plotB: Plot Bayes dose response curve and dose group means

Description

Plot a dose response curve fit by Bayes MCMC methods (with optional posterior interval bars). Also plot dose group means (with optional CI bars)

Usage

plotB(y, 
	dose, 
	parm, 
	sigma2,
	count=rep(1,length(y)),
	dgrid=sort(unique(c(seq(0,max(dose),length=50), dose))),
	predict= TRUE,plotDif=FALSE,plotMed=FALSE,
	plotResid=FALSE,clev=0.9,
	binary=c('no','logit','probit','BinRes'),BinResLev,
	BinResDir=c('>','

Value

Returns an object of class plotB. Three inputs are saved for later plotting: doses in the original design, dgrid, and clev. The following matrices are saved:

pairwise: The dose group means and their differences with placebo. If a baseline is supplied, the means are lsmeans adjusted to the mean baseline value.
modelABS: Model-based posterior mean, median, posterior (clev/2,1-clev/2) intervals for the population means and sample means. One row per dose group
modelABSG: Same as modelABS but computed on the input grid of doses.
modelDIF: Same as modelABS but with differences from placebo.
modelDIFG: Same as modelDIF but computed on the input grid of doses.

Arguments

y: Outcomes, which may be sample means (see counts). LSmeans from a saturated anacova model can be supplied, in which case it is assumed that the Bayesian dose response model also included the additive baseline covariates.
dose: Doses corresponding to outcomes
parm: Matrix of simultated parameter values (each row is a simulated parameter vector). The parm values must be constructed for use in the model function modFun. The default is a 4-parameter Emax model with parameters (log(ED50),lambda,Emax,E0). For a 3-parameter model, set lambda=1 for each simulated parameter vector.
sigma2: Simulated draws from the residual variance (assumed additive, homogeneous). The length of sigma2 must be the same as the number of rows of parm. Set sigma2 to all ones for binary data.
count: Sample sizes for means-only summarized data.
dgrid: The Bayes posterior summaries are evaluated and plotted on the dgrid dosing values
predict: If TRUE(default), the plotted intervals are predictive intervals for the dose group sample means.
plotDif: Plot difference between doses and placebo. It is assumed the lowest dose is placebo. If activeControl, the difference is with the active control mean, and the active controls are not plotted.
plotMed: If TRUE, model-based curves are medians rather than means.
plotResid: If TRUE, a plot of the residuals formed from the dose group means minus the posterior dose group means.
clev: Level for confidence and Bayes intervals
binary: If binary is 'logit' or 'probit', y is assumed to be binary and the appropriate backtransformation is applied to the Emax model output. If binary is 'BinRes', the continuous variable y is converted to a binary responder variable using BinResLev and BinResDir. The continuous Emax model output is converted to binary estimation and prediction assuming normally distributed residuals.
BinResLev: A cut level for a responder variable formed from a continuous endpoint. Rates are computed from the (continuous outcome) model parameters assuming normally distributed residuals. The input y variable is converted to a responder variable.
BinResDir: If BinResDir='>', the responder variable is 1 when y is greater than the cut level, otherwise, it is 1 when y is less than the cut level.
activeControl: When TRUE, active comparator data must be supplied. Each dose group (including PBO) are compared to the active comparator rather than PBO.
ac: Simulations from the posterior distribution of the mean response on active comparator. The number of simulations must match those for the dose response model. For binary data, the simulated values must be transformed to the proportion scale. This differs from the simulated model parameters.
yac: Outcomes for the active comparator group. The coding conventions for y are used.
countac: Sample sizes for summarized data corresponding to count.
labac: x-axis label for the active control group.
shapeac: Shape of the symbol for the active control group.
colac: Color of the symbol for the active control group.
symbol: An optional grouping variable for the dose group sample means.
symbolLabel: Label given to symbol in plot legend.
symbolShape: A character vector with names giving the shapes assigned to different levels of variable symbol. If a single shape is specified, it is replicated for all dose groups. See package ggplot2 for symbol mappings.
symbolColor: A character vector with names giving the colors assigned to different levels of variable symbol. If a single color is specified, it is replicated for all dose groups. See package ggplot2 for color mappings.
symbolSize: The size of the symbol for the dose group sample means. Set symbolSize=0 to supress plotting.
xlim: Plot limits for the x-axis
ylim: Plot limits for the y-axis
xat: The points at which tick-marks are to be drawn. Errors occur if the points are outside the range of xlim. By default (when NULL) tickmark locations are computed.
xlab: x-axis label
ylab: y-axis label
modelFun: The mean model function. The first argument is a scalar dose, and the second argument is a matrix of parameter values. The rows of the matrix are random draws of parameter vectors for the model. The default function is the 4-parameter Emax function emaxfun.
makePlot: If FALSE, create numerical output but no plot.
...: Parameters passed to generic plot function (not used)

Author

Neal Thomas

Details

A sample of parameters from the joint posterior distribution must be supplied (typically produced by BUGS). The Bayesian dose response curve is the Bayes posterior mean (or median) at each value on dgrid. The bar (interval) is the (clev/2,1-clev/2) Bayes posterior interval (which can differ from the Bayes HPD interval). The intervals are plotted only at the dose levels included in the study. Predictive intervals are formed by adding independent random draws from the sampling distributions of the dose group sample means to the population means.

The function generates random numbers when predict=TRUE, so the random number generator/seed must be set before the function is called for exact reproducibility.

References

Spiegelhalter, D., Thomas, A., Best, N., and Lunn, D. (2003), WinBUGS User Manual Version 1.4, Electronic version www.mrc-bsu.cam.ac.uk/bugs

Examples

Run this code

if (FALSE) {
data("metaData")
exdat<-metaData[metaData$taid==6 & metaData$poptype==1,]

prior<-emaxPrior.control(epmu=0,epsca=100,difTargetmu=0,difTargetsca=100,dTarget=80.0,
        p50=3.75,sigmalow=0.01,sigmaup=20)
mcmc<-mcmc.control(chains=3)

msSat<-sum((exdat$sampsize-1)*(exdat$sd)^2)/(sum(exdat$sampsize)-length(exdat$sampsize))
fitout<-fitEmaxB(exdat$rslt,exdat$dose,prior,modType=4,
				count=exdat$sampsize,msSat=msSat,mcmc=mcmc)
parms<-coef(fitout)[,1:4]  #use first intercept

outB<-plotB(exdat$rslt,exdat$dose,parms, sigma2=(sigma(fitout))^2,
			ylab="Change in EDD")
			
plot(outB,plotDif=TRUE)
}
# \dontshow{
data("metaData")
exdat<-metaData[metaData$taid==6 & metaData$poptype==1,]

parms<-matrix(c(1.611445,0.5160108,-77.73158,8.981834,
2.606269,0.3915947, -94.49052, 7.518533,
1.312385,0.9194756, -68.27381, 9.848746,
2.193071,0.5296447, -81.90235, 2.330520,
3.798612,0.3540451,-129.37529, 8.739098),ncol=4,byrow=TRUE)
sig2<-c(78.87238,87.05488,94.88178,78.54260,89.51093,79.74265)

outB<-plotB(exdat$rslt,exdat$dose,parms, sigma2=sig2,
			ylab="Change in EDD")
plot(outB,plotDif=TRUE)
# }

Run the code above in your browser using DataLab