The Duke Stroke Policy Model (SPM)

1
The Duke Stroke Policy Model (SPM)
2
Developers

• David Matchar, MD -- principal investigator
• Greg Samsa, PhD -- project director, statistician
• Giovanni Parmigiani, PhD -- statistician,
  software developer
• Joe Lipscomb, PhD -- health economist
• Greg Hagerty, MS -- software developer

3
Outline

• Rationale for modeling ()
• SPM described
• Applying the SPM to a randomized trial
• Extensions

4
Rationale for modeling

• Why model?
• Arguments for modeling
• Arguments against modeling
• Discussion
• Conclusions
• Application to stroke

5
Why model? (contd)

• To me, decision analysis is just the systematic
  articulation of common sense Any decent doctor
  reflects on alternatives, is aware of
  uncertainties, modifies judgements on the basis
  of accumulated evidence, balances risks of
  various kinds, . . .

6
Why model? (contd)

• considers the potential consequences of his or
  her diagnoses and treatments, and synthesizes all
  of this in making a reasoned decision that he or
  she decrees right for the patient
• (contd)

7
Why model?

• All that decision analysis is asking the
  doctor to do is to do this a lot more
  systematically and in such a way that others can
  see what is going on and can contribute to the
  decision process. -- Howard Raiffa, 1980

8
Advantages of modeling

• Clarifies decision-making
• Simplifies decision-making
• Provides comprehensive framework
• Allows best data to be applied
• Extrapolates short-term observations into
  long-term
• Encourages what if analyses

9
Disadvantages of modeling

• Ignores subjective nuances of patient-level
  decision-making
• Problem may be incorrectly specified
• Inputs may be incorrect / imprecise
• Usual outputs are difficult to interpret or
  irrelevant to decision-makers

10
Individual decision-making is subjective

• For individual decision-making, primary benefit
  of modeling is clarification.
• As normative process, decision-making works
  better for groups.
• Most applications involve group-, rather than
  individual-level, decisions (e.g., CEA,
  purchasing decisions, guidelines).

11
An aside

• Interactive software (possibly including models)
  shows great potential to help decision-makers
  (e.g., patients, physicians, pharmacy benefits
  managers) clarify and make better decisions.
• We are developing prototype for a user-friendly
  version of the SPM.

12
Some models are mis-specified

• A good model will simplify without
  over-simplifying.
• Poor models exist, but this need not imply that
  modeling itself is bad.
• We need more explicit standards under which
  models are developed, presented, and assessed.

13
An observation

• The fundamental problem with many of the poor
  models in circulation is that they assume the
  answer they are purporting to prove
• (often, that a treatment which is trivially
  effective or even ineffective is nevertheless
  cost-effective).
• Users are understandably wary.

14
Model inputs may be incorrect/ imprecise

• This problem is often most acute for utilities
  and costs, and least acute for natural history
  and efficacy.
• We need more and better data on cost and quality
  of life.
• The less certain the parameter, the greater the
  need for sensitivity analysis.

15
An aside

• In practice, the conclusions of a model / CEA are
  never stronger than the strength of the evidence
  regarding efficacy.
• If the evidence about efficacy is weak, then
  modeling / CEA should not be performed.

16
Usual outputs are difficult to interpret

• In academic circles, results are presented as
  ICERs using the societal perspective.
• Present this as a base case for purposes of
  publication / benchmarking.
• Also present multiple outcomes from multiple
  perspectives (vary cost categories, vary time
  periods, present survival, event-free survival,
  QALY, ).

17
General conclusions

• Modeling is of great potential benefit and indeed
  is sometimes the only reasonable way to proceed.
  However, models must be held to a high standard
  of proof.
• Although the standard reference model cannot be
  ignored, modeling should be done flexibly, with
  the needs of the end user in mind.

18
Application to acute stroke treatment

• RCTs follow patients in the short-term, but the
  large majority of benefits accrue in the
  long-term.
• Simple heuristics will not suffice to adequately
  trade off complex risks, benefits, and costs.
• Modeling allows a large body of evidence to be
  efficiently synthesized.

19
Outline

• Rationale for modeling
• Stroke model described ()
• Applying the SPM to a randomized trial
• Extensions

20
SPM described

• History / background
• Types of analysis
• Structure
• Validation / citations

21
SPM history / background
22
SPM development (contd)

• First version developed in 1993 by Stroke PORT
• Goals of stroke PORT
• Summarize epidemiology of stroke
• Describe best stroke prevention practices
• Describe current practices, and test methods for
  improving practice

23
SPM development

• SPM was used
• To summarize epidemiology of stroke
• To support CEA
• As a basic organizing structure for the PORT

24
SPM versions

• Original C code (uses waiting time
  distributions, research tool, difficult to
  extend)
• New S code (uses waiting time distributions,
  highly structured code used as development tool,
  inefficient)
• New C/Decision-Maker code (uses Markov-based
  cycles, intervention language, better interface,
  extendable)

25
New C version

• Decision-Maker used to specify natural history
  and effect of interventions in a decision tree
  format
• Efficient C code used as simulation engine
• Expandable into a web-based tool