n.opt {bdpopt}R Documentation

Optimise A Simple Normal Model

Description

Find an approximation of the optimal sample size and corresponding expected utility for a simple phase III clinical trial model with a single, normally distributed response and a utility function of a fixed form.

Usage

n.opt(nu = 0, tau = 1, sigma = 1, alpha = 0.025,
gain.constant = 1, gain.function = function(X, mu) 0,
fixed.cost = 0, sample.cost = 0.005,
k = 1, n.min = 1, n.max = 50, n.step = 1,
n.iter = 10000, n.burn.in = 1000, n.adapt = 1000,
regression.type = "loess",
plot.results = TRUE, independent.SE = FALSE,
parallel = FALSE, path.to.package = NA)

Arguments

nu

The mean of the conjugate normal prior distribution for the unknown population mean.

tau

The standard deviation of the conjugate normal prior distribution for the unknown population mean.

sigma

The known population standard deviation for each individual response in the trial.

alpha

The significance level in the one-sided test used by the regulatory authority to decide upon marketing approval for the new treatment.

gain.constant

A constant utility gain received upon treatment approval. The total gain consists of the sum of gain.constant and gain.function.

gain.function

A variable utility gain obtained in addition to the constant utility gain upon treatment approval.

fixed.cost

The fixed cost of performing the trial.

sample.cost

The marginal cost per observation for the trial.

k

The number independent, parallel trials. Must be an integer greater than zero.

n.min

Lower limit for the one-dimensional grid for the sample size.

n.max

Upper limit for the one-dimensional grid for the sample size.

n.step

The step size of the grid for the sample size.

n.iter

The number of iterations in the JAGS MCMC simulation.

n.burn.in

The number of burn iterations prior to the JAGS MCMC simulation.

n.adapt

The number of adaptation iterations prior to the burn in and JAGS MCMC simulation.

regression.type

If set to "loess", the default value, then local polynomial regression will be used (via a call to fit.loess) to fit the grid simulation results. If set to "gpr", GPR regression will be used instead. For any other value, no regression is performed and the optimisation done will consist of a maximisation over the values corresponding to the grid points.

plot.results

Set to TRUE if a plot of the results of the simulation over the grid is to be constructed.

independent.SE

If TRUE, then the standard errors of the sample means used to estimate the expected utility will be computed under the assumption of i.i.d. sampling. If FALSE, the standard errors are instead computed using the coda::spectrum0.ar function.

parallel

Set to TRUE if the simulations over the grid should be done in parallel on a multi-core processor. The default value FALSE leads to single-core computations.

path.to.package

The search path to the installation directory of bdpopt. For the default value, the function will attempt to find the path using search.

Value

A list with components

ns

A numeric, atomic vector containing the sample size grid points.

eus

A numeric, atomic vector containing the sample means of the simulated expected utilities corresponding to the sample size grid points.

opt.arg

The optimal sample size found by maximising the estimated expected utility.

opt.eu

The estimated optimal utility corresponding to the optimal sample size found.

Author(s)

Sebastian Jobjörnsson jobjorns@chalmers.se

See Also

optimise.eu


[Package bdpopt version 1.0-1 Index]