ideasIdea
Let the trial learn: response-adaptive allocation across many combination arms
As results come in, the trial sends more new patients to the arms that are working and fewer to those that are not, so more people benefit and bad arms die faster.
Multi-armed bandit and Bayesian response-adaptive randomisation, used in I-SPY 2 and in the REMAP-CAP platform during COVID-19, allocate patients towards promising arms while maintaining control of false positives. With ten or more combination arms, adaptive allocation shortens time to identify winners and reduces the number of patients on futile arms. Regulators accept these designs with pre-specified simulation of operating characteristics.
Hypothesis
Adaptive allocation across ten combination arms identifies the best arm with 30% fewer patients than equal randomisation at the same error rates, in simulation and in a live platform.
Rationale
REMAP-CAP found effective COVID-19 treatments faster than fixed designs. Oncology platforms have been slower to adopt because of endpoint latency; ctDNA or pathological response as intermediate endpoints removes that obstacle.
What would test it
Implement in a neoadjuvant platform using pathological response as the adaptive endpoint; compare patients-to-decision with a fixed-randomisation shadow analysis.
Maturity
early clinical
Who has to act
research
Cost to try
Medium ($1M to $50M)
Years to first evidence
4
Bottlenecks it attacks
- Too many combinations to test · There are thousands of possible drug pairs and sequences. Trials can test a few dozen a year.
- Trial design, endpoints and cost · A phase 3 trial takes years and hundreds of millions of dollars, and often answers a question that has already moved on.