R/simulatetrials.R
In Gtmille2/seamlessTrial: What the Package Does (One Line, Title Case)
Defines functions
simulatetrials
simulateest
simulateNoCar
simNoCarOrig
simCarOrig
simCarOrigBS
simNoCarOrigBS
Documented in
simCarOrig
simCarOrigBS
simNoCarOrig
simNoCarOrigBS
simulateest
simulateNoCar
simulatetrials
#' Simulate trials function
#'
#' Simulates 2 look trial and calculates the overall type I error rate.
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 1000
#' @param design The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simulatetrials <- function(N1=200, N=500, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=1000, design = "Pocock",tau1 = 1,tau2 = 1)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- seq(1,n.looks)/n.looks
for (sim in seq(1,nsim))
{
N1 = N1orig*(n.trt+1)
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N1)
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt)
# table(treat)
#Simulating data for these treatment assignments
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat, covValues,inspection = look)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.Current(data,n.looks,look,z.v.prev=NULL)
z = z.v[1:look,1]
v = z.v[1:look,2]
z.v
best = z.v[1,3] #The best treatment was found in this function
# boundaries = get.boundaries(n.looks = look,v = v, k = c(n.trt,rep(1,look-1)), alpha.star.u, alpha.star.l) # Getting stopping boundaries at this point
#Generating new covariate values
look = 2
left = sum(data$treat %in% c(0,best))
N = N*2 - left
covValuesNew = genCovValues(p=c(0.5,0.5),N=N)
covValues = rbind(covValues, covValuesNew) # Combining new covariate values with old covariate values
if (design == "Pocock" ) treat = psd(covValues,p1=3/4,best = best,tr = treat, n.trt = 1) else treat = spbd(covValues = covValues, m = 4, best=best, tr = treat, n.trt = n.trt) # Assigning new treatment values
# print(TRUE)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2,treat,covValues,data,inspection = look) #Simulating the new patients data
#Calculating test statistics z & v for this data at the second look.
z.v = get.z.v.Current(data,n.looks,look,z.v)
z = z.v[1:look,1] # Retrieving the Z statistic
v = z.v[1:look,2] # Retrieving the V statistic
boundaries = get.boundaries(n.looks = look,v = v, k = c(n.trt,rep(1,look-1)),alpha.star.u = alpha.star.u, alpha.star.l = alpha.star.l) # Calculating stopping boundaries at this look
selected[sim] = best
if (z.v$z[look] > boundaries$upper[look]) reject[sim] = 1
}
reject
}
#' Testing other simulate trials function
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' simulate.trials <- function(n1=20, N1=200, N=200, n.trt=3, mean.s=rep(0,3), mean.t=rep(0,3), p1, sigma0=1, sigma=1, rho=0.0, nsim=1000, save.boundary, simonly=0)
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 1000
#' @param design The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simulateest = function(n1 = 20, N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL,p1, p2 , sigma0, sigma, rho, nsim, design = "Pocock",tau1,tau2,save.boundary)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(n1/N1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig*(n.trt+1)
N = Norig
# data = simulate.data(mean.s,mean.t,N,sigma0,sigma,rho)
look = 1
covValues = genCovValues(p = c(p1,p2),N=N1)
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt)
table(treat)
#Simulating data for these treatment assignments
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat, covValues,inspection = look)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.Current(data,n.looks,look,z.v.prev=NULL)
best = z.v[1,3] #The best treatment was found in this function
selected[sim] = best
z = z.v[1:look,1]
v = z.v[1:look,2]
look = 2
left = sum(data$treat %in% c(0,best))
N = N*2 - left
covValuesNew = genCovValues(p=c(p1,p2),N=N)
covValues = rbind(covValues, covValuesNew) # Combining new covariate values with old covariate values
if (design == "Pocock" ) treat = psd(covValues,p1=3/4,best = best,tr = treat, n.trt = 1) else treat = spbd(covValues = covValues, m = 4, best=best, tr = treat, n.trt = n.trt) # Assigning new treatment values
# print(TRUE)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2,treat,covValues,data,inspection = look) #Simulating the new patients data
#Calculating test statistics z & v for this data at the second look.
