# Load bpp and rpact
library(bpp)Lade nötiges Paket: mvtnormlibrary(rpact)successR: exercises
Getting started
- Download the raw R markdown file that generated these exercises as a working file for all exercises and save it in your personal BEE area or on your local computer.
- Start up Rstudio on BEE or start up your local Rstudio installation. Then open the R markdown exercise file.
- Make sure that you have versions of
bppandrpactinstalled as specified here. - Start answering the questions below by adding your own
Rcode. Use thebpphelp and the example Rmd files for guidance.
\(\Rightarrow\) Good Luck!
Load bpp and rpact
Trial MO39193 (IMpassion132): update with external knowledge
This exercise does hypothetical DDCP computations for the below trial. The setup was simplified to be able to concentrate on aspects relevant to DDCP computations.
Background
Roche received accelerated approval for Triple-Negative Breast Cancer (TNBC) based on the IMpassion130 trial. As a post-marketing requirement (PMR) IMpassion131 was run, but that trial failed. As a consequence, it was evaluated whether yet another trial run by PDMA, IMpassion132, could stand in as PMR. To evaluate that feasibility it was of interest to update DDCP of IMpassion132 with the information from IMpassion 131, and that is what the task below is about.
Eventually, the TNBC label for Tecentriq based on IMpassion130 was withdrawn by Roche.
The trial
MO39193 (IMpassion132) is a phase III 1:1 randomized trial to investigate the efficacy and safety of Atezolizumab Plus Chemotherapy For Patients With Early Relapsing Recurrent (Inoperable Locally Advanced Or Metastatic) TNBC.
The primary endpoint of the trial is overall survival in PD-L1 positive patients.
The sample size section of the protocol states: The sample size is computed assuming that the trial gives 80% power for a hazard ratio of 0.7 and significance level of 0.05.
Questions and solutions
a) Compute the necessary number of events and the MDD for this trial.
Solution:
# add your R-code to answer the question here.b) Now assume a Normal prior centered around a hazard ratio of 0.75 and worth 50 events (1:1 randomization). What is the DDCP at design stage?
Solution:
# add your R-code to answer the question here.c) Now results of the related trial IMpassion131 become known: here, patients were randomized 2:1 and a hazard ratio of 1.12 was observed based on 120 events. How does DDCP change with this information?
Solution:
# add your R-code to answer the question here.d) Now assume that at trial onset there was little knowledge about the underlying effect, i.e. we assume that hazard ratios between 0.4 and 1.1 were equally likely. Specify such a prior and compute DDCP at the design stage.
Solution:
# add your R-code to answer the question here.The height of the pessimistic prior can be freely chosen as long as there is enough “area” left for the two tails, i.e. width \(\cdot\) height must be strictly \(<1\). See the tutorial on prior choice for a further discussion of this aspect. For the definition of the width we need to make sure we act on the log-scale, see also the tutorial on MIRROS.
e) Finally, plot the prior, the density of the external data, and the posterior for the Normal prior, as well as the density function of the pessimistic prior.
Solution:
# add your R-code to answer the question here.Protocol GA28950 v2 [binary endpoint]
This exercise does hypothetical DDCP computations for the below trial. The setup was simplified to be able to concentrate on aspects relevant to DDCP computations.
The trial
GA28950 is a phase III 4:1 randomized trial to investigate the efficacy and safety of etrolizumab during inducation and maintenance in patients with moderate to severe active ulcerative colitis who are refractory to or intolerant of TNF inhibitors.
The primary endpoint of the blinded induction cohort is the proportion of patients in remission at Week 14.
The sample size section of the protocol states (slightly modified): Cohort 2 patients will be randomized in a 4:1 ratio to etrolizumab or placebo. This will provide 90% power to detect a 10% difference in remission rates at Week 14 between the etrolizumab and placebo arms, under the assumption of a placebo remission rate of 5% and a two-sided \(\chi^2\) test at the 5% significance level.
Questions and solutions
a) Compute sample size and MDD.
Solution:
pi2 <- 0.05
# add your R-code to answer the question here.b) Assuming that 0.05 in the control arm is the true remission proportion, what proportion in the intervention arm do we need to observe to get a 2-sided \(p\)-value of 0.05 in this trial?
Solution:
c) Assume we want to be statistically significant at the final analysis, i.e. to beat the MDD. DDCP is the probability for this event, averaged over a prior distribution - write the formula for DDCP down.
Solution:
d) Write down the Normal approximation to the distribution of \(\delta\).
Solution:
e) Now assume in a 1:1 randomized Phase 2 trial with 50 patients we observed 4 and 2 remissions. Based on this result construct a Normal prior for the response proportion difference. The standard error is computed assuming an overall proportion of \(\pi_0 = 0.07\). Plot this prior distribution.
Solution:
# add your R-code to answer the question here.f) Now compute DDCP at the design stage using this prior.
Solution:
# add your R-code to answer the question here.Protocol WN29922 v2 (GRADUATE I, gantenerumab) [continuous endpoint]
This exercise does hypothetical DDCP computations for the above trial which was also discussed in the adaptR exercises. The setup below was simplified to be able to concentrate on aspects relevant to DDCP computations.
The trial
WN29922 is a phase III, multicenter, 1:1 randomized, double-blind, placebo-controlled, parallel-group, efficacy, and safety trial of gantenerumab in patients with early (prodromal to mild) Alzheimer's disease.
The primary endpoint of the trial is change from baseline (Day 1) to Week 104 in global outcome, as measured by the Clinical Dementia Rating - Sum of Boxes (CDR-SOB).
To cut the story short, the assumption was that for a continuous endpoint a sample size of approximately 500 patients for the trial has 80% power to detect a reduction in the change score from 2.5 to 1.75. The sample size calculation did not formally account for interim analyses but the protocol states that an interim analysis for efficacy and futility will be conducted after 50% of the targeted trial enrollment has been reached and that type I error control will be achieved by using O'Brien-Fleming boundaries approximated using the Lan-DeMets \(\alpha\)-spending method. The design specifications are as follows, just copying them over from the adaptR exercises to avoid hard-coding:
# Total sample size without inflation due to dropout
result <- getSampleSizeMeans(alternative = 2.5 * 0.3, stDev = 2.97,
sided = 2, alpha = 0.05, beta = 0.2)
design <- getDesignGroupSequential(sided = 1, alpha = 0.025, beta = 0.2,
informationRates = c(0.5,1),
typeOfDesign = "asOF",
futilityBounds = 0, bindingFutility = FALSE)
resultGS <- getSampleSizeMeans(design, alternative = 2.5 * 0.3, stDev = 2.97) Questions and solutions
a) Assume that at the design stage, a prior which puts equal weight on a reduction of 0% up to 50% is chosen. Specify such a prior and plot it.
Solution:
# add your R-code to answer the question here.b) Now, assuming a standard deviation of 3 in each arm compute the DDCP at the design stage.
Solution:
# add your R-code to answer the question here.c) Now assume that (unlike in the protocol) an interim for efficacy is performed after 50% of information, with MDD = -1.124. The trial does not stop at this interim, but the team does not know which estimate was observed. How does that change the team’s DDCP?
# add your R-code to answer the question here.