Most people know that clinical drug discovery is usually conducted using either Frequentist or Bayesian methods. These two statistical paradigms have enjoyed a degree of competition historically, with some statisticians tauting the statistical rigor of Frequentist designs and others the intuitiveness and flexibility of Bayesian clinical trials. Recently, though, a number of hybrid methods have arisen, that leverage the benefits of both paradigms for singularly powerful clinical trials. A new Cytel article outlines the benefits of these combined methods.
Historical Borrowing
Bayesian Borrowing is perhaps one of the fastest growing tactics for combining the benefits of both Frequentist and Bayesian methods. Suppose a Frequentist trial faces limited resources, or needs a little more power to convince regulators of its statistical rigor. Bayesian methods can be used to borrow from historical datasets, incorporating already existing data into a new clinical trial to augment new findings. A recent Cytel publication noted the particular importance of the Meta-analytic predictive prior (MAP) for this purpose.
Model-Based Meta-Analyses
Model-Based Meta-Analyses (MBMA) play a role in utilizing early phase data to see which molecules tested in phase 2 ought to be moved forward into phase 3 of clinical development. The use of MBMAs mobilizes seamless phase 2/3 designs by ensuring that those molecules which move forward have a robust evidence package justifying continued investment into their efficacy.
Benefit-Risk Profiles
Bayesian methods can be integrated with Frequentist clinical trial designs to obtain clearer Benefit-Risk profiles for a number of new therapies. Ethically, every ounce of data collected ought to be used for these calculations, and Bayesian methods allow this to occur.
These new hybrid methods represent only a few of the strategies that integrate Bayesian methods with Frequentist clinical trial design. For a full review of the benefits of Bayesian methods in modern clinical trials read the complimentary paper:
