DSBayes: Bayesian subgroup analysis in clinical trials

In DSBayes: Bayesian Subgroup Analysis in Clinical Trials

Bayesian subgroup analysis in clinical trials

Description

Calculate posterior modes and credible intervals of parameters of the Dixon-Simon model for subgroup analysis (with binary covariates) in clinical trials.

Usage

     DSBayes(obj, thetahat, C, lvector, control=list(), ...)

Arguments

obj	The object from a regression model, for example, linear regression or Cox proportional-hazards regression. If obj is specified, then thetahat and C should be set to NULL.
thetahat	A vector of regression coefficients without the intercept. If thetahat is specified, then C should be provided as well, and obj should be set at NULL.
C	A variance covariance matrix of regression. If C is given, then thetahat should also be provided, and obj should be set at NULL.
lvector	A vector or a matrix that denotes linear combination of the parameters for which posterior estimates are desired. Note that, the order of the lvector should be as follows: the first parameter should always be the treatment indicator, then a set of binary covariates, and then the interactions between the treatment with covariates. See *Examples*.
control	A list of control parameters. See *Details*.
...	Additional arguments.

Details

The control argument is a list that can supply any of the following components:

tol

A relative accuracy for numerical quadrature. Default is tol = 1.e-03.

epsilon

A small positive quantity to ensure proper posterior resulting from Jeffreys' prior. Default is epsilon = 0.005.

ci

Level of the credible interval. Default is ci = 0.95.

k

A constant value to determine the interval width for searching the Bayesian credible interval, from lower to upper for a maximum of the density function. Default value for k is, k = qnorm((6+ci)/7) = 2.45.

transform

= NULL, then no transformation is performed. If transform = "logit", which is at default, then logit transformation is applied for posterior density function to find the credibile interval, logit(x) = log(x/(1-x)).

print

= TRUE or FALSE, indicating whether or not we want to print control parameters and progress. Default is FALSE.

Author(s)

Ravi Varadhan <[e-mail]> and Wenliang Yao (maintainer) <[e-mail]>

References

Dixon D. and Simon R. (1991). Bayesian Subset Analysis. Biometrics, 47, 871-881

Examples

# ex1 - use given thetahat and C matrix, and set "obj=NULL".
# an example from the clinical trial reported by Fisher(1988)

thetahat  	<- c(-1.57,-0.52,-0.39,.68, 1.09, 0.68, 0.91)
names(thetahat) <- c("trt","Sex","Age","Stage","trt*sex","trt*age","trt*stage")
p <- length(thetahat)

C <- matrix(NA, p, p)
C[upper.tri(C, diag=TRUE)] <- c( .1502, .0141, .0505, .0198, .0042, .0506,
 .0389, -.0038, .0041, .0538, -.0361, -.0505, -.0042, .0039, .1037, -.0445,
 -.0042, -.0507, -.0041, -.0046, .1066,-.1209, .0037, -.0041, -.0536, -.0025,
 .0120, .1474)
C[lower.tri(C)] <- t(C)[lower.tri(t(C))]

# define lvector
trt 	<- rep(1,8)
cov 	<- as.data.frame(matrix(rep(0,24), ncol=3))
lmatrix<-as.matrix(cbind(trt,cov,rep(1:0,each=4),rep(rep(0:1,each=2),2), rep(0:1,4)))
dimnames(lmatrix)[[2]]<-c("trt","Sex","Age","Stage","trt*sex","trt*age","trt*stage")

lvector <- lmatrix[2,]    # for 1 subset  
#> lvector
#      trt       Sex       Age     Stage   trt*sex   trt*age trt*stage 
#        1         0         0         0         1         0         1 
# treatment effect for the subset of Female under 65 at stage C.
# in this case the reference group is Male, under 65 years, at stage B.

#lvector <- lmatrix       # for all 8 subsets

result <- DSBayes(NULL, thetahat, C, lvector)


################################################################################
# ex2 - use "obj" option, and set "thetahat=NULL" and "C=NULL" 
# To run ex2, you need to remove hashmark(#).

#data(simsolvd)
#simsolvd$event <- 1-simsolvd$censor
#obj <- glm(event~trt*(age+beat+lvef+cardratio+sodium),
#                 family = "binomial", data = simsolvd)
#
#para    <- as.data.frame(matrix(rep(rep(0,5),5), ncol=5))
#lmatrix <- as.matrix(cbind(rep(1,5),para[1:5,],diag(1,5)))
#dimnames(lmatrix)[[2]] <- c("trt","age","beat","lvef","cardratio","sodium",
#"trt*age","trt*beat","trt*lvef","trt*cardratio","trt*sodium")
  
#lvector   <- lmatrix[2,] 	    # for 1 subset 
#out <- DSBayes(obj, NULL, NULL, lvector)

DSBayes documentation built on Oct. 14, 2023, 5:06 p.m.