Charge

1
Charge

• Can one apply a Bayesian analysis where the a
  priori data comes from an adult patient
  population and the new data comes from a
  pediatric population?
• Dr. S. Hirschfeld, Halloween, 2001

2
What Can Bayesian Methods Do For Us?

• FDA ODAC, Pediatric Subcommittee
• November 28, 2001
• Steven Goodman, MD, PhD
• Oncology Biostatistics
• Johns Hopkins University

3
What are Bayesian Methods?

• Methods based on Bayes theorem

a.k.a. Bayes factor
4
What are Bayesian Methods?

• Approaches that combine information of
  different types.
• Provide a formal way to make statistical
  inferences from a clinical trial, incorporating
  prior knowledge.
• A calculus of uncertainty.
• A calculus of belief.
• A calculus of evidence.

5
In what settings have Bayesian designs/analyses
been used?

• Pharmacokinetics
• Phase I - CRM
• Phase II - Thall/Simon
• Phase III - Spiegelhalter, Parmar, many others
• Meta-analysis

6
of Bayes Articles in Medical Journals
7
What Bayesian Methods Cannot Do

• They cannot tell us, in the absence of empirical
  or biological information, how alike children
  and adults are, and how relevant adult
  information is for children.

8
The Charge, Redux

• Can one apply a Bayesian analysis where the a
  priori data comes from an adult patient
  population and the new data comes from a
  pediatric population?
• Yes, but only if one makes an a priori judgment
  about how relevant the information from adults is
  for children.

9
What information comes from adults?

• Pharmacokinetics
• Pharmacogenomics
• Dose - Toxicity relationship
• Types of toxicity
• Frequency of toxicity / MTD
• Efficacy
• Covariate effects on all of above.
• Uncertainty in all of above.

10
What allows extrapolation to children?

• Empirical comparisons
• Knowledge of mechanism
• Known adult-child biologic and clinical
  properties of analgous drugs.
• Known sensitivity of children to specific
  toxicities.

11
How is prior information represented?

• Probability distributions on key parameters,
  expressing our best guess and degree of
  uncertainty.
• Key parameters
• MTD
• Response / Survival rate
• Toxicity rate
• Shape/slope of dose-toxicity curve
• Pharmacokinetic parameters

12
Bayesian Representation of Prior Knowledge
13
Another view of prior information

• Prior probability distributions are
  mathematically equivalent to information from N
  prior individuals.
• Made-up data.

14
Even weak knowledge Subjects not needed

• Confidence that cure rate lies within a 40 range
  (e.g. 20 -gt 60) corresponds to 25 patients
  worth of experimental information.
• Confidence that cure rate lies within a 20 range
  (e.g. 20 -gt 40) corresponds to 100 patients
  worth of experimental information.

15
What do these methods do for us?

• Properly account for uncertainty/ knowledge in
  both previous and current experimental data.
• Minimize the amount of information necessary from
  the current experiment - of value only if
  priors are reasonably accurate.
• Promote research treatments that reflect, as
  closely as possible, our best guess about what
  would be best for the child based on all prior
  information.
• Allow flexibility in design, because all Bayesian
  designs can be adaptive, i.e. responsive to
  accumulating data.

16
What do these methods do for us? (cont.

• Encourage extremely valuable discussions about
  prior knowledge and uncertainty, and about goals
  of study. e.g. toxicity target in CRM.
• Most standard approaches, used flexibly and
  with common sense, can become operationally
  indistinguishable from Bayesian ones, but this
  often requires more ad-hocery, and lack of
  theoretical foundation.

17
Charge

• Can one apply a Bayesian analysis where the a
  priori data comes from an adult patient
  population and the new data comes from a
  pediatric population?

18
Final Answer

• Yes, but only if the adult data is deemed
  relevant or informative.
• More empirical studies of this relevance need to
  be conducted, and ongoing.
• This may require incorporating additional
  endpoints and measurements into both adult and
  pediatric trials to facilitate this comparison.