Preface

Training overview

In this course, we will present concepts, theory, software, and a Shiny interface. Mainly we will focus on designs that you might consider for time-to-event endpoints. In addition to classical approaches assuming a proportional hazards assumption, we will provide methods for designing under non-proportional hazards assumptions. While most studies still use a logrank test, we will also touch on some alternatives along with their potential advantages and disadvantages.

Disclaimer

All opinions expressed are those of the presenters and not Merck & Co., Inc., Rahway, NJ, USA.

Chapters and training sections

  • Background theory (30 minutes)

    • Extension to non-proportional hazards
    • Group sequential design asymptotic distribution
    • Spending function bounds

  • Proportional hazards applications with Shiny app (40 minutes)

    • Lachin and Foulkes method for sample size derivation
    • Design setup with exponential distribution
    • Design setup with cure model
    • Updating bounds at time of analysis
    • Event-based and calendar-based spending bounds
    • Exercise

  • Break (15 minutes)

  • Non-proportional hazards model with logrank test (60 minutes)

    • Piecewise model
    • Average hazard ratio
    • Statistical information and time
    • Introduction to gsdmvn, gsDesign2 and simtrial

  • Break (10 minutes)

  • Weighted logrank and combination tests (55 minutes)

    • Introduction to methods
    • Weighted logrank
    • MaxCombo
    • Exercise

Software and supporting materials

  • Useful directories in course repository at https://github.com/keaven/gsd-deming:
    • data/: contains design files for examples; also simulation results
    • vignettes/: reports produced by Shiny app to summarize designs
    • simulation/: R code and simulation data for the last part of the course

Installing R packages

If you choose to install R packages locally:

  • The gsDesign package (v3.2.1) is available at CRAN
install.packages("gsDesign")
  • For non-proportional hazards, the following 3 R packages would be useful to install
remotes::install_github("Merck/simtrial@87cd828")
remotes::install_github("Merck/gsDesign2@fc3a2d3")
remotes::install_github("Merck/gsdmvn@ef2bb74", upgrade = "never")

You will need reasonably recent versions of R and packages.