Transcript Bayesian Health Technology Assessment: An Industry Statistician`s

Bayesian Health Technology
Assessment: An Industry
Statistician's Perspective
John Stevens
AstraZeneca R&D Charnwood
Bayesian Statistics Focus Team Leader
Overview
• Some history
• Why Bayesian statistics?
• Summary
Some History
“something could be done”
Why is Bayesian statistics appropriate for health
technology assessment?
Incremental Cost-Effectiveness Plane
c
K
e
Inferences about Cost-Effectiveness
• Net (Monetary) Benefit,
(K) = K e - c > 0
• Objective to make inferences :
Q(K) = P( K e - c > 0 )
– Q(K) is the posterior probability that NB > 0
– Q(K) is the C/E-acceptability curve
This is intrinsically Bayesian!
Cost-Effectiveness Acceptability Curve
CEACs for three different prior structures
1.0
Exch
Nonpar
Weak
0.9
Q
0.8
0.7
0.6
0.5
0.4
1
10
100
1000
K
10000
100000
Why is Bayesian statistics appropriate for health
technology assessment?
More intuitive and meaningful inferences.
Why is Bayesian statistics appropriate for health
technology assessment?
Prior Information
• A fundamental feature of Bayesian statistics
• Represents information that is available in addition to
observing the data
• Prior information almost always exists and this should
be used to strengthen inferences
– Does not mean that inferences are necessarily more
favourable
– Particularly important where the design objectives of a
clinical trial may not relate to the effectiveness measure
– Can be structural information as well as numerical
information
• Prior information should always be used for internal
planning
Subjectivity
• Prior information is intrinsically subjective
• An unscrupulous analyst can concoct any desired
result
• The potential for manipulation is not unique to
Bayesian statistics
• But …..
– nowhere is the discipline of statistics carried out with more
discipline than in the pharmaceutical industry, and
– nowhere will Bayesian statistics be carried out with greater
discipline than in the pharmaceutical industry
• Synthesis of evidence should follow a formal process
for justifying the prior information
Synthesis of Evidence
“The elicitation and description of prior information
should be given at least as much care and attention
as the planning, execution and data scrutiny of the
trial itself.”
O’Hagan, Stevens and Montmartin (2001)
Why is Bayesian statistics appropriate for health
technology assessment?
Incorporates prior information in addition to the trial
data.
Why is Bayesian statistics appropriate for health
technology assessment?
Sensitivity Analysis
• Economic models are necessary in support of
arguments of cost-effectiveness
• Parameter estimates are typically drawn from a variety
of sources
• Univariate sensitivity analysis is still common
• Uncertainty and inaccuracy in parameter estimates
should be acknowledged
• Correlations between input parameters should also be
acknowledged
• Bayesian probabilistic sensitivity analysis has
advantages to both sponsors and decision makers
Guidance (1)
HTBS: Guidance for manufacturers on submission of
evidence relating to clinical and cost-effectiveness in
Health Technology Assessments 2002
It may be appropriate in some cases to combine the
cost-effectiveness evaluation with one or more meta
analyses of the clinical data. This may be done most
naturally in a Bayesian framework.
Guidance (2)
HTBS: Guidance for manufacturers on submission of
evidence relating to clinical and cost-effectiveness in
Health Technology Assessments 2002
Comprehensive sensitivity analyses should be
conducted. For parameters with substantial
uncertainty, sensitivity analysis using probability
distributions in a Bayesian framework is preferred.
Guidance (3)
NICE: Guidance for manufacturers and sponsors
2001
When data are drawn from a variety of sources and
used in a modelling framework, probabilistic
sensitivity analysis is recommended in order to take
account of the uncertainty around data values.
Why is Bayesian statistics appropriate for health
technology assessment?
Bayesian PSA allows the output uncertainty to be
analysed, with a range of diagnostics to identify the
most influential model inputs that are driving the
output uncertainty.
Why is Bayesian statistics appropriate for health
technology assessment?
Answering More Complex Questions
• Frequentist theory relies on large sample
approximations
• Sample means are not necessarily good estimators
of population means when data are skew in relatively
small samples
• Bayesian inferences can be computed exactly in
highly complex models
Why is Bayesian statistics appropriate for health
technology assessment?
The ability to tackle more complex problems.
Summary
• Bayesian statistics provides:
– More intuitive and meaningful inference
– The ability to incorporate prior information
– The ability to tackle more complex problems
• Bayesian PSA allows the output uncertainty to be
analysed, with a range of diagnostics to identify the
most influential model inputs that are driving the
output uncertainty.
Win! Win!