3/12/2009 Decision and Cost-Effectiveness Analysis Eran Bendavid When Rationality Falters: Limitations and Extensions of Decision Analysis

3/12/2009Decision and Cost-Effectiveness Analysis

Eran Bendavid

When Rationality Falters: Limitations and Extensions of

Decision Analysis

Experiment Part 1


Assume that the United States is preparing for the outbreak of an unusual Icelandic disease, which is expected to kill 600 people in the absence of intervention.

Two alternative programs to combat the disease have been proposed. Assume that the exact estimates of the programs are as follows:

Experiment 1


If program A is adopted, 200 people will be saved.

If program B is adopted, there is 1/3 chance that 600 people will be saved and 2/3 probability that no people will be saved.

Experiment 1 – Switch Groups


If program A is adopted, 400 people will die.

If program B is adopted, there is a 1/3 chance that nobody will die, and 2/3 chance that 600 people will die.

Experiment 2 – Everyone


Imagine yourself $3000 richer than you are right now. You have to choose between (a) a sure gain of $1000, (b) a 50% chance of gaining $2000 and 50% of gaining nothing.

Imagine yourself $5000 richer than you are right now. You have to choose between (a) a sure loss of $1000, (b) a 50% chance of losing nothing and 50% chance of losing $2000.

Cartoon


Three Topics


FramesEquityEconomic epidemiology

Normative Problem Formulation


Classical decision theory axioms Ordering of preference Transitivity of preference Quantification of

judgment Comparison of

alternatives Substitution

Cost benefit rationale

“Risky prospects arecharacterized by their possible outcomes and by the probabilities of these outcomes.

The same option, however, can be framed or described

in different ways.” -- Tversky & Kahneman, 1981

Formulation Effects


Positive formulation Keep the status quo Risk averse

Negative formulation Gamble to achieve a better result Risk seeking

Mental Accounting


You set off to buy an iPod shuffle at what you believe to be the cheapest store in your neighborhood. When you arrive, you discover that the price of the Shuffle is $75, a price you believe is consistent with low estimates of the retail price.

A friend walks into the store and tells you a store 10 minutes away sells Shuffles for $55.

Do you go to the other store?Now suppose you are buying a MacBook Pro for

$1960, and a friend tells you it sells for $1940 in a store 10 minutes away. Do you go?

Different Frames


Real versus Hypothetical Experiment 1: What do you think? Experiment 2: Framing with hypothetical payoffs Experiment 3: Framing with real payoffs

Framing the choice to the civil jury can greatly affect the award

Framing the choice to the criminal jury Can help decide guilt or innocence Can affect the sentencing of the guilty

Framing Effects in Medical Decision-Making: Treatments


When framed positively (i.e. survival vs. mortality): Respondents 1.5 x more likely to choose surgery

over other treatments (i.e. radiotherapy) Respondents demonstrated increased preference

for invasive/toxic treatments No framing effect noted in hypothetical vs.

real life treatment decisionsMedicine use intention higher when results

presented as RRR vs. ARR or NNT

RRR, ARR, and NNT


RRR = Relative Risk ReductionARR = Absolute Risk ReductionNNT = Numbers Needed to Treat Dead AliveMeds 404 921CABG 350 974

Risk of death (from having CABG) = 350/1324 = 0.264Relative risk of death = 0.264/0.305 = 0.87 = 87%RRR = Amt of risk of death is reduced by surgery: 100% - 87%

= 13% ARR = Absolute amt of risk surgery reduces death: 30.5% -

25.4% = 4.1% NNT = # pts needing surgery to prevent 1 death: 1/ARR = 24

Source: http://www.ebm.worcestervts.co.uk/trial_results.htm

Conclusions on Frames


Humans are inconsistent. Framing is effective Framing can be manipulated to achieve desired outcomes Awareness of framing effects can make you a better

decision maker

Crucial in understanding discrepancies and inconsistencies in individual preferences.

