Platform trials
Ten new drugs to test. Do you run ten trials, or one that keeps running?
Three kinds of master protocol
A master protocol runs many sub-studies under one roof.
Steer patients to the better drug
Adaptive: more new patients go to the drug doing better so far. Will more patients get the better drug?
The lab's rule is Bayesian (Thompson-style): treatments with a higher chance of being the best, given the data so far, get more patients.
The cost: this can cut power (module 3). A drift in the response rate over time can also raise false positives.
1. Scenario
2. Enroll patients
Option
Keeps enough control patients for a fair comparison.
Share a control group
Each person icon stands for 10 patients: drug or control. 80 patients per arm.
Arms come and go
Suppose the drug does nothing. Pick one and compare it only with control patients enrolled at the same time (dark orange).
Predict, then check
Real example 1: I-SPY 2
A breast cancer platform that ran from 2010 to 2022. An arm graduates when its chance of winning a later trial reaches 85%.
Source: Sokolova, Sokolov and Polson (2026) and the platform trial chapters of Clinical Trials in Practice.
Real example 2: STAMPEDE
A UK prostate cancer platform shared one control arm across 10 other arms, and abiraterone cut the death rate by about a third.
Sources: STAMPEDE final recruitment figures; abiraterone result, James et al. (2017), as cited in Clinical Trials in Practice.
Regulator's corner
The FDA's revised draft guidance on master protocols (June 2026) says some master protocols may:
"improve efficiency by using a shared control arm in evaluating multiple drugs"
FDA draft guidance, Master Protocols for Drug and Biological Product Development. Revised draft (June 2026). In the lab: fewer control patients per drug.
Takeaway
- One shared control group saves patients: 480 instead of 800 for five drugs.
- Steering patients to the better drug helps them, but one trial can steer the wrong way.
- Compare each drug with controls from the same time.