optgs {OptGS} | R Documentation |
Finding optimal and balanced group-sequential designs
Description
optgs is used to find a one-sided multi-stage design that balances four optimality criteria for a RCT with normally distributed outcomes
Usage
optgs(delta0 = 0, delta1 = 1/3, J = 2, sigma = 1, sd.known = TRUE,
alpha = 0.05, power = 0.9, weights = c(0.95, 0, 0, 0.05),
initial = NULL)
Arguments
delta0 |
mean difference in treatment effect under the null hypothesis (default: 0) |
delta1 |
clinically relevant difference used to power the trial (default: 1/3) |
J |
number of stages in the trial (default: 2) |
sigma |
assumed standard deviation of treatment responses (default: 1) |
sd.known |
logical value indicating if sigma will be treated as known; if FALSE, a quantile substitution method will be used to modify the stopping boundaries (default TRUE) |
alpha |
one-sided type-I error rate required (default: 0.05) |
power |
power required (default: 0.9) |
weights |
vector of length 4 giving the weights put on the four optimality criteria (default: c(0.95,0,0,0.05)). See details for more information |
initial |
starting values for the Nelder-Mead algorithm if the user wishes to override the default (default: NULL). Initial values must be specified as a two-dimensional vector where both entries are between -0.5 and 0.5. |
Details
optgs uses the extended power-family of group-sequential tests, and searches for the values of the futility and efficacy shape parameters that optimise the specified weighting. A description of the extended power-family and optgs is provided in Wason (2012). The ‘weights’ argument corresponds to the weight put on: 1) the expected sample size at delta=delta0; 2) the expected sample size at delta=delta1; 3) the maximum expected sample size; 4) the maximum sample size (i.e. J*groupsize).
Value
groupsize |
the number of patients required per arm, per stage |
futility |
the futility boundaries for the design |
efficacy |
the efficacy boundaries for the design |
ess |
the expected sample size at the delta0; the expected sample size at the delta1; and the maximum expected sample size |
typeIerror |
the actual type-I error rate of the design |
power |
the actual power of the design |
References
Wason, J.M.S. OptGS: an R package for finding near-optimal group-sequential designs. Journal of Statistical Software, 66(2), 1-13. https://www.jstatsoft.org/v66/i02/
Examples
##Find a three-stage design that minimises the maximum expected sample size.
threestagedeltaminimax=optgs(J=3,weights=c(0,0,1,0))
plot(threestagedeltaminimax)