Implications for Cost-Effectiveness Analysis


Important when considering perspective for analysis.

Preferences are dependent on framing and point of reference. Individual preferences vs. community preferences Preferences at time of illness or during recovery Availability of alternative treatments

Three Topics



Equity


Efficiency and Equity Both important for health care resource allocation

decisions Few guidelines for measuring or incorporating

equity Equity ~ Values

How can equity concerns be incorporated in cost-effectiveness analyses?

What is equity?


An equal and fair distributionAre treatments fairly allocated? Or

Are benefits fairly distributed?Canadian Common Drug Review Pharmacoeconomic

Review Template: “What equity assumptions were made in the analysis?”

No guidance on how to assess

Vertical Equity


Principle of vertical equity = allocation linked to “need”

Greater care is given to people with greater health needs

Sicker patients first priority for funding Goal is to create equity in eventual health status

Neglecting Vertical Equity


Implies all health outcomes are valued equally

Regardless of to whom they accrue

Conversely, paying attention to equity: Could make some relatively inefficient technologies more

attractive If benefits groups with greater claim to treatment Or could make efficient options less attractive

NICE (UK) Decisions, 1999 to 2002


Cost per QALY, £

Accepted Restricted Rejected

<20,000 14 3 1

20,000–30,000 0 4 0

>30,000 1 4 3

Sculpher, M.The use of quality-adjusted life-years in cost-effectiveness studies.Allergy 61 (5), 527-530.


Vertical equity may be controversialIf your definition of “need” is different than mineAssume we accept vertical equityWhat characterizes equity?How should it measured?

Controversy


The Incremental Cost-Effectiveness Ratio Comparing treatments A and B:

The cost of obtaining one extra unit of health effect

Cost-effectiveness analysis A measure of efficiency

Tradeoff between made explicit between scarce resources potential changes in health

AB

AB

effecteffect

costcostICER

Review of Efficiency

3/5/2009MS&E 292 - Health Policy Modeling

QALYs as a Measure of Health

Quality Adjusted Life Year

Life expectancy = 10 yearsQuality adjusted LE = 6.45 QALYs

0.5 0.5 0.5

0.750.750.8 0.8 0.8 0.8

0.25

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 2 3 4 5 6 7 8 9 10Years

Quality of LifeWeight

Are All QALYs Gains Equivalent?


4 QALYs

0

5

10

15

20

25

0 0.2 0.4 0.6 0.8 1

Quality of Life

Life

Exp

ect

an

cy

B

A

CD

E

1 QALY

7 QALYs

A’

B ′

E ′

C ′D ′

Each associated with a gain of 3 QALYs!

Steps in Applying Equity to CEA


1.Define groups which should receive priority to advance equity

2.Derive equity weights3.Determine how equity weights can be applied to

results of cost-effectiveness analyses (CEA) 4.Apply equity weighting to CEA results as a form of

sensitivity analysis

Some Possible Equity Factors


Baseline life expectancy

Baseline quality of life

Improvement in or final life expectancy

Improvement in or final quality of life

Duration of health benefits

Direction of health benefits

Distribution of benefits (number of people)

Health care endowment (prior expenditure)

Age

Personal behaviours

Relation to others

Social status

Lifetime health








Attribute Levels

Baseline quality of life (0 to 100) 306085

Gain in quality of life (0 to 100) 05

15

Baseline life expectancy (years) 210

Gain in life expectancy (years) 12

10

Number of beneficiaries 10010,000

1,000,000

Pre-program financial resources allocated to treat the condition ($)

05,000

50,000

Age (years) 154575

Survey to Understand Equity

3/5/2009MS&E 292 - Health Policy Modeling

Pilot in elected officials, municipal and provincial public clerks.

Participants recruited from waiting rooms at major Toronto downtown teaching hospital.

Asked to imagine they were voting in a referendum between 2 programs.

