Bayesian Statistics in Medical Device Clinical Trials: Design, Analysis and Implementation

What You Need to Know? 👇

What is Bayesian statistics and how does it differ from traditional statistical methods in medical device trials?

Bayesian statistics is an approach for learning from evidence as it accumulates, using Bayes’ Theorem to formally combine prior information with current trial data. Unlike traditional frequentist methods that only use previous studies at the design stage, Bayesian methods continuously update inferences as new data becomes available.

When should I consider using Bayesian methods for my medical device clinical trial?

Consider Bayesian methods when you have good prior information from previous studies, need adaptive trial designs, want to incorporate historical controls, or require complex modeling that’s difficult with frequentist approaches. They’re particularly useful for medical devices due to their predictable physical mechanisms of action.

What are the main regulatory challenges when proposing a Bayesian trial to FDA?

Key challenges include extensive preplanning requirements, need for early FDA collaboration, complex model-building, specific computational expertise requirements, and ensuring the prior information doesn’t overwhelm current data. FDA agreement on prior information and models is crucial before trial initiation.

How do I determine appropriate sample sizes for Bayesian clinical trials?

Sample size depends on variability, prior information, mathematical models, parameter distributions, and decision criteria. Unlike fixed frequentist designs, Bayesian trials can use stopping criteria based on sufficient information gathering, with predictive distributions continuously updating expected additional observations needed.

What software and computational tools are recommended for Bayesian medical device trials?

WinBUGS is commonly used non-commercial software for Bayesian calculations, along with BRUGS and OpenBUGS variants. These programs use Markov Chain Monte Carlo (MCMC) sampling methods. FDA recommends consulting with their statisticians before choosing specific software for your submission.

How should I handle missing data and interim analyses in Bayesian trials?

Bayesian methods offer flexibility for missing data through predictive probabilities, assuming missing data are exchangeable with observed data. For interim analyses, use posterior probabilities or predictive distributions for stopping decisions, but ensure all methods are pre-specified and agreed upon with FDA.

---

What You Need to Do 👇

Recommended Actions

1. Meet with FDA early to discuss planned use of Bayesian methods and obtain agreement on:
   • Prior information to be used
   • Statistical models and assumptions
   • Trial design and adaptations
   • Success criteria
2. Document in protocol:
   • Justification for using Bayesian approach
   • Sources and validation of prior information
   • Statistical analysis plan including models
   • Operating characteristics through simulation
   • Sample size justification
   • Interim analysis plans if applicable
3. Implement appropriate controls:
   • Verify computational methods and convergence
   • Conduct sensitivity analyses
   • Check model assumptions
   • Document all analyses
4. For submission:
   • Submit data and analysis code electronically
   • Provide clear presentation of results
   • Include sensitivity analyses
   • Justify all modeling choices
   • Present labeling in understandable terms
5. Plan for post-market:
   • Define how Bayesian updating will be used
   • Specify data collection and analysis methods
   • Document process for incorporating new information

Key Considerations

Clinical testing

• Clinical trial design principles remain the same as traditional trials (randomization, blinding, etc.)
• Prior information from previous clinical studies can be incorporated if sufficiently similar
• Minimum sample size should be defined based on safety and effectiveness endpoints
• Operating characteristics (type I and II error rates) should be evaluated

Software

• Software used for Bayesian analysis should be discussed with FDA statisticians
• Program code and data used for simulations should be submitted electronically
• Convergence of computational algorithms should be verified

Labelling

• Results from Bayesian trials should be expressed clearly in device labeling
• Bayesian terminology should be explained in a way that is easy to understand
• Credible intervals should be reported

Safety

• Safety endpoints may require larger sample sizes than effectiveness endpoints
• Prior information on safety should be carefully evaluated
• Sensitivity analyses should be performed for safety claims

Other considerations

• Prior information should be carefully selected and justified
• Adaptive trial designs should be pre-specified
• Model assumptions should be checked and sensitivity analyses performed
• Missing data should be appropriately handled
• Post-market surveillance can utilize Bayesian updating of pre-market data

---

Relevant Guidances 🔗

• Bayesian Statistics in Medical Device Clinical Trials
• Design Considerations for Medical Device Pivotal Clinical Studies

Related references and norms 📂

• ISO 14155: Clinical investigation of medical devices for human subjects - Good clinical practice

---

Original guidance

• Bayesian Statistics in Medical Device Clinical Trials: Design, Analysis and Implementation
• HTML
• Issue date: 2010-02-04
• Last changed date: 2020-01-19
• Status: FINAL
• Official FDA topics: Medical Devices, Biostatistics
• ReguVirta ID: bf385674933d595f298ce2cd3ac8c220