/vera-master-trial-designing
Designs and simulates basket, umbrella, and platform master protocols. Public-scope baselines: no-borrowing/complete-pooling basket, MAMS umbrella, concurrent-control platform. Built by Vera.
A biostatistician’s toolkit for designing and simulating master-protocol clinical trials across the three families: basket, umbrella, and platform. Public release implements one canonical method per family, so users learn the taxonomy and can compare any hand-tuned design against a standard baseline. Built by Vera; advanced borrowing and adaptation methods are intentionally out of scope (with references provided).
What it does
| Family | Public-scope method | Endpoint |
|---|---|---|
| Basket | No borrowing (independent Beta-Binomial) or complete pooling | Binary |
| Umbrella | Multi-arm multi-stage (MAMS) with stage-wise futility | Continuous or binary |
| Platform | Naïve concurrent-control comparison | Binary |
Monte Carlo simulation produces per-arm rejection rate, FWER under any-null configurations, 1-minimum power across active arms, and (for umbrella) traditional separate-trial vs. MAMS expected N. Outputs CSV + PDF.
What it does not do
Listed in the skill’s Beyond This Skill section with primary references:
- Basket information-borrowing (Simon’s Bayesian, Chen confirmatory, Wathen S-TI/SEP, Calibrated BHM, full BHM with Gibbs)
- Umbrella adaptive selection (drop-the-losers, Bayesian adaptive randomization)
- Platform NCC adjustment (regression, time-machine) and RAR (Thompson sampling)
- TTE and incidence-rate endpoints in master-protocol context
- HTA / reimbursement strategy
- External design-review checklists
Who it’s for
- Biostatisticians producing a first-pass operating-characteristics table for a master-protocol concept paper or early protocol draft.
- Clinical scientists sanity-checking a CRO’s basket / umbrella / platform proposal at the baseline level.
- Postdocs and PhD / MS holders who want a clean reference implementation of the master-protocol taxonomy.
Tested
Yes — built by a working biostatistician for internal use, then trimmed to a public-scope subset before release. The full internal version is preserved on the v1.0-internal git tag. Smoke-tested with three end-to-end examples (basket, umbrella, platform); operating characteristics match expectations under both null and active-arm configurations.
Verdict
The skill teaches the master-protocol taxonomy through working code. The advanced methods that turn a basket / umbrella / platform protocol into a defensible regulatory submission — borrowing, adaptive randomization, NCC adjustment — stay with the biostatistician.