z.v = get.z.v.Current(data,n.looks,look,z.v)
z = z.v[1:look,1] # Retrieving the Z statistic
v = z.v[1:look,2] # Retrieving the V statistic
t1percent = min(99,round(100*v[1]/v[2]))
boundary.value = sqrt(v[2])*save.boundary[t1percent]
if (z[2] > boundary.value) reject[sim] = 1
# # using short-term and long-term endpoints
# z.v = get.z.v(data,n1,N1,N)
# print(z.v)
# selected[sim] = seq(1,3)[z.v$v2>0]
# t1percent = min(99,round(100*z.v$v1[1]/z.v$v2[selected[sim]]))
# boundary.value <- sqrt(z.v$v2[selected[sim]])*save.boundary[t1percent]
# if (z.v$z2[selected[sim]]>boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simulateNoCar = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,save.boundary)
{
Norig =N
N1orig = N1
# set.seed(10101)
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
# trialprogress <<- seq(1,n.looks)/n.looks
trialprogress <<- c(n1/N1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig*(n.trt+1)
N = Norig
# data = simulate.data(mean.s,mean.t,N,sigma0,sigma,rho)
look = 1
covValues = genCovValues(p = c(p1,p2),N=N1)
#Getting treatment value assignments
treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
# table(treat)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.Current(data,n.looks,look,z.v.prev=NULL)
# z.v = get.z.v.simulate(data, n.looks = 2, look = 1,n1 = 20, N1 = 100, N = 200)
best = z.v[1,3] #The best treatment was found in this function
selected[sim] = best
z = z.v[1:look,1]
v = z.v[1:look,2]
look = 2
left = sum(data$treat %in% c(0,best))
N = N*2 - left
covValuesNew = genCovValues(p=c(p1,p2),N=N)
covValues = rbind(covValues, covValuesNew) # Combining new covariate values with old covariate values
treat = norandom(covValues = covValues, best = best , tr = treat, n.trt = 1)
# print(TRUE)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2,treat,covValues,data,inspection = look) #Simulating the new patients data
#Calculating test statistics z & v for this data at the second look.
z.v = get.z.v.Current(data,n.looks,look,z.v.prev = z.v)
z = z.v[1:look,1] # Retrieving the Z statistic
v = z.v[1:look,2] # Retrieving the V statistic
t1percent = min(99,round(100*v[1]/v[2]))
boundary.value = sqrt(v[look])*save.boundary[t1percent]
if (z[look] > boundary.value) reject[sim] = 1
# # using short-term and long-term endpoints
# z.v = get.z.v(data,n1,N1,N)
# print(z.v)
# selected[sim] = seq(1,3)[z.v$v2>0]
# t1percent = min(99,round(100*z.v$v1[1]/z.v$v2[selected[sim]]))
# boundary.value <- sqrt(z.v$v2[selected[sim]])*save.boundary[t1percent]
# if (z.v$z2[selected[sim]]>boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
# data.frame(reject = reject, selected=selected)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simNoCarOrig = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,save.boundary)
{
Norig =N
N1orig = N1
# set.seed(10101)
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.simulate(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = z.v[1,3] #The best treatment was found in this function
look = 2
z = z.v[1:look,1]
v = z.v[1:look,2]
t1percent = min(99,round(100*v[1]/v[2]))
boundary.value = sqrt(v[look])*save.boundary[t1percent]
if (z[look] > boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simCarOrig = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,design = "Pocock",save.boundary, block.size = 12)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt, block.size = block.size)
# treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.simulate2(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = seq(1,n.trt)[z.v[,2,1]!=0]
selected[sim] = best
look = 2
Bhat = z.v[best,look,3]
# Test 1: Using the bootstrap variance for Z2, V1, and V2
v1 = z.v[best,1,2]
v2 = z.v[best,2,2]
z2 = z.v[best,2,1]
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2 > boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simCarOrigBS = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,design = "Pocock",save.boundary, block.size = 20)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected =rep(0,nsim)
reject = rep(0,nsim)
rejectTest1 = rep(0,nsim)
rejectTest2 = rep(0,nsim)
rejectTest3 = rep(0,nsim)
rejectTest4 = rep(0,nsim)
rejectTest5 = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt, block.size = block.size)
# treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
z1.bs = rep(0,200)
v1.bs = rep(0,200)
z2.bs = rep(0,200)
b1.bs = rep(0,200)
b2.bs = rep(0,200)
for ( i in 1:200) {
N1 = N1orig
