I-SPY 2

I-SPY 2 opened in 2010 as a Bayesian adaptive platform trial, sponsored by the non-profit Quantum Leap Healthcare Collaborative and led by Laura Esserman at UCSF with Don Berry as its statistician. Women with stage 2 or 3 breast cancer at high risk of recurrence (by hormone receptor, HER2 and MammaPrint status) receive standard paclitaxel then anthracycline chemotherapy, with or without an experimental drug, before surgery. The endpoint is pathological complete response, read within months rather than the years a survival endpoint takes. Randomisation is adaptive: as each arm's results accumulate within ten biomarker signatures, new patients are more likely to be assigned to the arm doing best for their subtype.

A drug graduates when the model predicts at least an 85 percent chance of success in a 300-patient phase 3 trial in a given signature; it is dropped for futility when that chance falls below 10 percent. Several arms run at once against a shared control, so each new drug needs far fewer patients than a stand-alone trial.

Graduates and what became of them: veliparib with carboplatin (triple-negative disease, published in the New England Journal of Medicine in 2016, which led to the BrighTNess phase 3 and confirmed the value of carboplatin); neratinib (HER2-positive, hormone receptor-negative disease, NEJM 2016); pembrolizumab (HER2-negative disease, reported at ASCO in 2017), which fed into KEYNOTE-522 and the approval of pembrolizumab for early triple-negative breast cancer; the AKT inhibitor MK-2206; trastuzumab emtansine with pertuzumab (HER2-positive disease, replacing paclitaxel and trastuzumab); durvalumab with olaparib (HER2-negative disease); and the PD-1 and LAG-3 antibody pair cemiplimab and fianlimab (HER2-negative disease, 2022). Other arms, including ganitumab, trebananib, ganetespib and talazoparib with irinotecan, did not graduate, which is also the point: the platform retires drugs early and cheaply.

More than two thousand women have taken part. The trial has also shown that long-term outcomes track pathological complete response across every subtype and treatment, that response-predictive subtypes defined by immune, DNA repair and hormone signatures sort patients better than receptor status alone, and that MRI plus biopsy during treatment can identify women who are responding early. Those findings became I-SPY 2.2, which changes treatment during the trial according to early response.