PMED: APPM Workshop: Challenges in Using Bayesian Analysis Approaches for Regulation of Medical Products - Telba Irony, March 15, 2019
1. Challenges in using
Bayesian and Decision Analysis Approaches for Regulation of
Medical Products
Telba Irony, PhD
Deputy Director, Office of Biostatistics and Epidemiology
Center for Biologics Evaluation and Research
FDA
SAMSI- Advances in Precision and Personalized Medicine
Panel: Regulatory Advancements at the FDA, March 15, 2019
Disclaimer
This presentation reflects the views of the author and should not be
construed to represent the policies of the U.S. FDA.
2. 2
• Subjectivity
How to choose the prior? Whose prior? All priors? Selection bias?
How to discount the prior?
Hierarchical models: how to choose the hyper parameters?
• Legal: prior information may not be legally available
• Need to control type I error rates; significance level fixed at
traditional values: 5% or 2.5%
If tradition cannot be relaxed, all prior information is discounted
no gains in using prior information
Need agreement to increase the traditional value to a higher level
1. Prior Information to Power Clinical Trials
• Agreement to be reached in advance between sponsor and FDA
(exchangeability; suitability of the prior, discount, etc.)
3. 3
Promising areas for using prior information
• Pediatric trials: extrapolation from adult population
https://www.fda.gov/downloads/medicaldevices/devicereg
ulationandguidance/guidancedocuments/ucm444591.pdf
• Rare diseases or Small populations
• Safety
• Unmet medical need for life threatening or irreversible
debilitating diseases
• Expedited Access Program (EAP) (CDRH and CBER)
EAP guidance
http://www.fda.gov/downloads/MedicalDevices/DeviceRegula
tionandGuidance/GuidanceDocuments/UCM393978.pdf
4. 4
Interim analyses for decisions on stopping or continuing recruiting
based on predictive distributions sample size decided and
optimized during the trial “Goldilocks” trials
Most advantageous when modeling is used in the likelihood:
results at early follow-up times predict results at the final follow-
up. Model refined at interim looks when all follow-up results from
patients recruited early are available.
Adaptive randomization: probability of assignment to a treatment
depends on data obtained thus far
Simulations are used to assess operating characteristics of the trial
design - control type I error rates and power
2. Bayesian Adaptive Designs
5. 5
3. Decision Analysis and Benefit-Risk Determinations
Benefit-Risk determinations for regulation of medical product
have been Qualitative
Drugs and Biologics Qualitative Assessment
Decision Factor Evidence and Uncertainties Conclusion & Reasons
Analysis of Condition
Unmet Medical Need
Clinical Benefit
Risk
Risk Management
Goal: Make B-R considerations explicit and transparent to audiences
within and outside the FDA
7. 7
a. Prophylactic Use of Inferior Vena Cava (IVC) Filters
Prevent Pulmonary Embolism when
anticoagulation drugs are contraindicated
Public Health Notification
Optimal time to remove the IVC filter after
implantation: between 1 and 3 months
Harms Weights
No device
Death 10
SAE 8
Device in
Situ
Occlusion 6
AE 1
Migration 3
Fracture 4
Removal AE 3
Postmarket Signal
8. 8
8
b. Postmarket Signal on Diagnostic Device
Possible Actions Consequences
Status Quo • Deaths and AEs
• Devices replaced slowly
Issue PHN • Deaths and AEs + deaths and AEs due to PHN
• Devices replaced faster
Removal • Deaths and AEs due to temporary shortage
• Devices replaced much faster
Influence Diagram
9. 9
c. Weight-Loss and Patient Preferences
Weight LossBenefit
Risk
New
Decision Aid Tool
• Calculates the minimum benefit patients would require for a treatment with
a given risk and other attributes (or maximum risk for a given benefit)
• Estimated values inform clinicians in the assessment of the “minimum
clinically significant weight loss ” to be used in trial design and analysis.
10. 10
4. The value of Bayes in the regulatory setting
• 1998 – Center for Devices Workshop
• 2004 FDA Workshop: Can Bayesian Approaches to Studying New
Treatments Improve Regulatory Decision Making?
• Slow progress
Strict control of type I error (a=5%) independently of context hinders
the use of informative priors
More success in adaptive trials with modeling predicting final results
at interim looks (not possible for frequentist adaptive designs)
Limited use of decision analysis and utility based decision rules
11. 11
Highest Value of Bayes in the Regulatory Setting
1. Account for the totality of external evidence via prior info
2. Use the likelihood principle for flexible clinical trial designs
3. Use modeling to build likelihood functions
4. Use Bayesian decision analysis to develop rational and
transparent decision rules
5. Define thresholds for approval by considering:
• Factors for benefit-risk determination (context)
• Patients’ and stakeholders utilities
6. Required strength of evidence scientifically determined by:
• Medical need
• Patient tolerance for risk and perspective on benefit
• Severity and chronicity of the disease, etc.