Traditionally the use of Bayesian analysis in clinical research has been restricted by concerns over how easily it can be understood by statistically non-technical colleagues or how regulatory authorities would receive the results. However, the use of Bayesian study designs and analysis methods have become increasingly popular due to communications from regulatory authorities such as the FDA, and the advantages over frequentist methods such as greater cost-effectiveness and flexibility in trial design it can offer in some situations.
Bayesian analyses offer a transformative approach to decision-making processes and statistical analysis in clinical trials. Some key advantages of the use of Bayesian methods include:
- Greater-informed decision making, leading to more intuitive and easily interpretable results
- Cost-effectiveness through reduced study duration and sample sizes
- Reducing the ethical burden of allocating patients to potentially ineffective treatment groups (i.e. placebo or standard of care) in debilitating disease areas
- Trial design flexibility by using Bayesian adaptive designs for a range of studies including dose-finding or in rare disease areas.
One of the main features of Bayesian methods which enable the advantages listed above is the ability to incorporate prior knowledge, expert opinions and accumulating study data into the analysis. Prior knowledge can be accrued from one or multiple relevant historical studies using a method such as network meta-analysis. By incorporating the prior information, greater-informed decisions can be made with less uncertainty than frequentist approaches, ensuring the best chance of making the correct decision. With defined clinical or trial success criteria; calculations such as the probability of success (PoS) can empower decision-makers when considering going ahead with a study design initially, or at an interim stage the predictive power of success (PPoS) can aid decisions to change study sample size or stop the study early for efficacy or futility. This can lead to improved cost-effectiveness.
In addition to informed decision making, Bayesian designs can boast efficiency in terms of sample size requirements. By leveraging prior information and continuously updating the probability distributions as new data is observed, Bayesian methods can reduce the number of participants or observations required to achieve the same level of statistical power as frequentist methods. This reduction in sample size can accelerate the research process translating to lower costs, and reduces ethical concerns in therapeutic areas where patient exposure to potentially ineffective or less-effective treatments should be minimized.
Bayesian adaptive designs utilize the accumulation of information as a trial progresses to modify key trial design features such as treatment group allocation and required sample size following a pre-defined adaptation algorithm. Once a stopping rule has been achieved, if successful a study could then go on to the next phase in a seamless design. Use of seamless designs is typically used for moving from phase II to phase III.
Some logistical advantages of a seamless design include:
- Potential elimination of the need for separate approvals for each phase
- Continuity in recruitment of patients and data collection
- Integration of data across phases, making programming more efficient
- Retaining patients between phases (especially beneficial in rare disease areas)
- Streamlining of data monitoring and regulatory submissions
While there are many advantages to using Bayesian approaches in clinical trials, more traditional frequentist approaches may be more appropriate depending on the situation. For example if there is little relevant prior information available, a group-sequential design may yield similar test results as a Bayesian adaptive design with identical stopping rules for both approaches at a fixed interim analysis.
Frequentist methods are more widely used in clinical trials than Bayesian methods, resulting in fewer case studies and less available information. This can create a greater challenge for regulatory authorities and stakeholders to accept the Bayesian approaches. With growing popularity, access to better Bayesian examples is likely to improve, consequently providing greater familiarity amongst stakeholders in the industry including regulatory authorities.
It is important to note that, in a similar manner to using a frequentist approach, a strategy (e.g. gatekeeping procedures, multiplicity adjustments, stopping rule definitions) should be considered when using a Bayesian adaptive design to ensure control of the Type-I error rate.
In summary, Bayesian methods can provide a robust framework for making well-informed decisions with smaller sample sizes, making them a powerful tool for cost-effective, flexible and logistically neat modern clinical trials. However, the decision to use a Bayesian approach should be carefully considered on a case-by-case basis.
Veramed’s Bayesian Expertise
Veramed has a wealth of statisticians experienced in a wide variety of Bayesian applications in clinical trials including; probability of success, continual reassessment methods (CRM), defining informative priors, meta-analytic predictive (MAP) priors, dynamic borrowing, meta-analysis, forecasting and event-tracking, sample sizing with a Bayesian primary endpoint, and Bayesian analysis methods including; Emax, MCMC, logistic regression, MMRM, fixed effects models, random effects models, hierarchical models and survival analysis.
Veramed has also recently constructed a Bayesian support team of approximately 20 statisticians with the mission of becoming an industry-leading provider of expertise in Bayesian trial design and analysis, by developing company-wide training ranging from the basics to more advanced topics.
Take two minutes to read our blog on ‘4 ways Statistical consultancy can save your trial‘
Or
Find out more about Veramed’s specialist consultancy services and contact our team today to learn how Veramed’s consultancy team could provide Bayesian support for your study.
