Repository files navigation

The goal ofgoldilocks is to implement the Goldilocks Bayesianadaptive design proposed by Broglio et al. (2014) for time-to-eventendpoint trials, both one- and two-arm, with an underlying piecewiseexponential hazard model.

The method can be used for a confirmatory trial to select a trial’ssample size based on accumulating data. During accrual, frequent samplesize selection analyses are made and predictive probabilities are usedto determine whether the current sample size is sufficient or whethercontinuing accrual would be futile. The algorithm explicitly accountsfor complete follow-up of all patients before the primary analysis isconducted. Final analysis tests include the log-rank test, Coxproportional hazards regression Wald test, and a Bayesian test thatcompares the absolute difference in cumulative incidence functions at afixed time point.

Broglio et al. (2014) refer to this as aGoldilocks trial design, asit is constantly asking the question, “Is the sample size too big, toosmall, or just right?”

Other software and R packages are available to implement this algorithm.However, when designing studies it is generally required that manythousands of trials are simulated to adequately characterize theoperating characteristics, e.g. type I error and power. Hence, acomputationally efficient and fast algorithm is helpful. Thegoldilocks package takes advantage of many tools to achieve this:

• Log-rank tests are implemented via code from thefastlogranktestpackage, which uses a lightweight C++ implementation

• Piecewise exponential simulation is implemented via thePWEALL package, whichuses a lightweight Fortran implementation

• Simulation of multiple trials can be performed in parallel using thepbmcapply package

Note: becausefastlogranktest is no longer available on CRAN, acopy of the C++ code and wrapper have been incorporated directly intothis package.

Broglio KR, Connor JT, Berry SM. Not too big, not too small: aGoldilocks approach to sample size selection.Journal ofBiopharmaceutical Statistics, 2014;24(3): 685–705.

You can install the development version ofgoldilocksGitHub with:

# install.packages("devtools")devtools::install_github("graemeleehickey/goldilocks")