Investigators published details of a revised breast cancer trial design called I-SPY2.2. The I-SPY2 platform, known for testing neoadjuvant therapies in locally advanced breast cancer, has been updated into a SMART format sequential multiple assignment randomized trial. Under the new design, patients who are unlikely to achieve a pathologic complete response (pCR) with their initial assigned therapy are re-randomized to a second subtype-specific treatment. If a satisfactory response is still not predicted, standard rescue therapy then applies.

The revised protocol also uses a response-adaptive randomization algorithm. That means as data accrue, randomization probabilities shift in favor of treatment regimes that appear more effective in achieving pCR. The I-SPY2.2 design uses techniques like Thompson sampling to update the probabilities dynamically based on posterior evidence that certain regimes will be optimal. Analyses so far (simulations and internal empirical studies) show that this design improves within-trial regime-specific pCR rates compared to uniform randomization, without sacrificing the ability to recommend optimal treatment paths.

Why this matters

Breast cancer treatment increasingly moves toward tailoring therapy based on tumor subtype, biomarker status, and individual patient response. The I-SPY2.2 design builds on this trend by allowing trial participants to change course if early indicators suggest the initial treatment may not work well. This could reduce exposure to ineffective therapies and minimize unnecessary toxicity.

Adaptive designs like this also help make trials more efficient: fewer patients may be exposed to inferior arms; resources can be focused on promising regimens; therapeutic hypotheses can be tested faster. For patients, the hope is clearer information, potentially faster transitions to better therapies, and more personalized treatment journeys.

Things to watch

Real-world implementation of SMART designs can be challenging. The criteria for re-assigning therapies need to be rigorously validated so patients aren’t switched too early or too late. Biomarker and subtype assays must be reliable and timely.

Also, trials using adaptive randomization must have strong statistical oversight to ensure bias is minimized and the trial remains robust, especially for secondary endpoints, long-term outcomes, and safety.

Regulators will also observe how I-SPY2.2 handles endpoint definitions and subgroup analyses, since adaptive designs sometimes draw criticism over how well results generalize beyond the trial setting.