bs.s1 = sample(nrow(covValues), nrow(covValues), replace = TRUE)
covValuesbs = covValues[bs.s1,]
if (design == "Pocock" ) treat = psd(covValuesbs, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValuesbs, m = 4, best=0, tr = NULL, n.trt = n.trt, block.size = block.size)
#Calculating test statistics z & v for this data, and selecting the best treatment
databs = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValuesbs,inspection = look,data = NULL)
b.s1 = get.z.v.bootstrap(databs,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
bestcur = seq(1,n.trt)[b.s1[,2,1]!=0]
b1.bs[i] = b.s1[bestcur,1,3]
b2.bs[i] = b.s1[bestcur,2,3]
}
z.v = get.z.v.simulate2(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = seq(1,n.trt)[z.v[,2,1]!=0]
selected[sim] = best
look = 2
Bhat = z.v[best,look,3]
# Test 1: Using the bootstrap variance for Z2, V1, and V2
v1bootstrap = 1/var(b1.bs)
v2bootstrap = 1/var(b2.bs)
z2bootstrap = Bhat/(var(b2.bs))
t1percent = min(99,round(100*v1bootstrap/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest1[sim] = 1
# Test 2: Using the bootstrap variance for Z2 only
v2 = z.v[best, 2, 2]
v1 = z.v[best, 1, 2]
z2bootstrap = Bhat/sqrt(var(b2.bs))
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest2[sim] = 1
# Test 3: Using the bootstrap variance for Z2 and V2
v2bootstrap = 1/var(b2.bs)
v1 = z.v[best, 1, 2]
z2bootstrap = Bhat/var(b2.bs)
t1percent = min(99,round(100*v1/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest3[sim] = 1
#Test 4: Using square root of variance to calculate the statistic z2, v2, and v1
v1bootstrap = 1/sqrt(var(b1.bs))
v2bootstrap = 1/sqrt(var(b2.bs))
z2bootstrap = Bhat/sqrt(var(b2.bs))
t1percent = min(99,round(100*v1bootstrap/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest4[sim] = 1
#Test 5: Using square root of variance to calculate the statistic z2 only
v1bootstrap = 1/(var(b1.bs))
v2bootstrap = 1/(var(b2.bs))
z2bootstrap = Bhat/sqrt(var(b2.bs))
t1percent = min(99,round(100*v1bootstrap/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest5[sim] = 1
#Test6 Without boostrapping
v1 = z.v[best,1,2]
v2 = z.v[best,2,2]
z2 = z.v[best,2,1]
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2 > boundary.value) reject[sim] = 1
}
data.frame(rejectTest1 = rejectTest1,rejectTest2 = rejectTest2,rejectTest3 = rejectTest3,
rejectTest4 = rejectTest4, rejectTest5 = rejectTest5, selected = selected)
data.frame(powerTest1=sum(rejectTest1)/nsim,power3Test1=sum(rejectTest1[selected==3])/nsim,
powerTest2=sum(rejectTest2)/nsim,power3Test2=sum(rejectTest2[selected==3])/nsim,
powerTest3=sum(rejectTest3)/nsim,power3Test3=sum(rejectTest3[selected==3])/nsim,
powerTest4=sum(rejectTest4)/nsim,power3Test4=sum(rejectTest4[selected==3])/nsim,
powerTest5=sum(rejectTest5)/nsim,power3Test5=sum(rejectTest5[selected==3])/nsim,
powerTest6=sum(reject)/nsim,power3Test6=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simNoCarOrigBS = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,design = "Pocock",save.boundary, block.size = 12)
{
Norig =N
N1orig = N1
# set.seed(10101)
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
z1.bs = rep(0,200)
v1.bs = rep(0,200)
z2.bs = rep(0,200)
v2.bs = rep(0,200)
for ( i in 1:200) {
bs.s1 = sample(nrow(data), nrow(data), replace = TRUE)
databs = data[bs.s1,]
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.simulate(databs,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = z.v[1,3] #The best treatment was found in this function
# selected[sim] = best
look = 2
z = z.v[1:look,1]
v = z.v[1:look,2]
z1.bs[i] = z[1]
v1.bs[i] = v[1]
z2.bs[i] = z[2]
v2.bs[i] = v[2]
}
z.v = get.z.v.simulate(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = z.v[1,3] #The best treatment was found in this function
selected[sim] = best
look = 2
z = z.v[1:look,1]
v = z.v[1:look,2]
z2 = z[look]/sqrt(var(z2.bs))
v2 = v[look]/sqrt(var(v2.bs))
v1 = v[1]/sqrt(var(v1.bs))
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2 > boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
Gtmille2/seamlessTrial documentation built on Nov. 18, 2019, 5:16 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions simulatetrials simulateest simulateNoCar simNoCarOrig simCarOrig simCarOrigBS simNoCarOrigBS
Documented in simCarOrig simCarOrigBS simNoCarOrig simNoCarOrigBS simulateest simulateNoCar simulatetrials
#' Simulate trials function
#'
#' Simulates 2 look trial and calculates the overall type I error rate.
