The ground-breaking I-SPY2 trial of neoadjuvant treatment for locally advanced breast cancer established a new benchmark for efficiency of phase II clinical trials. Widely regarded as a pioneer of the ‘platform’ trial, I-SPY2’s success continues to be a major influence on the development of next-generation trial designs in oncology and beyond.
I-SPY2 breaks from the traditional randomized clinical trial model, employing an 'adaptive' clinical trial model designed to increase trial efficiency by minimizing the number of participants and time required to evaluated an experimental agent.
At consent, a new participant’s breast cancer is classified into one of 10 molecular subtypes.
Then, for each participant in the trial:
1) I-SPY2’s adaptive randomization engine assigns a participant to a study arm with the higher Bayesian prior probability of efficacy.
2) The is assessed at time of surgery. Primary endpoint is RCB 0 or pathologic complete response meaning the tumor has disappeared completely.
3) Based on the participant's tumor subtype, outcome (ie. MRI volume, pCR) and treatment received, the predictive probabilities of the agent in the various subtypes are updated in real time, and affects the next participant to enter the arm.
4) If predictive probabilities for an experimental agent reach a pre-determined level of efficacy in one or more molecular subtypes, it “graduates”. Alternatively, it may be removed from the trial for futility or after reaching a maximum number of participants. At any point new agents can enter the trial through a protocol amendment.
5) The participant's serial MRI measures, RCB scores and tumor subtype are used to update the prior probabilities of the randomization engine -- over time this refines the targeting of subsequent participants.