Early stage Phase 2 clinical trials are often designed as multi-stage single arm trials, which quickly identify inefficacious molecules and interventions, without subjecting too many patients to treatments with questionable standard of care. As the primary purpose of these designs is the early stopping for futility, it is often the case that very small cohorts enroll in early stages of the design. A larger cohort is only allowed to enroll when results from earlier enrollment suggest that there is clinical benefit to the new treatment.
The rise of Bayesian methods has meant that predictive power can be used to assess efficacy during these single arm Phase 2 studies, but how do they differ from traditional designs and when should they be used?
Simon’s Two Stage Designs
The most commonly used Phase 2 designs are Simon’s Two Stage Designs. There are in fact two such designs, the Optimal Two-Stage Design and the Minimax Design. Each of these has special properties that increase efficiencies while protecting patients. Both operate by enrolling a small number of patients in earlier stages to gauge treatment effect, before enrolling a larger number to make more precise estimates of response and toxicity. Response and toxicity are of course critical information to design optimal Phase 3 clinical trials, but arguably, a beneficial treatment effect should be established in Phase 2 prior to making these calculations. Hence Simon’s Two Stage Designs achieve the competing needs of ensuring patient safety while collecting vital data to optimize a Phase 3 trial.
The Optimal Two-Stage Design minimizes the expected sample size within these operating constraints, thereby ensuring the fewest number of patients are tested before robust toxicity can be calculated and other critical information be generated. The Minimax Design aims to minimize the maximum sample size for this same purpose. Both designs are able to control Type 1 and Type 2 error.
Three Stage Designs
If Two Stage Designs are lauded for ensuring that the fewest number of patients are subject to the most uncertainties, then in principle three stage designs would be even more superior. Unfortunately, operationalizing a three stage design has practical consequences, and many studies reveal that trying to “fix” the statistical properties of such trials after they have gone awry can compromise findings. On the other hand, gains in efficiency and the ethical gains may be such that the operational concerns are worth the gain.
Predictive Probabilities
The fixed multi-stage designs described above are sometimes charged with being insufficiently flexible. If our objective is to stop a clinical trial the moment we realize that a treatment is not efficacious, then why should clinicians wait for an interim look to make the decision to stop?
As trial sponsors know, it is impossible to tell before a trial begins, exactly when we will have sufficient information to make the call to stop a trial.
Predictive probabilities are estimations made during a clinical trial, that can tell a sponsor the probability that the clinical trial will have a positive outcome, should the trial continue. Using Bayesian updating methods, these predictive probabilities can be continuously calculated as every new piece of data is gathered about patients.
The Decision to Stop Early
Essentially, a well designed clinical trial that makes use of predictive probabilities can enable sponsors to select unplanned early stopping due to the ability to continuously monitor and calculate the chance of success. When using fixed designs like Simon’s Two Stage Designs, prospectively planned decision-rules become an essential feature of clinical trial design. These are Go/No-Go Rules that tell a sponsor how to respond to evidence generated during an interim analysis. Using fixed designs requires sponsors to invest in the search for optimal Go/No-Go Decisions.
Technology to Support Phase 2 Trials
When choosing whether to pursue a fixed design or a Bayesian design, a number of factors might play into consideration. Cytel’s new East Bayes software enables sponsors to compare the benefits of fixed multistage designs and Bayesian designs in an intuitive tool, for clear-sighted and quantitatively justified clinical design selection. Click below to learn more.
About Pantelis Vlachos
Pantelis is Principal/Strategic Consultant for Cytel, Inc. based in Geneva. He joined the company in January 2013. Before that, he was a Principal Biostatistician at Merck Serono as well as a Professor of Statistics at Carnegie Mellon University for 12 years. His research interests lie in the area of adaptive designs, mainly from a Bayesian perspective, as well as hierarchical model testing and checking although his secret passion is Text Mining. He has served as Managing Editor of the journal “Bayesian Analysis” as well as editorial boards of several other journals and online statistical data and software archives.
