Loading
The randomised trial is how oncology knows what works, and it is slow, expensive and enrols fewer than one in ten patients. The roadmap is the set of designs and tools that keep the rigour while cutting the time, cost and exclusions: platform trials that never close, blood-test endpoints, remote consent, real-world data used honestly, and doses chosen by evidence.
The randomised controlled trial, introduced to medicine in 1948, and the cooperative groups that ran thousands of them, are why most of what oncology does is evidence-based. But a phase 3 trial takes years and hundreds of millions of dollars, often answers a question the field has moved past, excludes the older and sicker patients who make up most of the disease, and is joined by fewer than one in ten adults with cancer.
The first generation of fixes changed the design: STAMPEDE, the longest-running platform trial, showed that arms can be added and dropped under one protocol; basket and umbrella trials (Pediatric MATCH) assigned drugs by mutation; tissue-agnostic approvals followed. The second changed operations: remote consent and telehealth visits, electronic symptom monitoring, broadened eligibility, diversity plans, and real-world data with a regulatory framework. The third is changing what trials measure and how doses are chosen: ctDNA residual disease as an endpoint that reads out in months (DYNAMIC, CIRCULATE-Japan, SERENA-6, IMvigor011), pre-surgery windows as the biomarker engine, and the FDA's Project Optimus (final guidance August 2024) requiring randomised dose comparison instead of the maximum tolerated dose.
Ahead are AI in trial operations, standing platform infrastructure per cancer, registry-embedded randomisation, and endpoints that weigh how people live alongside how long. The pace is set by enrolment, cost, exclusion, weak real-world data and the incentives that keep failures hidden.
The first randomised controlled trial in medicine (streptomycin, 1948) gave oncology its method, and the US cooperative groups, the EORTC and later national groups in Canada, the UK and Europe ran the trials that made chemotherapy, adjuvant therapy and combined-modality treatment evidence-based. ClinicalTrials.gov (2000) made registration public. The model worked; it also fixed the template of one question, one comparator, years of follow-up and hundreds of millions of dollars.
STAMPEDE, opened in 2005, showed that a multi-arm multi-stage platform can add and drop arms under one protocol and answer several questions for the price of one, establishing docetaxel and abiraterone in prostate cancer along the way. Basket and umbrella trials assigned drugs by mutation across cancers; Pediatric MATCH did it nationally for children. Tissue-agnostic approvals (2017, 2018) followed, and accelerated approval on surrogate endpoints became the norm for oncology, with confirmatory trials that were often late or never done.
Real-world evidence gained a regulatory framework; the pandemic forced remote consent, telehealth visits and home delivery of study drugs and the FDA wrote them into guidance (2024); weekly electronic symptom reporting became a trial tool as well as a care tool. ASCO and Friends of Cancer Research rewrote eligibility so that brain metastases, prior cancers, controlled HIV and modest organ dysfunction no longer excluded patients by default, and diversity action plans became a filing requirement. Federated real-world networks and curated clinico-genomic databases made external comparison possible, if not yet trusted.
Molecular residual disease in blood reads out in months rather than years and has now changed treatment in randomised trials: DYNAMIC, CIRCULATE-Japan, SERENA-6 and IMvigor011, the first ctDNA-guided approval. Pre-surgery windows turned pathological response into a fast biomarker engine (NICHE-2, PHERGain, KEYNOTE-522). The FDA's Project Optimus (final guidance August 2024) ended the maximum-tolerated-dose default for new oncology drugs by requiring randomised dose comparison; the harder task of re-optimising approved doses falls to public funders and trials such as PERSEPHONE. Central imaging reads and companion diagnostics are being standardised as trial infrastructure.
Language models that read the record and flag a matching trial at the moment a treatment is chosen, eligibility simulated against real-world data before a protocol is locked, AI-assisted central imaging reads to cut endpoint cost, target-trial emulation in real-world data to decide which randomised trials are worth running, validated real-world progression endpoints so pragmatic trials can use them, and digital twins as virtual controls where randomisation is unethical. Each has a pilot; none has a standard.
The proposal that would change the economics most is a perpetual platform trial in every major cancer, funded as infrastructure, with one ethics approval and one consent form across countries, arms added as drugs arrive and dropped as answers come in, and response-adaptive allocation that learns as it goes. Variants: a national platform every ctDNA-positive patient can join, a platform that assigns treatment by resistance mechanism, one umbrella for all rare cancers, a RECOVERY-style platform for cheap repurposed drugs, and a DRUP-style protocol for off-label generics. Registry-embedded randomisation answers everyday questions inside routine care.
Trials that report patient-reported side-effects as rigorously as efficacy, win-ratio endpoints that weigh survival, toxicity and quality of life together, tolerability defined as carefully as efficacy, a mandatory over-70s cohort with geriatric assessment in every pivotal trial, a pragmatic trial after approval in the patients the pivotal trial excluded, sponsor-funded sites in Africa, South Asia and Latin America, and participants paid for their time. An independent programme to validate surrogate endpoints setting by setting would tell everyone which shortcuts are safe.
Fewer than one in ten adults with cancer joins a trial, and trials close for lack of patients rather than lack of questions. A phase 3 costs hundreds of millions and takes years. Older, poorer, rural and minority patients are under-represented, so results do not transfer. Real-world data are too weak to fill the gap. Negative results and abandoned programmes are rarely published, so mistakes repeat; a reversal registry and mandatory disclosure of top-line data are the proposed fixes. Regulators still review the same dossier separately in each region.
Every era's records, trial outcomes and papers, and every watch item, as JSON.
Probability ranges are named estimates that the claim is borne out on roughly a five-year horizon. They are meant to be argued with: propose a revision with your name and reasoning via a pull request to src/data/confidence.ts.
Loading this step…
Loading this step…
Loading this step…
Loading this step…
Loading this step…
Loading this step…
Loading this step…
Loading this step…
Shares A RECOVERY-style permanent platform trial of cheap drugs added to cancer care, One ethics approval and one consent form for a platform trial across countries, A standing platform trial that assigns treatment by how the tumour escaped, Let the trial learn: response-adaptive allocation across many combination arms.
Shares Diversity action plans with consequences: unmet targets trigger post-approval requirements, Turn Project Orbis into a work-sharing review with one shared assessment report, Conditional approvals that lapse automatically if the confirmatory trial is late, A public registry of cancer treatments that were later shown not to work.
Shares Massive Bio, Flatiron Health and Foundation Medicine Clinico-Genomic Database, Trial Library, Trial matching inside the electronic record at the moment a treatment is chosen.
Shares Remote consent and tele-screening so the first trial visit is a video call, Pay oncologists for the time it takes to enrol a patient, AI trial matching & clinical decision support, Trials do not represent the people who get cancer.
Shares One ethics approval and one consent form for a platform trial across countries, Let the trial learn: response-adaptive allocation across many combination arms, A standing platform trial for every major cancer, funded as infrastructure, A perpetual platform trial in every major cancer, funded as infrastructure.
Shares Emulate the trial in real-world data first to decide which trials to run, Validate real-world progression endpoints so pragmatic trials can use them, AI-assisted central imaging reads to cut endpoint cost and variability, A short pre-surgery drug window as the default early test of new agents.
Shares Let the trial learn: response-adaptive allocation across many combination arms, An independent programme that validates surrogate endpoints, setting by setting, PHERGain, Pre-surgery platform trials that test combinations on pathological response in months.
Shares Pay oncologists for the time it takes to enrol a patient, Default-inclusive eligibility: sponsors must justify every exclusion criterion, An independent programme that validates surrogate endpoints, setting by setting, Older and multimorbid patients are excluded and undertreated.