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 1000
#' @param design The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simulatetrials <- function(N1=200, N=500, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=1000, design = "Pocock",tau1 = 1,tau2 = 1)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- seq(1,n.looks)/n.looks
for (sim in seq(1,nsim))
{
N1 = N1orig*(n.trt+1)
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N1)
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt)
# table(treat)
#Simulating data for these treatment assignments
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat, covValues,inspection = look)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.Current(data,n.looks,look,z.v.prev=NULL)
z = z.v[1:look,1]
v = z.v[1:look,2]
z.v
best = z.v[1,3] #The best treatment was found in this function
# boundaries = get.boundaries(n.looks = look,v = v, k = c(n.trt,rep(1,look-1)), alpha.star.u, alpha.star.l) # Getting stopping boundaries at this point
#Generating new covariate values
look = 2
left = sum(data$treat %in% c(0,best))
N = N*2 - left
covValuesNew = genCovValues(p=c(0.5,0.5),N=N)
covValues = rbind(covValues, covValuesNew) # Combining new covariate values with old covariate values
if (design == "Pocock" ) treat = psd(covValues,p1=3/4,best = best,tr = treat, n.trt = 1) else treat = spbd(covValues = covValues, m = 4, best=best, tr = treat, n.trt = n.trt) # Assigning new treatment values
# print(TRUE)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2,treat,covValues,data,inspection = look) #Simulating the new patients data
#Calculating test statistics z & v for this data at the second look.
z.v = get.z.v.Current(data,n.looks,look,z.v)
z = z.v[1:look,1] # Retrieving the Z statistic
v = z.v[1:look,2] # Retrieving the V statistic
boundaries = get.boundaries(n.looks = look,v = v, k = c(n.trt,rep(1,look-1)),alpha.star.u = alpha.star.u, alpha.star.l = alpha.star.l) # Calculating stopping boundaries at this look
selected[sim] = best
if (z.v$z[look] > boundaries$upper[look]) reject[sim] = 1
}
reject
}
#' Testing other simulate trials function
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' simulate.trials <- function(n1=20, N1=200, N=200, n.trt=3, mean.s=rep(0,3), mean.t=rep(0,3), p1, sigma0=1, sigma=1, rho=0.0, nsim=1000, save.boundary, simonly=0)
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 1000
#' @param design The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simulateest = function(n1 = 20, N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL,p1, p2 , sigma0, sigma, rho, nsim, design = "Pocock",tau1,tau2,save.boundary)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(n1/N1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig*(n.trt+1)
N = Norig
# data = simulate.data(mean.s,mean.t,N,sigma0,sigma,rho)
look = 1
covValues = genCovValues(p = c(p1,p2),N=N1)
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt)
table(treat)
#Simulating data for these treatment assignments
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat, covValues,inspection = look)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.Current(data,n.looks,look,z.v.prev=NULL)
best = z.v[1,3] #The best treatment was found in this function
selected[sim] = best
z = z.v[1:look,1]
v = z.v[1:look,2]
look = 2
left = sum(data$treat %in% c(0,best))
N = N*2 - left
covValuesNew = genCovValues(p=c(p1,p2),N=N)
covValues = rbind(covValues, covValuesNew) # Combining new covariate values with old covariate values
if (design == "Pocock" ) treat = psd(covValues,p1=3/4,best = best,tr = treat, n.trt = 1) else treat = spbd(covValues = covValues, m = 4, best=best, tr = treat, n.trt = n.trt) # Assigning new treatment values
# print(TRUE)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2,treat,covValues,data,inspection = look) #Simulating the new patients data
#Calculating test statistics z & v for this data at the second look.
