BDP2 workflow"

In BDP2: Bayesian Adaptive Designs for Phase II Trials with Binary Endpoint

options(width=80)
# out.format <- knitr::opts_knit$get("out.format")
# img_template <- switch( out.format,
#                      word = list("img-params"=list(fig.width=6,
#                                                    fig.height=6,
#                                                    dpi=150)),
#                      {
#                        # default
#                        list("img-params"=list( dpi=150,
#                                                fig.width=6,
#                                                fig.height=6,
#                                                out.width="504px",
#                                                out.height="504px"))
#                      } )
# 
# knitr::opts_template$set( img_template )

knitr::opts_chunk$set(echo = F,fig.width=7,fig.height = 3,dpi=300)



  pF = params$pF
  pE = params$pE
  p0 = params$p0
  p1 = params$p1
  shape1F = params$shape1F
  shape2F = params$shape2F
  shape1E = params$shape1E
  shape2E = params$shape2E
  n.range_1 = params$n.range_1
  n.range_2 = params$n.range_2
  n.range_4 = params$n.range_4
  cF = params$cF
  cE.range_1 = params$cE.range_1
  cE = params$cE
  firstInterim.at = params$firstInterim.at
  furtherInterims.at = params$furtherInterims.at
  nfinal = params$nfinal
  nfinal.vec = params$nfinal.vec 
  ptrue.range_1 = params$ptrue.range_1

Chosen Settings

• Uninteresting and target response rate: $p_0$=r p0, $p_1$=r p1
• Beta prior for futility criterion: Beta(r shape1F, r shape2F)

• Beta prior for efficacy criterion: Beta(r shape1E, r shape2E)

• First interim at: r firstInterim.at

• Second and potentially further interims at (selected based on logistic considerations): r furtherInterims.at
• $p_F$=r pF and $c_F$=r cF, i.e. futility criterion $P(p>$ r pF $|Data) <$ r cF
• $p_E$=r pE and $c_E$=r cE, i.e. efficacy criterion $P(p>$ r pE $|Data) \geq$ r cE

Figures

2a. Select timing of first interim analysis and futility stopping criterion, cF, such that the trial stops after a run of treatment failures

n.range_1=as.numeric(n.range_1)
par(mfrow=c(1,2),cex=.7,cex.main=.9) #cex.main=.7,cex.lab=.8,cex.axis=.7,
plotBDP2(x="n",y="Prob0Successes",n=c(n.range_1[1],n.range_1[2]),p0=p0,p1=p1)
plotBDP2(x="n",y="PostProb0or1Successes",n=c(n.range_1[1],n.range_1[2]),pF=pF,shape1F=shape1F,shape2F=shape2F)

The left figure shows true (for $p_{true}=p_0$, in green) and false (for $p_{true}=p_1$, in red) stopping probability. False stopping probability should be bounded to an acceptable limit, e.g. 5%. Any n with false stopping probability below this limit can be used as timing of first interim. The actual choice may be guided by logistical considerations.

From left figure, false stopping probability was evaluated and first interim chosen to be performed at r firstInterim.at patients.

From right figure read P(p>$p_F$| 0 of n), in red, and P(p>$p_F$| 1 of n), in green, for the selected timing of first interim, i.e. for r firstInterim.at patients. The value of $c_F$ can be chosen anywhere between these two numbers and will ensure that the trial is stopped at first interim if no successes have been observed so far.

From right figure $c_F$ was chosen: $c_F$ = r cF.

Note: If it is not intended to stop the trial only after a run of treatment failures but also in case that a low number of successes is observed, the timing of first interim and $c_F$ can be chosen on basis of the operating characteristics and expected patient number without evaluating these plots.

Timings of second and potentially further interims are selected based on logistical considerations. The choice was to perform (a) further potential interim(s) at r furtherInterims.at patients.

3. Selecting cE

cE.range_1=as.numeric(cE.range_1)
cEaccuracy=0.001
cEvec=seq(from=cE.range_1[1],to=cE.range_1[2],by=cEaccuracy)
interims.at= c(firstInterim.at,as.numeric(unlist(strsplit(furtherInterims.at," "))))
n=nfinal

par(cex.lab=.9,cex.axis=.9,mar=c(5.1, 4.1, 0.5, 2.1))
res=plotBDP2(x="cE",y="PEcall",n=n,interim.at=interims.at,pF=pF,cF=cF,pE=pE,cE=cEvec,p0=p0,p1=p1,
             shape1F=shape1F,shape2F=shape2F,shape1E=shape1E,shape2E=shape2E,col=c("green","red"),cex.sub=.8)

abline(v=cE,col="gray",lty="dashed")

The plot shows power (in red) and type I error (in green) as function of $c_E$.

Threshold $c_E$ should be selected such that type I error probability can be tolerated and power is maximized.

Selected was $c_E$ = r cE. This leads to a type I error of r round(res$y.p0[which.min(abs(res$x.p0-cE))[1]],3) at $p_0$=r p0, and a power of r round(res$y.p1[which.min(abs(res$x.p1-cE))[1]],3) at $p_1$=r p1.

