Transcript Document

Cost-Effectiveness Analysis
and the Value of Research
David Meltzer MD, PhD
The University of Chicago
Overview
• Cost-effectiveness analysis has long been used to
assess the value of medical treatments and the
information that comes from diagnostic tests
• Newer value of information techniques have extended
these tools to assess the value of medical research
• Understanding behaviors determining use of medical
interventions in the context of heterogeneity is key to
assessing their value and priorities for research
• Research may be especially valuable when it can be
used to individualize care
Value of Medical Treatments
• Health effects
– Length/quality of life: QALYs
• Cost effects
• Choose all interventions for which
Dcost/DQALY < threshold
– Often $50-100K/QALY
• Widely accepted, >> 1000 applications
Value of Diagnostic Testing
U(T|S)
S
pU(T|S)+(1-p)U(N|H)
Test
U(N|H)
H
Don’t Test
S
H
Max{pU(T|S)+(1-p)U(T|H),
pU(N|S)+(1-p)U(N|H)}
Cost-Effectiveness of Medical
Interventions
Intervention
Neonatal PKU screening
Cost/LY
<0
Sec. prev. hyperchol. men age 55-64
2,000
Sec. prev. hyperchol. men age 75-84
25,000
Pri. prev. mild hyperchol. men age 55-64
99,000
Screening exercise test men age 40
124,000
Screening ultrasound every 5 yr. for AAA
907,000
Cost-Effectiveness of Pap Smears
Frequency
Increase in
LE vs.
no screening
Increase in
Cost vs.
no screening
Average
Cost per
Life-Yr
Saved
3 years
70 days
$500
$2,600/LY
70 days
$500
$2,600/LY
2 years
71 days
$750
$3,900/LY
1 day
$250
$91,000/LY
1 year
71 days
8 hours
$1,500
$7,300/LY
8 hours
$750
$830,000/LY
Marginal
Increase
in LE
Marginal
Increase
in Cost
Marginal
Cost per
Life-Yr Saved
Testing as Value of Information
U(T|S)
S
pU(T|S)+(1-p)U(N|H)
Test
U(N|H)
H
Don’t Test
S
H
Max{pU(T|S)+(1-p)U(T|H),
pU(N|S)+(1-p)U(N|H)}
Research as Value of Information
U(T|S)
S
pU(T|S)+(1-p)U(N|H)
Test
U(N|H)
H
Don’t Test
S
H
Max{pU(T|S)+(1-p)U(T|H),
pU(N|S)+(1-p)U(N|H)}
Value of Information Approach to Value of Research
• Without information
– Make best compromise choice not knowing true state of the
world (e.g. don’t know if intervention is good, bad)
• With probability p:
get V(Compromise|G)
• With probability 1-p: get V(Compromise|B)
• With information
– Make best decision knowing true state
• With probability p:
get V(Best choice|G)
• With probability 1-p: get V(Best choice|B)
• Value of information
= E(outcome) with information - E(outcome) w/o information
= {p*V(Best choice|G) + (1-p)*V(Best choice|B)} {p*V(Compromise|G) + (1-p)*V(Compromise|B)}
= Value of Research
Practical Applications of Value of Information
• Several full applications
– UK (NICE): Alzheimer’s Disease Tx, wisdom teeth removal
– US (AHRQ): Hospitalist research
– But needed data can be hard to obtain
• Bound with more limited data
– Murphy/Topel: DLE 3mo/yr*$50K/LY = $10K/person/yr = $3 Trillion/yr
– Real value of research may be far less than expected, e.g., for prostate
cancer:
• Maximal value of research
= $ 5 Trillion
• Expected value of perfect information
= $21 Billion
• Expected value of information
= $ 1 Billion
• Area of active investigation
– Most promising clearly for applied research
“Bayesian Value of information analysis: An
application to a policy model of Alzheimer's disease.”
