Get Optimal Design — get_opt

This function search for the optimal design for a study, returning the interim sample size, total sample size and other measurements of the design minimizing the expected sample size under H0: EN0. For any overall sample size N, the function only search for interim sample size ranging from 0.3N to 0.7N.

Usage

get_opt_des(
  n,
  sim_size,
  acc_time,
  cen_time,
  int_step,
  method,
  lambda_H0,
  lambda_H1,
  H1_type,
  HR1,
  HR2,
  change_time,
  tau = NULL,
  alpha,
  power = NULL
)

Arguments

n: Overall sample size for each group. This should maintain consistency in inputs.
sim_size: Number of simulations to perform.
acc_time: Accrual time for the study.
cen_time: Time until censoring occurs.
int_step: The difference in interim sample size between two searching steps. For example, int_step = 4 means increasing int_n by 4 (2 for each group) each time.
method: Design method to use. Options are: - "logrank": Two-stage log-rank test - "Simple": Simple RMST difference test - "Complex": RMST with sculpted critical region.
lambda_H0: Rate parameter(exponential) under the null hypothesis.
lambda_H1: Rate parameter(exponential) of experimental group under the alternative hypothesis.
H1_type: Distribution of H1: - "PH": Proportional hazards _ "NPH": Non-proportional hazards (piecewise exponential for experimental group)
HR1: For piecewise exponential only. The hazard ratio before change_time
HR2: For piecewise exponential only. The hazard ratio after change_time
change_time: For piecewise exponential only. The time when hazard ratio changes from HR1 to HR2
tau: Cut-off time for RMST at the final stage. Default is NULL.
alpha: Significance level for the test.
power: Desired power of the test. If NULL, returns the most powerful result.

Value

A dataframe containing the optimal interim sample size and other relevant design measurements.

Examples


set.seed(2024)
lambda_H1 <- 0.9
HR <- 1.7
lambda_H0 <- lambda_H1 * HR
sim_size <- 5000
r <- 60
cen_time <- 1
alpha <- 0.1
power <- 0.85
result_scu <- c()
# Give a searching range of total sample size N:
for (N in seq(from = 74, to = 76, 2))
{
    n <- ceiling(N / 2) 
    acc_time <- N/r
    opt_rmst <- get_opt_des(n = n, sim_size = sim_size, 
                acc_time = acc_time, cen_time = cen_time, int_step = 4, 
                lambda_H0 = lambda_H0, lambda_H1 = lambda_H1, H1_type = 'PH', 
                alpha = alpha, power = power, method = 'Complex') 
    res <- our_rmst[, c('PET0','PET1','alpha','power',
                        'PET','EN0','EN1','EN','interim_n')]
    res$N <- c(N)
    result_scu <- rbind(result_scu, res)
    saveRDS(result_scu, "tem_result_scu.rds")
}
#> Error in makeCluster(n_cores): could not find function "makeCluster"

# All result in result_scu