In tpq/rarsim: Response-Adaptive Randomization Simulation Station

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

Quick start

Welcome to the rarsim GitHub page! Let's get started.

library(devtools)
devtools::install_github("tpq/rarsim")
library(rarsim)

library(rarsim)

Response-adaptive randomization (RAR) is a novel approach to experimental design that has the potential to reduce the cost of expensive clinical trials. However, it may be susceptible to inflated Type 1 or Type 2 errors. Simulations can help ensure that the trial design will not introduce spurious results. The rarsim package is designed to make it easier to run simulations for RAR experiments.

To use this package, you must consider two things. First, how do you want to simulate your data? Second, how do you want to schedule your trial?

Simulate the data

By convention, this package uses a list of length $K$ to represent the $K$ arms of the experiment. Each element in the list is a vector of the $N_k$ rewards associated with that group. This structure is used for both the simulator and the scheduler objects. Below, we make a data pool that has 4 groups with 1000 patients in each group.

set.seed(3220)
pool <- list(
  rpois(1000, 10),
  rpois(1000, 20),
  rpois(1000, 10),
  rpois(1000, 5)
)

It is easy to create a simulator object from any data pool.

simulator <- simulator.start(pool = pool)

Alternatively, we make a simulator object in one step.

simulator <- simulator.start.from.unif(mins = c(1, 1, 1, 1), maxes = c(7, 7, 6, 6))

Schedule a trial

There are a few trial parameters that you must set. This includes the number of patients in each group at the start of the trial during the "burn-in" phase, the number of patients to allocate at each subsequent step in the trial, the total number of steps, the prior means and prior variances, and the sampling method used for response-adaptive allocation.

N.burn.in = 20
N.allocate = 15
N.trials = 10
K.arm = 4
prior.means = rep(4, K.arm)
prior.vars = rep(1, K.arm)
sampler = sampler.thompson

Now we can make the scheduler object.

scheduler <- scheduler.start(prior.means, prior.vars, N.burn.in = N.burn.in, sampler = sampler)

A trial is run for N.trials steps. In each step, the scheduler requests the patient allocations to the simulator, and the simulator returns the observed rewards for the simulated patients. In pratice, the "rewards"" are generated from an experiment.

for(trial in 1:N.trials){

  request <- getAllocation(scheduler)
  simulator <- simulator.draw(simulator, request)
  rewards <- getDraw(simulator)
  scheduler <- scheduler.update(scheduler, rewards, N.allocate = N.allocate)
}

This iterative procedure is wrapped by the run.trial function.

scheduler <- scheduler.start(prior.means, prior.vars, N.burn.in = N.burn.in, sampler = sampler)
simulator <- simulator.start.from.unif(mins = c(1, 1, 1, 1), maxes = c(7, 7, 6, 6))
scheduler <- run.trial(scheduler, simulator, N.trials = N.trials, N.allocate = N.allocate)