z.v = get.z.v.Current(data,n.looks,look,z.v)
z = z.v[1:look,1] # Retrieving the Z statistic
v = z.v[1:look,2] # Retrieving the V statistic
t1percent = min(99,round(100*v[1]/v[2]))
boundary.value = sqrt(v[2])*save.boundary[t1percent]
if (z[2] > boundary.value) reject[sim] = 1
# # using short-term and long-term endpoints
# z.v = get.z.v(data,n1,N1,N)
# print(z.v)
# selected[sim] = seq(1,3)[z.v$v2>0]
# t1percent = min(99,round(100*z.v$v1[1]/z.v$v2[selected[sim]]))
# boundary.value <- sqrt(z.v$v2[selected[sim]])*save.boundary[t1percent]
# if (z.v$z2[selected[sim]]>boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simulateNoCar = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,save.boundary)
{
Norig =N
N1orig = N1
# set.seed(10101)
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
# trialprogress <<- seq(1,n.looks)/n.looks
trialprogress <<- c(n1/N1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig*(n.trt+1)
N = Norig
# data = simulate.data(mean.s,mean.t,N,sigma0,sigma,rho)
look = 1
covValues = genCovValues(p = c(p1,p2),N=N1)
#Getting treatment value assignments
treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
# table(treat)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.Current(data,n.looks,look,z.v.prev=NULL)
# z.v = get.z.v.simulate(data, n.looks = 2, look = 1,n1 = 20, N1 = 100, N = 200)
best = z.v[1,3] #The best treatment was found in this function
selected[sim] = best
z = z.v[1:look,1]
v = z.v[1:look,2]
look = 2
left = sum(data$treat %in% c(0,best))
N = N*2 - left
covValuesNew = genCovValues(p=c(p1,p2),N=N)
covValues = rbind(covValues, covValuesNew) # Combining new covariate values with old covariate values
treat = norandom(covValues = covValues, best = best , tr = treat, n.trt = 1)
# print(TRUE)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2,treat,covValues,data,inspection = look) #Simulating the new patients data
#Calculating test statistics z & v for this data at the second look.
z.v = get.z.v.Current(data,n.looks,look,z.v.prev = z.v)
z = z.v[1:look,1] # Retrieving the Z statistic
v = z.v[1:look,2] # Retrieving the V statistic
t1percent = min(99,round(100*v[1]/v[2]))
boundary.value = sqrt(v[look])*save.boundary[t1percent]
if (z[look] > boundary.value) reject[sim] = 1
# # using short-term and long-term endpoints
# z.v = get.z.v(data,n1,N1,N)
# print(z.v)
# selected[sim] = seq(1,3)[z.v$v2>0]
# t1percent = min(99,round(100*z.v$v1[1]/z.v$v2[selected[sim]]))
# boundary.value <- sqrt(z.v$v2[selected[sim]])*save.boundary[t1percent]
# if (z.v$z2[selected[sim]]>boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
# data.frame(reject = reject, selected=selected)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simNoCarOrig = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,save.boundary)
{
Norig =N
N1orig = N1
# set.seed(10101)
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.simulate(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = z.v[1,3] #The best treatment was found in this function
look = 2
z = z.v[1:look,1]
v = z.v[1:look,2]
t1percent = min(99,round(100*v[1]/v[2]))
boundary.value = sqrt(v[look])*save.boundary[t1percent]
if (z[look] > boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simCarOrig = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,design = "Pocock",save.boundary, block.size = 12)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt, block.size = block.size)
# treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.simulate2(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = seq(1,n.trt)[z.v[,2,1]!=0]
selected[sim] = best
look = 2
Bhat = z.v[best,look,3]
# Test 1: Using the bootstrap variance for Z2, V1, and V2
v1 = z.v[best,1,2]
v2 = z.v[best,2,2]
z2 = z.v[best,2,1]
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2 > boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simCarOrigBS = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,design = "Pocock",save.boundary, block.size = 20)
{
Norig =N
N1orig = N1
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected =rep(0,nsim)
reject = rep(0,nsim)
rejectTest1 = rep(0,nsim)
rejectTest2 = rep(0,nsim)
rejectTest3 = rep(0,nsim)
