Bayesian Sequential Monitoring Calculator
Designs interim monitoring for a trial using posterior-probability stopping boundaries: at each planned look, the trial may stop for efficacy (or futility) when the posterior probability that the treatment is favorable crosses a threshold. Frequentist operating characteristics are then estimated by Monte Carlo simulation.
How it works
You set the number of looks, their spacing, and a posterior-probability threshold for stopping. Using a conjugate model (Beta–Binomial for binary, Normal–Normal for continuous, a normal approximation on the log hazard ratio for survival), the tool computes the stopping boundary at each look on both the posterior-probability and the data scale, then simulates many trials to estimate the achieved type I error, power, and expected sample size. The reported operating characteristics can be used to iteratively tune the threshold toward a target type I error.
When to use it
- You want to monitor a trial at interim looks and stop early using a posterior-probability rule.
- You are working with a binary, continuous, or time-to-event endpoint and want simulated operating characteristics.
- You need to evaluate the achieved type I error for a chosen threshold and compare candidate thresholds.
Assumptions & limitations
- The posterior-probability threshold does not inherently control the frequentist type I error across repeated looks; the achieved type I error is estimated by simulation, and if a frequentist error target is required, candidate thresholds must be iteratively evaluated or calibrated in a separate design procedure.
- With a vague prior the boundary is approximately constant on the z-scale (not an O'Brien–Fleming shape); an O'Brien–Fleming-like profile requires a deliberate choice, not the default.
- The survival endpoint uses a normal approximation on the log hazard ratio, and continuous endpoints assume a known variance.
- Simulated operating characteristics describe expected behavior across trials under the modeled assumptions, not a guarantee for a single realized trial.
For the full methodology, derivation, and worked examples, read the complete guide.