4. Operating characteristics

  n.range_2=as.numeric(n.range_2)
  nvec=c(n.range_2[1]:n.range_2[2])
  interims.at= c(firstInterim.at,as.numeric(unlist(strsplit(furtherInterims.at," "))))

  par(mfrow=c(1,2),cex.main=1,cex.sub=.9,cex=.6,cex.lab=1.1,cex.axis=1.1)
  plotBDP2(x="n",y="PFstopEcall",n=nvec,interim.at=interims.at,pF=pF,cF=cF,pE=pE,cE=cE,ptrue=p0,
                   shape1F=shape1F,shape2F=shape2F,shape1E=shape1E,shape2E=shape2E, progress=T,#cex.legend=.8,
         main="Type I error and probability of true stopping \nfor varying n")
  plotBDP2(x="n",y="PFstopEcall",n=nvec,interim.at=interims.at,pF=pF,cF=cF,pE=pE,cE=cE,ptrue=p1,
                   shape1F=shape1F,shape2F=shape2F,shape1E=shape1E,shape2E=shape2E,progress=T,#cex.legend=.8,
         main="Power and probability of false stopping \nfor varying n")

The left plot shows the probabilities for declaration of efficacy and cumulative stopping for futility at final for $p_{true}=p_0$. The right plot shows the same quantities for $p_{true}=p_1$.

5. Power function for multiple sample sizes at final analysis and corresponding expected number of patients in the trial as a function of ptrue

nvec= as.numeric(unlist(strsplit(nfinal.vec," ")))
pvec=seq(ptrue.range_1[1],ptrue.range_1[2],by=.01)
interims.at= c(firstInterim.at,as.numeric(unlist(strsplit(furtherInterims.at," "))))

par(mfrow=c(1,2),cex=.9,mar=c(5.1, 4.1, 0.5, 2.1))
n=nvec[1]
vn.int=interims.at[interims.at<n]
plotBDP2(x="ptrue",y="PEcall",n=n,interim.at=vn.int,pF=pF,cF=cF,pE=pE,cE=cE,ptrue=pvec,
                  shape1F=shape1F,shape2F=shape2F,shape1E=shape1E,shape2E=shape2E,col=1,cex.sub=.7)
abline(v=p0,col="grey")
abline(v=p1,col="grey")

for (jj in 2:length(nvec)) {
  n=nvec[jj]
  vn.int=interims.at[interims.at<n]
  plotBDP2(x="ptrue",y="PEcall",n=n,interim.at=vn.int,pF=pF,cF=cF,pE=pE,cE=cE,ptrue=pvec,
                    shape1F=shape1F,shape2F=shape2F,shape1E=shape1E,shape2E=shape2E,add=TRUE,col=jj)
}
legend("bottomright",title="final analysis at ",legend=nvec,text.col = 1:length(nvec),cex=.8)

n=nvec[1]
vn.int=interims.at[interims.at<n]
plotBDP2(x="ptrue",y="ExpectedNumber",n=n,interim.at=vn.int,pF=pF,cF=cF,pE=pE,cE=cE,ptrue=pvec,shape1F=shape1F,shape2F=shape2F,col=1,ylim=c(0,max(nvec)),cex.sub=.7)
abline(v=p0,col="grey")
abline(v=p1,col="grey")

for (jj in 2:length(nvec)) {
  n=nvec[jj]
  vn.int=interims.at[interims.at<n]
  plotBDP2(x="ptrue",y="ExpectedNumber",n=n,interim.at=vn.int,pF=pF,cF=cF,pE=pE,cE=cE,ptrue=pvec,shape1F=shape1F,shape2F=shape2F,col=jj,add=TRUE)
}
legend("bottomright",title="final analysis at ",legend=nvec,text.col = 1:length(nvec),cex=.8)

6. Decision boundaries for futility and efficacy (Design specification for clinician) and check for contradictory results

vn.int=firstInterim.at
n=n.range_4[2]
plotBDP2(x="n",y="bFbE",n=n,pF=pF,cF=cF,pE=pE,cE=cE,
          shape1F=shape1F,shape2F=shape2F,shape1E=shape1E,shape2E=shape2E,col=c("red","green"),cex.sub=.8,cex.main=.9)

The plot shows the decision boundaries for the selected trial design in terms of number of successes per number of enrolled patients.

The trial is stopped for futility if the observed number of successes among enrolled patients is in the red part of the plot, including the boundary. Efficacy can be called if the observed number of successes among patients at final analysis is in the green part of the plot, including the boundary.

The range of final sample size should be increased to the maximal feasible number to check for contradictory results, i.e. for overlap between red and green areas.

Reference

Kopp-Schneider, A., Wiesenfarth, M., Witt, R., Edelmann, D., Witt, O. and Abel, U. (2018). Monitoring futility and efficacy in phase II trials with Bayesian posterior distributions - a calibration approach. Biometrical Journal, to appear.

BDP2 documentation built on May 1, 2019, 9:03 p.m.