Basket Trial Design Calculator
Designs and evaluates a basket trial — one treatment tested across several diseases, conditions, or biomarker-defined subtypes (baskets) — and estimates the operating characteristics of each basket by Monte Carlo simulation. Supports independent analysis and two information-borrowing methods, BHM and EXNEX.
How it works
Each basket is analyzed with a Bayesian model, and a Go/No-Go decision is based on the posterior probability that the response rate exceeds a reference. You can analyze baskets independently, or borrow information across them with a Bayesian hierarchical model (BHM) or the more robust EXNEX mixture. Borrowing improves precision when baskets are similar but can bias an outlying basket when they are not; simulation estimates per-basket power and false-positive rates under different truth patterns.
When to use it
- You are testing a single treatment across multiple diseases or biomarker-defined subtypes and want per-basket decisions.
- You want to borrow strength across baskets while controlling the risk that a truly different basket is distorted.
- You need simulated per-basket operating characteristics under homogeneous and heterogeneous scenarios.
Assumptions & limitations
- Information borrowing is a bias–variance tradeoff: it helps when baskets are exchangeable but can inflate or deflate an individual basket's results under heterogeneity, so operating characteristics should be evaluated across both similar and dissimilar scenarios.
- Decisions use posterior probabilities; per-basket type I error is estimated by simulation. Consistent with FDA's 2026 draft master-protocols guidance, strong family-wise error control across the separate baskets is generally not required by default (each basket is a distinct clinical question).
- The exchangeability weight (EXNEX) and hierarchical settings should be pre-specified from clinical plausibility rather than tuned to the data.
- Simulated operating characteristics describe expected behavior under the modeled scenarios, not a guarantee for a realized trial.
For the full methodology, derivation, and worked examples, read the complete guide.