datasim_cont()
The function datasim_cont() enables data simulation of a
platform trial with continuous endpoint and an arbitrary number of
treatment arms entering at different time points.
Assumptions
- equal sample sizes across all treatment arms, different sample size for the control group (resulting from allocation ratio 1:1:…:1 in each period)
- block randomization is used, a factor to multiply the number of
active arms with in order to get the block size in each period can be
specified as input argument (
period_blocks, default=2)
Notation
| Paper | Software |
| \(N\) | n_total |
| \(K\) | num_arms |
| \(d\) | d |
| \(n\) | n_arm |
| \(\eta_0\) | mu0 |
| \(\theta\) | theta |
| \(\lambda\) | lambda |
| \(\sigma\) | sigma |
| \(N_p\) | N_peak |
Usage
Input
The user specifies the number of treatment arms in the trial, the sample size per treatment arm (assumed equal) and the timing of adding arms in terms of patients recruited to the trial so far.
-
num_armsNumber of treatment arms in the trial -
n_armSample size per arm (assumed equal) -
dVector with timings of adding new arms in terms of number of patients recruited to the trial so far (of lengthnum_arms) -
period_blocks- number to multiply the number of active arms with in order to get the block size per period (block size =period_blocks\(\cdot\) #active arms) -
mu0- response in the control arm. Default=0 -
theta- vector with treatment effects for each treatment arm (of lengthnum_arms) -
lambda- vector with strength of time trend in each arm (of lengthnum_arms+1, as time trend in the control is also allowed) -
sigma- residual variance -
trend- indicates the time trend pattern (“linear”, “stepwise” or “inv_u”) -
N_peak- point at which the inverted-u time trend switches direction in terms of overall sample size -
full- Boolean. Indicates whether the full dataset should be returned. Default=FALSE
Output
Per default (using full=FALSE), the function outputs a
dataframe with simulated trial data needed for the analysis. If the
parameter full is set to TRUE, the output is a
list containing an extended version of the dataframe (also including
lambdas and underlying responses) and all input parameters.
Examples
# Dataset with trial data only (default)
head(datasim_cont(num_arms = 3, n_arm = 100, d = c(0, 100, 250),
theta = rep(0.25, 3), lambda = rep(0.15, 4),
sigma = 1, trend = "linear")) j response treatment period
1 1 -0.3888938 1 1
2 2 0.1484187 0 1
3 3 -0.3865791 1 1
4 4 -0.3564704 0 1
5 5 0.1704230 1 1
6 6 1.0764260 0 1
# Full dataset
head(datasim_cont(num_arms = 3, n_arm = 100, d = c(0, 100, 250),
theta = rep(0.25, 3), lambda = rep(0.15, 4),
sigma = 1, trend = "linear", full = T)$Data) j response treatment period means lambda0 lambda1 lambda2 lambda3
1 1 -0.7948134 1 1 0.2500000000 0.15 0.15 0.15 0.15
2 2 1.5167820 0 1 0.0003006012 0.15 0.15 0.15 0.15
3 3 0.1309096 0 1 0.0006012024 0.15 0.15 0.15 0.15
4 4 0.7464175 1 1 0.2509018036 0.15 0.15 0.15 0.15
5 5 -0.2801921 0 1 0.0012024048 0.15 0.15 0.15 0.15
6 6 -0.7003466 1 1 0.2515030060 0.15 0.15 0.15 0.15