This shiny app shows Bayesian optimal group sequential design with time to event endpoint.
- To test superiority and non-inferiority with time-to-event data under a unified framework.
- It's optimal in minimizing the expected sample size under null hypothesis while controlling the Type I error.
- It allows predefined number and timing of interim analysis based on number of events.
- There is no distributional assumption imposed on the time-to-event data.
library(shiny)
runGitHub(repo = 'BayesianOptimalDesign', username = 'stat-li', ref = 'main')
Assume the survival functions between experimental arm and historical control (or reference control) are proportional at time
- Futility stopping boundary using a power function:
where
- Superiority stopping using the O'Brien-Fleming boundary:
-
Prior of
$\delta$ in single-arm design:$$\delta \sim \rm{Gamma}(a, b),$$ where$a$ is the shape parameter and$b$ is the rate parameter. In the simulations,$b$ is set to$\frac{2a}{1+\delta_1}$ and$\delta_1$ is the expected hazard ratio under alternative hypothesis. -
Prior of
$\theta=-\log(\delta)$ in two-arm RCT design:$$\theta \sim N(\theta_0, \sigma_0^2),$$ where$\theta_0=-0.5\log(\delta_1)$ .
This app will require user to specify the x-year survival probability of the historical control (or reference control), e.g.