Uncertainty in Incremental Net Benefits
Cost-Effectiveness Acceptability Curve
Value of Research by Time Horizon
Value of Research by Value of Health
Contributors to Value of Research
Practical Applications of Value of Information
• Several full applications
– UK (NICE): Alzheimer’s Disease Tx, wisdom teeth removal
– US (AHRQ): Hospitalist research
– But needed data can be hard to obtain
• Bound with more limited data
– Murphy/Topel: DLE 3mo/yr*$50K/LY = $10K/person/yr = $3 Trillion/yr
– Real value of research may be far less than expected, e.g., for prostate
cancer:
• Maximal value of research
= $ 5 Trillion
• Expected value of perfect information
= $21 Billion
• Expected value of information
= $ 1 Billion
• Area of active investigation
– Most promising clearly for applied research
– Increasing interest among pharma
Behavioral Cost-Effectiveness Analysis
• Value of health interventions depend on how they
are used
– Especially in the presence of heterogeneity
– True for treatments and for diagnostics
• Understanding behaviors determining use of
health interventions key to their evaluation
– Optimizing behavior: self-selection/diagnostic testing
– Non-optimal behavior: non-selective use
Standard CEA with Heterogeneous Individuals
CE
D costs
m
D effectiveness
Blue Dots = Treated Patients
Optimal Selection with Heterogeneity:
via Self-selection or Diagnostic Testing
CE
D costs
m
D effectiveness
Blue Dots=Pts gain from Tx; Orange Dots=Pts lose from Tx
Effect of Perfect Selection on CEA
CE
D costs
m
m’
D effectiveness
Blue Dots=Pts gain from Tx; Orange Dots=Pts lose from Tx (reject)
Empirical Selection
CE
D costs
m
D effectiveness
Blue Dots=Pts choose Tx; Orange Dots=Pts reject Tx
Background: Diabetes in the Elderly
• Diabetes care guidelines call for intensive lowering of
glucose among younger patients
• However, unclear if this should apply to older patients
– Gains in life expectancy smaller
– Side effects of treatment may dominate
– CE models of intensive therapy in older patients:
• Minimal or even negative effects on QALYs
• Not cost-effective
– Know many patients refuse intensive therapy
• Suggests self-selection may have important effects on
CEA in diabetes
Methods
• Interviewed 500 older diabetes patients to obtain data on
preferences
– Conventional and intensive glucose lowering (using insulin or oral
medications)
– Blindness, end-stage renal disease, lower extremity amputation
• Collected data on treatment choices and patient
characteristics by medical records review
• Used CDC simulation model of intensive therapy for type
2 diabetes and patient-specific demographic, health, and
preference data to get person-specific estimates of lifetime
costs and benefits
• Analyses of cost-effectiveness of intensive vs.
conventional therapy contrasting all patients vs. perfect
self-selection vs. empirical self-selection
Results: Intensive vs. Conventional Therapy
CE Approach
Group
Standard
Full Population
N
543
Change in Change in CE Ratio
Costs ($)
QALYs
($/QALY)
8076
-0.49
--
Perfect Self-Selection Effect for Intensive Therapy
20000
CE
15000
10000
5000
m
m’
0
-8
-6
-4
-2
0
2
4
-5000
Blue dots--the cost-effectiveness values of individuals with an expected benefit from intensive therapy.
Orange dots-- the cost-effectiveness values of individuals with a decrement in expected benefits with intensive therapy.
M-- CE ratio for whole population. M’—CE ratio after self-selection.
Results: Intensive vs. Conventional Therapy
CE Approach
Group
N
Change in Change in CE Ratio
Costs ($)
QALYs
($/QALY)
Standard
Full Population
543
8076
-0.49
--
Perfect SelfSelection
DQALY>0
403
8165
0.40
20K
DQALY<0
131
7906
-3.25
--
Empirical Self-Selection Effect for Intensive Therapy
20000
15000
10000
5000
0
-8
-6
-4
-2
0
2
-5000
Blue dots-- cost-effectiveness values for individuals who identify their care as intensive therapy.
Orange dots-- cost-effectiveness values for all other individuals.
M-- CE ratio for orange dot individuals. M’-- CE ratio for blue dot individuals.