rejectTest4 = rep(0,nsim)
rejectTest5 = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
if (design == "Pocock" ) treat = psd(covValues, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValues, m = 4, best=0, tr = NULL, n.trt = n.trt, block.size = block.size)
# treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
z1.bs = rep(0,200)
v1.bs = rep(0,200)
z2.bs = rep(0,200)
b1.bs = rep(0,200)
b2.bs = rep(0,200)
for ( i in 1:200) {
N1 = N1orig
bs.s1 = sample(nrow(covValues), nrow(covValues), replace = TRUE)
covValuesbs = covValues[bs.s1,]
if (design == "Pocock" ) treat = psd(covValuesbs, p1 = 3/4, best = 0, tr = NULL, n.trt = n.trt) else treat = spbd(covValues = covValuesbs, m = 4, best=0, tr = NULL, n.trt = n.trt, block.size = block.size)
#Calculating test statistics z & v for this data, and selecting the best treatment
databs = simulatedata.car(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValuesbs,inspection = look,data = NULL)
b.s1 = get.z.v.bootstrap(databs,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
bestcur = seq(1,n.trt)[b.s1[,2,1]!=0]
b1.bs[i] = b.s1[bestcur,1,3]
b2.bs[i] = b.s1[bestcur,2,3]
}
z.v = get.z.v.simulate2(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = seq(1,n.trt)[z.v[,2,1]!=0]
selected[sim] = best
look = 2
Bhat = z.v[best,look,3]
# Test 1: Using the bootstrap variance for Z2, V1, and V2
v1bootstrap = 1/var(b1.bs)
v2bootstrap = 1/var(b2.bs)
z2bootstrap = Bhat/(var(b2.bs))
t1percent = min(99,round(100*v1bootstrap/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest1[sim] = 1
# Test 2: Using the bootstrap variance for Z2 only
v2 = z.v[best, 2, 2]
v1 = z.v[best, 1, 2]
z2bootstrap = Bhat/sqrt(var(b2.bs))
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest2[sim] = 1
# Test 3: Using the bootstrap variance for Z2 and V2
v2bootstrap = 1/var(b2.bs)
v1 = z.v[best, 1, 2]
z2bootstrap = Bhat/var(b2.bs)
t1percent = min(99,round(100*v1/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest3[sim] = 1
#Test 4: Using square root of variance to calculate the statistic z2, v2, and v1
v1bootstrap = 1/sqrt(var(b1.bs))
v2bootstrap = 1/sqrt(var(b2.bs))
z2bootstrap = Bhat/sqrt(var(b2.bs))
t1percent = min(99,round(100*v1bootstrap/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest4[sim] = 1
#Test 5: Using square root of variance to calculate the statistic z2 only
v1bootstrap = 1/(var(b1.bs))
v2bootstrap = 1/(var(b2.bs))
z2bootstrap = Bhat/sqrt(var(b2.bs))
t1percent = min(99,round(100*v1bootstrap/v2bootstrap))
boundary.value = sqrt(v2bootstrap)*save.boundary[t1percent]
if (z2bootstrap > boundary.value) rejectTest5[sim] = 1
#Test6 Without boostrapping
v1 = z.v[best,1,2]
v2 = z.v[best,2,2]
z2 = z.v[best,2,1]
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2 > boundary.value) reject[sim] = 1
}
data.frame(rejectTest1 = rejectTest1,rejectTest2 = rejectTest2,rejectTest3 = rejectTest3,
rejectTest4 = rejectTest4, rejectTest5 = rejectTest5, selected = selected)
data.frame(powerTest1=sum(rejectTest1)/nsim,power3Test1=sum(rejectTest1[selected==3])/nsim,
powerTest2=sum(rejectTest2)/nsim,power3Test2=sum(rejectTest2[selected==3])/nsim,
powerTest3=sum(rejectTest3)/nsim,power3Test3=sum(rejectTest3[selected==3])/nsim,
powerTest4=sum(rejectTest4)/nsim,power3Test4=sum(rejectTest4[selected==3])/nsim,
powerTest5=sum(rejectTest5)/nsim,power3Test5=sum(rejectTest5[selected==3])/nsim,
powerTest6=sum(reject)/nsim,power3Test6=sum(reject[selected==3])/nsim)
}
#' Simualte for no CAR
#'
#' This function doesn't use an actual second point, just the projected based on the initial point
#' @param N1 Number of patients with secondary endpoint available at first analysis
#' @param N The total number of patients in the trial
#' @param n.trt The number of treatments in the trial
#' @param mean.s The mean for short term endpoint sample groups
#' @param mean.t Mean for the long term endpoint sample groups
#' @param p1 The covariate value for the first covariate
#' @param p2 THe covariate value for the second covariate
#' @param sigma0 sigma0 in the bivariate normal distribution
#' @param sigma is the known sigma for the population
#' @param rho is the known correlation between endpoints.
