Master protocol (platform, basket and umbrella trials)

A conventional trial asks one question about one drug in one cancer and is dismantled when it answers it. A master protocol keeps the infrastructure (the sites, the consent, the central testing, the data systems, the statistics) and lets questions come and go inside it. Three shapes are common. A platform trial compares several experimental arms with one shared control and can add new arms and close failing ones while it runs; STAMPEDE in prostate cancer (2005), I-SPY 2 in breast cancer (2010), GBM AGILE in glioblastoma (2019) and Beat AML in leukaemia (2016) are platforms. A basket trial gives one drug to patients whose tumours share a molecular feature regardless of where the cancer started; NCI-MATCH, TAPUR, DART and MyPathway are baskets. An umbrella trial screens patients with one cancer and sends each to the arm that matches their tumour's biomarker; Lung-MAP, the National Lung Matrix Trial, FOCUS4 and plasmaMATCH are umbrellas. Many trials mix the shapes, and most use adaptive statistics: Bayesian response-adaptive randomisation in I-SPY 2 and GBM AGILE, multi-arm multi-stage stopping rules in STAMPEDE and FOCUS4.

What the designs have delivered is uneven and instructive. Platforms with a shared control and a survival endpoint have changed practice directly: STAMPEDE established docetaxel, abiraterone and prostate radiotherapy at first diagnosis of advanced prostate cancer; I-SPY 2 graduated pembrolizumab into the trial that won its approval in early triple-negative breast cancer. Baskets and umbrellas that matched single targeted drugs to single mutations mostly found low response rates outside strong drivers (fusions, BRAF V600, dMMR, HER2 amplification), a lesson repeated by NCI-MATCH, Lung-MAP's first sub-studies, the National Lung Matrix Trial and Pediatric MATCH, and now pursued with combinations in ComboMATCH. Individualised designs (I-PREDICT, WINTHER) showed that how well a regimen matches the tumour predicts outcome, and CUPISCO showed in a randomised comparison that molecularly guided therapy can beat chemotherapy in cancer of unknown primary.

The practical gains are large: one screening step serves many arms, so rare subgroups fill; a shared control halves the patients each new drug needs; sites keep one contract and one consent. The costs are real too: control arms drift as standard care changes over a decade, arms added late are compared with patients recruited early, sponsors must share governance with academic groups or charities, and regulators have had to write new rules (the FDA's guidance on master protocols in oncology was finalised in 2022). The open questions are whether platform results can carry approvals on their own, how to keep the control arm honest across eras, and whether national screening platforms such as DETERMINE and TARGET National can route patients into trials fast enough to matter.