4
Results: Intensive vs. Conventional Therapy
CE Approach
Group
Standard
Full Population
543
8076
-0.49
--
Perfect SelfSelection
DQALY>0
403
8165
0.40
20K
DQALY<0
131
7906
-3.25
--
Self-identified
intensive
therapy
154
7948
0.17
47K
All others
364
8164
-0.80
--
Empirical
Self-Selection
N
Change in Change in CE Ratio
Costs ($)
QALYs
($/QALY)
Implications - I
• Results of standard CEA may be misleading
– In contrast to the suggestion of standard CEA, offering
intensive glucose lowering to all older people likely
cost-effective
– CEAs should consider the importance of selfselection
• Distinction between perfect and empirical selfselection is potentially important
– Data on who will use a treatment if it is offered is
important
Implications - II
• A framework to account for heterogeneity
in patient benefits is key to valuing
diagnostic tests, guidelines, decision-aids,
or improved patient-doctor communication
that can make care more consistent with
variation in patient benefits
Motivation for Diagnostic Test/Decision Aids
CE
D costs
m
D effectiveness
Blue Dots=Pts choose Tx; Orange Dots=Pts reject Tx
Aim of Diagnostic Test/Decision Aids
CE
D costs
m
D effectiveness
Blue Dots=Pts choose Tx; Orange Dots=Pts reject Tx
Value of Diagnostic Test/Decision Aids
CE
D costs
m
Dc
D effectiveness
De
Blue Dots=Pts choose Tx; Orange Dots=Pts reject Tx
Value of Diagnostic Test/Decision Aid
• Effectiveness = Pts D De
• Costs = Pts D Dc
• Total Benefit
Cost-Benefit =
(1/l) Pts D De +
Pts D Dc
Net Health Benefit =
Pts D De + l Pts D Dc
Per Capita Value of Identifying Best
Population-level and Individual-level
Treatment in Prostate Cancer
Value
Best Population-level Treatment
$29
Best Individual-level Treatment
$2958
Implications - III
• Modeling heterogeneity and selection suggests a
framework to design co-payment systems to
enhance the cost-effectiveness of therapies
Motivation for Copayment (pc)
D costs
pc
CE
m
D effectiveness
Blue Dots=Pts choose Tx; Orange Dots=Pts reject Tx
Motivation for Copayment (pc)
D costs
pc
CE
m
D effectiveness
Blue Dots=Pts choose Tx; Orange Dots=Pts reject Tx
Per Capita Value of Identifying Best
Population-level and Individual-level Care
in Prostate Cancer with Full Insurance
Value
Best Population-level Therapy
$29
Best Individual-level Therapy
$2958
Best Individual-level Therapy
with Full Insurance
$41
Conclusions
• Cost-effectiveness analysis can be used to value
diagnostic testing and research on diagnostic testing
– Approaches exist to bound calculations with limited data
• Understanding behaviors determining use of
medical interventions in the context of
heterogeneity is key to assessing their value and
priorities for research
– Research may be especially valuable when it can be used
to individualize care
– Insurance and other determinants of use can significantly
alter value of research
Implications of Empirical CEA
• Need to consider how a treatment will be used in deciding
if it will be welfare improving
• Highlights importance of efforts to promote selective use
of treatments
– Biomarkers valuable if encourage selective use of
treatments
• Need to consider how a biomarker will be used in deciding
if it will be welfare improving
• Highlights importance of efforts to promote selective use
of biomarkers
– Biomarkers valuable if encourage selective use of
treatments
Non-selective Use and Empirical
Cost-effectiveness
• Cost-effectiveness analyses of interventions often
stratify cost-effectiveness by indication
• Yet technologies are often used non-selectively
• The actual (empirical) costs and effectiveness of
an intervention may be strongly influenced by
patterns of use
Example: Cox-2 Inhibitors vs. NSAIDs
DQALY
DCOST ($)
$/QALY
Fraction
Users
High Risk
0.085
4,721
56K
39%
Low Risk
0.026
14,123
537K
61%
Overall
0.042
11,584
276K