It’s no secret that drug development is a long and arduous process. Products routinely take many years to go from discovery through Phase I, II, and III testing and on to final approval. The situation is especially challenging in rare and pediatric disease, where small patient populations mean that traditional statistical methods cannot deliver the power that sponsors desire while protecting against bias and false positive signals, as regulators desire.
In recent years, Bayesian statistical methods have emerged as a helpful alternative in data-poor settings. These methods facilitate the combining of information from similar but not identical sources. For example, we may want to supplement a small concurrent control group in a clinical trial with historical information on controls accumulated in past trials, or from real-world data sources like disease registries or billing records.
Bayesian methods also facilitate the modeling of various biases that might arise from using less-than-perfect data sources, and can thus correct for them. Although Bayesian methods are often more computationally intensive than traditional methods, the ready availability of parallel high-speed computing and the burgeoning collection of both commercial and open-source software programs to implement these methods means that they are broadly available to sponsors looking to modernize and streamline their statistical toolkits.
While regulatory science is often more conservative, the FDA’s Center for Drug Evaluation and Research (CDER) and its Center for Biologics Evaluation and Research (CBER) have ramped up their support for Bayesian approaches in recent years. For example, the Complex Innovative Trial Design (CID) program, offers applicants considering novel trial designs extra meetings with regulators to further discuss key aspects of these designs[1].
Complex innovative trials are typically implemented using Bayesian methods, and often feature in-trial adaptations such as dose finding, dose-dropping, stopping early for either efficacy or futility, and adaptive borrowing from historical controls. A key “innovation” in all of these designs is that their adaptive nature means their operating characteristics (power and false positive rate) must be evaluated using simulation.
FDA also offers explicit help with orphan drug development through its Accelerating Rare Disease Cures (ARC) program[2]. This program has strengthened internal and external partnerships with stakeholders, including a variety of patient-focused initiatives in the rare disease space. A particularly novel and welcome aspect of this program is its Rare Disease Endpoint Advancement (RDEA) pilot program, which supports novel endpoint development for sponsors with an active IND or pre-IND for a rare disease[3].
In a similar vein, recent changes in U.S. drug and biologics regulations regarding early phase testing in oncology inspired by FDA Project Optimus suggest dramatic changes in statistical approach[4]. In the traditional paradigm, Phase 1 testing considers only safety outcomes, and seeks a single recommended Phase 2 dose (RP2D), typically the maximum tolerated dose (MTD), to advance to Phase II testing of preliminary efficacy.
While effective with traditional cytotoxic agents, this approach is inappropriate for more modern cancer therapies (e.g., therapeutic vaccines), for which the efficacy curve may rise rapidly with dose, and then flatten for doses far below the MTD. As such, Project Optimus suggests a search not for a single R2PD, but rather a recommended dose range (RDR), from the lowest minimally effective dose to the highest safe dose. A subset of doses from within the RDR can then be advanced to a fully randomized Phase 2 trial focusing on comparative efficacy.
Again, in this setting, Bayesian statistical methods are a helpful strategy, since they are more easily understood by clinical researchers and facilitate combining information across multiple data types and sources, including PK/PD, clinical, registry, natural history, and even expert opinion.
FDA’s new Bayesian Supplemental Analysis (BSA) Demonstration Project takes a slightly different approach to promoting innovative methods at FDA[5]. Unlike the CID program, this program is not intended to help sponsors use historical information-borrowing designs as primary analyses. Rather, it seeks to partner with sponsors to integrate Bayesian approaches in parallel with traditional frequentist analysis in simple, non-adaptive design settings where the traditional analysis continues to be primary. Here, Bayesian methods are used to supplement the primary analysis and may be used to evaluate the primary endpoint in the overall study population and/or in relevant subgroups (i.e., for subgroup analysis).
Finally, EMA regulators are also showing more interest in Bayesian and other novel statistical methods. A good recent example is their reflection paper on the use of single arm trials (SATs) as pivotal evidence[6]. This document acknowledges that clinical trials in many rare and pediatric disease areas do not permit randomization to placebo, often on ethical grounds. While this precludes traditional statistical methods for evaluating treatment efficacy, it does not eliminate the need for rigorous statistical evaluation and corresponding regulatory decision-making.
Once again, Bayesian methods carefully combined with modern methods for causal inference can offer sponsors a way forward even in the absence of a large number of randomized comparisons.
About the author:
Brad Carlin, Ph.D., is Senior Advisor, Data Science and Statistics, for PharmaLex.
[1] Complex Innovative Trial Design Meeting Program, FDA. https://www.fda.gov/drugs/development-resources/complex-innovative-trial-design-meeting-program.
[2] Accelerating Rare disease Cures (ARC) Program, FDA. https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/accelerating-rare-disease-cures-arc-program
[3] Rare Disease Endpoint Advancement Pilot Program, FDA. https://www.fda.gov/drugs/development-resources/rare-disease-endpoint-advancement-pilot-program
[4] Project Optimus, FDA. https://www.fda.gov/about-fda/oncology-center-excellence/project-optimus
[5] Bayesian Supplemental Analysis (BSA) Demonstration Project, FDA. https://www.fda.gov/about-fda/cder-center-clinical-trial-innovation-c3ti/bayesian-supplemental-analysis-bsa-demonstration-project
[6] Single-arm trials as pivotal evidence for the authorisation of medicines in the EU, EMA, April 2023. https://www.ema.europa.eu/en/news/single-arm-trials-pivotal-evidence-authorisation-medicines-eu