#' @param nsim The number of simulation runs. The default is 10000
#' @param tau1 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @param tau2 The chosen covariate adaptive randomization procedure. Default is Pocock's design
#' @export
simNoCarOrigBS = function(n1 = 20,N1=100, N=200, n.trt=3, mean.s=NULL, mean.t=NULL, p1 = .5, p2 = .5, sigma0=1, sigma=1, rho=0.5, nsim=10000,tau1 = 1,tau2 = 1,design = "Pocock",save.boundary, block.size = 12)
{
Norig =N
N1orig = N1
# set.seed(10101)
data = NULL
z.v = NULL
treat = NULL
covValues = NULL
n.looks = 2 #There are 2 planned analyses in this trial
alpha.star.u <<- c(0.0,0.025)
alpha.star.l <<- c(0.0,0.975)
if (is.null(mean.s)) mean.s = rep(0,n.trt+1)
if (is.null(mean.t)) mean.t = rep(0,n.trt+1)
selected=rep(0,nsim)
reject = rep(0,nsim)
trialprogress <<- c(1, 1)
for (sim in seq(1,nsim))
{
N1 = N1orig
N = Norig
look = 1
covValues = genCovValues(p = c(p1,p2),N=N*(n.trt+1))
#Getting treatment value assignments
treat = norandom(covValues = covValues, best = 0 , tr = NULL, n.trt = n.trt)
#Simulating data for these treatment assignments
# set.seed(101)
data = simulatedata.nocar(mean.s = mean.s, mean.t = mean.t, sigma = sigma, sigma0 = sigma0, rho = rho, tau1 = tau1, tau2 = tau2, treat=treat, covValues=covValues,inspection = look,data = NULL)
z1.bs = rep(0,200)
v1.bs = rep(0,200)
z2.bs = rep(0,200)
v2.bs = rep(0,200)
for ( i in 1:200) {
bs.s1 = sample(nrow(data), nrow(data), replace = TRUE)
databs = data[bs.s1,]
#Calculating test statistics z & v for this data, and selecting the best treatment
z.v = get.z.v.simulate(databs,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = z.v[1,3] #The best treatment was found in this function
# selected[sim] = best
look = 2
z = z.v[1:look,1]
v = z.v[1:look,2]
z1.bs[i] = z[1]
v1.bs[i] = v[1]
z2.bs[i] = z[2]
v2.bs[i] = v[2]
}
z.v = get.z.v.simulate(data,n.looks,look,z.v.prev=NULL, n1 = n1,N1= N1, N = N)
best = z.v[1,3] #The best treatment was found in this function
selected[sim] = best
look = 2
z = z.v[1:look,1]
v = z.v[1:look,2]
z2 = z[look]/sqrt(var(z2.bs))
v2 = v[look]/sqrt(var(v2.bs))
v1 = v[1]/sqrt(var(v1.bs))
t1percent = min(99,round(100*v1/v2))
boundary.value = sqrt(v2)*save.boundary[t1percent]
if (z2 > boundary.value) reject[sim] = 1
}
data.frame(power=sum(reject)/nsim,power3=sum(reject[selected==3])/nsim)
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.