DoseFinding (version 0.6-3)

getUpdDesign: Calculate Bayes estimates and optimal design for next cohort

Description

Wrapper function that calculates Bayesian dose-response model parameter estimates, model probabilities and optimal design for next cohort of patients. Internally two main function are called (i) calcBayesEst to calculate the Bayes estimates and (ii) calcOptDesign to calculate the optimal design.

Usage

getUpdDesign(data = NULL, doses, n2, clinRel = NULL, models, prior, scal = NULL,
             meanInd = TRUE, sWeights, sDoses, method = c("Nelder-Mead", "nlminb", "solnp", "exact"),
             type = c("MED", "Dopt", "MED&Dopt"),  control = list())

Arguments

data

Data frame containing the dose and the response data. The code assumes the columns to be named "dose" and "resp". If equal to NULL, the design specified via sWeights is used.

doses

Numeric vector given the doses available for the adaption

n2

Numeric specifying the sample size of the next cohort

clinRel

Clinical relevance threshold

models

A list specifying the candidate models. The names of the list entries should be equal to the names of the model functions. See help page for MCPMod function.

prior

list specifying the parameters of the prior (see calcBayesEst for details)

scal

scale parameter for beta dose-response model

meanInd

indicator, whether posterior means (if meanInd = TRUE) or the posterior mode should be used for updating (see calcBayesEst for details).

sWeights

Allocation weights for the start design (used when data=NULL)

sDoses

Doses to be used in start design (these might be different from the doses available for adaption)

method

Algorithm used for calculating the optimal design (see also calcOptDesign for details).

type

Determines which type of design to calculate. "MED&Dopt" uses both optimality criteria with equal weight.

control

control list passed down to calcOptDesign

Details

The function uses a Bayesian updating rule to update parameter estimates and model probabilities. Then it calculates the optimal design for the next cohort of patients. When there is a dose-response model, for which no dose in the dose-range achieves clinical relevance, this model is removed from the model set (for calculation of the optimal design), and the model probabilities are recalculated to sum to 1. When there is no model with a clinically relevant dose, a balanced allocation to all doses available is used.

References

Dette, H., Bretz, F., Pepelyshev, A. and Pinheiro, J. C. (2008). Optimal Designs for Dose-Finding Studies. Journal of the American Statistical Association, 103, 1225-1237

Bornkamp, B., Bretz, F., Dette, H. and Pinheiro, J. C. (2011). Response-Adaptive Dose-Finding under model uncertainty, Annals of Applied Statistics, 5, 1611-1631

Examples

## examples
models <- list(emax = 0.5, linear = NULL, logistic = c(4.5, 0.8), betaMod = c(1.5,1.5))
prior <- list(S=3, a = 5*6, d = 4, m = c(0, 1.65), V = c(4.5,0,0,4.5))
dfdata <- NULL  
doses <- c(0,2,4,6,8)
startDes <- getUpdDesign(dfdata, doses, n2 = 100, clinRel = 0.4, models, prior, scal=10, 
                         meanInd = FALSE, sWeights = rep(1/4,4), sDoses = c(0,2,4,8))
dfdata <- genDFdata("emax", c(e0 = 0.2, eMax = 1, ed50 = 0.5),
                    startDes[,1], startDes[,2], sqrt(4.5))
getUpdDesign(dfdata, doses, n2 = 225, clinRel = 1.3, models, prior,
             scal=10, meanInd = FALSE, method = "solnp",
             sWeights = rep(1/4,4), sDoses = c(0,2,4,8), type="MED&Dopt")
## now with 9 doses (0:8)
getUpdDesign(dfdata, 0:8, n2 = 225, clinRel = 1.3, models, prior,
             scal=10, method = "solnp", meanInd = FALSE,
             sWeights = rep(1/4,4), sDoses = c(0,2,4,8), type="MED&Dopt")