R/lrsim_pkTrtSel.R
lrsim_pkTrtSel.RdSimulates the oncology case study of Carreras, Gutjahr, and Brannath (2015). Two experimental regimens and a control are randomized in stage 1, one regimen is selected at an interim analysis, and the selected regimen and control continue in stage 2. Overall survival is tested using the follow-up-wise, patient-wise, conservative Dunnett, stage-2-only, and conventional three-arm procedures considered in the article.
lrsim_pkTrtSel(
medianSurvival = c(9, 7.5, 6),
meanExposure = c(300, 200),
exposureSD = 120,
corrSurvivalExposure = 0.5,
interimPatients = 100L,
interimFollowup = 3,
totalPatients = 410L,
finalAnalysisTime = 29,
selectionRule = c("exposure", "os", "perfect"),
maxNumberOfIterations = 1000L,
seed = 0L,
nthreads = 0L
)Median overall survival in months for the high-dose, low-dose, and control arms, in that order.
Mean AUC for the high-dose and low-dose arms.
Common standard deviation of AUC on its original scale.
Gaussian copula correlation between survival time and log exposure.
Number of patients in the stage-1 cohort. The interim
analysis occurs after the last of these patients has at least
interimFollowup months of follow-up. Accrual continues in all three arms
until that analysis. These additional recruits are excluded from interim
OS statistics but contribute exposure data to treatment selection. For
patient-wise testing, they belong to the pre-selection patient cohort.
Minimum follow-up in months at the interim analysis.
Total sample size across both stages.
Calendar time in months from first enrollment to the final analysis.
Regimen-selection rule: "exposure" selects high dose
when its observed mean AUC is at least 1.5 times the low-dose mean;
"os" selects the regimen with the better interim log-rank statistic;
"perfect" selects the regimen with the smaller final hazard-ratio
estimate across the two counterfactual adaptive continuations.
Number of simulated trials.
Random-number seed.
Number of simulation threads. Zero leaves the current RcppParallel setting unchanged.
An object of class lrsim_pkTrtSel. It contains selection
probabilities, selection bias in the hazard-ratio estimate, estimated
combination-test weights, rejection probabilities by method, and a
trial-level sumdata data frame.
Accrual follows the article's piecewise rates of 7 patients/month for the first 4 months, 15 patients/month for the next 4 months, and 22 patients/month thereafter. Stage 1 uses 1:1:1 allocation. Stage-2 patients accrued before treatment selection also use 1:1:1 allocation; subsequent patients use 1:1 allocation to control and the selected regimen. Exposure is lognormal, parameterized by the supplied arithmetic mean and standard deviation, and is joined to exponential survival through a Gaussian copula. Combination weights are estimated from the average simulated event counts as described in the article. The patient-wise procedure splits patients at the interim decision time, so all patients enrolled before selection, including overrun patients, contribute to its first-stage statistics.
Carreras M, Gutjahr G, Brannath W. Adaptive seamless designs with interim treatment selection: a case study in oncology. Statistics in Medicine. 2015;34:1317-1333. doi:10.1002/sim.6407 .
sim <- lrsim_pkTrtSel(maxNumberOfIterations = 20, seed = 314159,
nthreads = 1)
sim$probabilitySelectHigh
#> [1] 0.4
sim$byMethod$dunnett$rejectionProbability
#> [1] 0.7