and applications of adaptive randomization in clinical trials. Professor Parmigiani co-authored an article published in The New England Journal of Medicine that accompanied 2 articles reporting results from I-SPY 2, which summarizes the potential benefits of adaptive randomization and the differences between Bayesian and frequentist trial designs.3
RELATED: Contralateral Mastectomy Rates Up in Patients With Breast Cancer
A major benefit to the I-SPY 2 study is the “combination of adaptive randomization and the openness of the study. The platform has the ability to add drugs into the trial and the ability to graduate drugs that show benefit while remaining open to accrual,” explained Professor Parmigiani. “Adaptivity increases efficiency in identifying the most promising treatment when several are available, limits putting patients on drugs that have no benefits, and focuses resources and sample sizes on beneficial treatment arms.”