Skip to contents

Select Experimental Treatments Using a Bayesian Rule

SelectExpUsingBayesianRule.Rd

This function implements the MAMS design for binary outcomes and performs treatment selection at the interim analysis (IA) using a Bayesian decision rule. At IA, an experimental treatment is selected for stage 2 if its posterior probability of exceeding a user-specified historical response rate (UserParam$dHistoricResponseRate) is greater than a user-defined threshold (UserParam$dMinPosteriorProbability): Pr(pj > UserParam$dHistoricResponseRate | data) > UserParam$dMinPosteriorProbability. If no treatment satisfies this criterion, the treatment with the highest posterior probability is selected. All experimental arms assume the same prior distribution: pj ~ Beta(UserParam$dPriorAlpha, UserParam$dPriorBeta). For stage 2, selected treatments are randomized against the control arm in a 2:1 ratio (experimental:control).

Usage

SelectExpUsingBayesianRule(
  SimData,
  DesignParam,
  LookInfo = NULL,
  UserParam = NULL
)

Arguments

SimData

Dataframe containing data generated in the current simulation.

DesignParam

List of design and simulation parameters required to perform treatment selection.

LookInfo

List containing design and simulation parameters, which might be required to perform treatment selection.

UserParam

A list of user-defined parameters in East or East Horizon. The default is NULL. The list must contain the following named elements:

UserParam$dPriorAlpha

A value (0,1) defining the prior alpha parameter of the beta distribution.

UserParam$dPriorBeta

A value (0,1) specifying the prior beta parameter of the beta distribution.

UserParam$dHistoricResponseRate

A value (0,1) specifying the historic response rate.

UserParam$dMinPosteriorProbability

A value (0,1) specifying the posterior probability needed to exceed the historic response rate for experimental treatment selection.

Value

A list containing:

TreatmentID

A vector of experimental treatment IDs selected to advance, e.g., 1, 2, ..., number of experimental treatments.

AllocRatio

A vector of allocation ratios for the selected treatments relative to control.

ErrorCode

An integer indicating success or error status:

ErrorCode = 0

No error.

ErrorCode > 0

Nonfatal error, current simulation aborted but subsequent simulations will run.

ErrorCode < 0

Fatal error, no further simulations attempted.

Note

  • The length of TreatmentID and AllocRatio must be the same.

  • The allocation ratio for control is always 1, and AllocRatio values are relative to this. For example, an allocation value of 2 means twice as many participants are randomized to the experimental treatment compared to control.

  • The order of AllocRatio should match TreatmentID, with corresponding elements assigned their respective allocation ratios.

  • The returned vector includes only TreatmentID values for experimental treatments. For example, TreatmentID = c(0, 1, 2) is invalid because control (0) should not be included.

  • At least one treatment and one allocation ratio must be returned.