Like with any randomized clinical trial, researchers interpreting results from studies using adaptive randomization have to be aware of assumptions made during study design, conduct, and analysis of results.

For adaptive randomization to work seamlessly with new enrollments, outcomes data must be measured carefully and reported accurately in real-time. Because past patient experiences influence randomization probabilities and therefore treatment assignments for future patients, it greatly benefits investigators and study participants to use short-term outcomes that are still clinically meaningful.

Continue Reading

In the case of I-SPY 2, the randomization probabilities are determined using baseline biomarker signature, change in tumor size at 3 interim MRI timepoints, and pCR at surgery. I-SPY 2 also assesses recurrence-free survival and overall survival as secondary objectives, but these outcomes would not be suitable for informing the adaptive randomization algorithm in an adjuvant setting for breast cancer.

In a traditional 1:1 randomized clinical trial with a large sample size, the demographic and clinical characteristics are typically evenly distributed between treatment arms. This is a particular advantage as one can expect this to hold true for unknown or unmeasured factors that could otherwise confound the analysis.

In I-SPY 2, randomization against the control occurs before the adaptive randomization algorithm assigns a patient to an investigational treatment, so the distribution of baseline factors may differ between experimental and control arms. While final analyses will be adjusted for these factors, readers interpreting results have to consider the potential for unmeasured factors that differ between groups and affect the outcome.

RELATED: Study Examines Costs of Breast Cancer Chemotherapy Regimens

As more targeted therapies move into phase 2 testing, adaptive randomization designs are appealing options to efficiently match new drugs with subsets of patients who will benefit the most. But investigators designing these trials will need to weigh potential assumptions and uncertainty when choosing biomarkers to categorize patients and outcomes that will determine the randomization probabilities for future participants.


  1. Rugo HS, Olopade OI, DeMichele A, et al. Adaptive randomization of veliparib-carboplatin treatment in breast cancer. 2016 Jul 7. N Engl J Med. doi: 10.1056/NEJMoa1513749
  2. Park JW, Liu MC, Yee D, et al. Adaptive randomization of neratinib in early breast cancer. 2016 Jul 7. N Engl J Med. doi: 10.1056/NEJMoa1513750
  3. Harrington D, Parmigiani G. I-SPY 2 – a glimpse of the future of phase 2 drug development? 2016 Jul 7. N Engl J Med. doi: 10.1056/NEJMp1602256