An Example


AttributesScenario

A B

Baseline QOL 30 30

Gain in QOL 0 15

Baseline LE 10 10

Gain in LE 10 2

N 10,000 100

Prior Allocation 50,000 5,000

Age 75 15

Number Selecting (%) 79 (29) 191 (71)

Solve the problem of equity?


Personal circumstances made such decision making challenging.

Several disliked the conceptual basis of the study, Fairness factors “aren’t measurable”

Trade-offs between attributes too complex

Individual or group values should dominate over centralized decision making

More Comments


7: interesting and thought-provoking14: challenging (3 both interesting and

challenging)5 wanted “equal” option to indicate that they

considered some scenarios equivalentSome felt that any rationing was objectionable

“Everyone should have a chance to be treated. It is up to the patient and his doctor to decide whether it is worthwhile”.

Choices are best left in the hands of God

Significant factors in equity…


Consistent with prioritization for those with poorer health

Less prior resource allocation viewed as having priority

Equal priority two groups alike except: 1st had a quality of life that was 50 points worse 2nd had an expected 10 year increase in life

expectancy

Equal priority two groups alike except: 1st 10 years younger 2nd had received about $13,000 less in prior

resources

Some Factors Not Significant


Number of people expected to benefit

Potential improvement in quality of life

Could have important implications for resource allocation models

Distributional aspects (“how many benefit?”) may be less important than the characteristics of individuals (“who benefits?”)







Equity-Weighted QALYs


Vertical equity Implies society values some health gains

more than othersFor example

A QALY gain a sick person more valuable than a QALY gain for a well person

Cancer drug vs. lifestyle drugOne often proposed solution is to adjust

QALYsQALYs transformed into “eQALYs”

= equity-weighted QALYs

An alternative to focusing on QALYs


Rather than focusing on health outcomeFocus on resource allocation decisionReframe vertical equity as “more willing to pay

for some health outcomes than others”i.e., a higher (or lower) willingness to pay

threshold

Advantage: Policy implications more transparent Accommodated by current CEA methods

Disadvantage Measurement more difficult “Knee-jerk” rejections more common?

Limitations of eQALYs


QALYs already controversialConstruct is artificial, somewhat foreignMeasurement issuesAlready conflate survival, quality of lifePutting equity in might confuse more than it

illuminatesAnd exacerbate concerns about subjectivity, valuesi.e. eQALY components:

Survival Objective Quality of life (preference) Subjective Equity weight Subjective and

value-laden

The Net Benefit Approach


Consider an ICERΔ C / Δ EDecision favorable if:

ICER < society’s willingness to pay for an extra unit of E (λ) Δ C / Δ E < λ

Define Net Monetary Benefit (NMB) asNMB = λ∙ Δ E- Δ CDecision favorable if NMB>0

Equity-weighted NMB


Assume an interventionWe want to assign an equity weight to the health

effectCall the equity weighting function f(∙)Equity-weighted health effect is f(Δ E,q)Where q is a vector of equity factors

So equity-weighted NMB is

NMB = λ∙ f(Δ E,q)- Δ C







Conclusions


Equity weighting the willingness to pay threshold is algebraically equivalent to equity adjusting QALYs

A form of sensitivity analysis, offers transparency, reproducibility

Focus on methods to estimate the relative attractiveness of allocating to different groups

Doesn’t obviate need for determining societal willingness to pay threshold

Equity Considerations


Fairness in process Accountability for reasonableness Fairness in outcomes

A decision that is: Transparent Principled Defensible

Three Topics



Traditional View of Epidemics


How is an epidemic started?

Index case The first case to start an epidemic Not necessarily the first case of the disease Epidemic is an interaction of the disease, the host, and the

susceptible population

Finding the Index Case


• Detective work to find “Patient Zero”• Gaëtan Dugas

• the French-Canadian gay flight attendant reputed to introduce HIV to the US.

• Made famous by the book “And the Band Played On”• The research used for that study was later repudiated• Introduction probably through Haiti rather than Africa

• Typhoid Mary• The SARS outbreak

• On Feb 21, 2003, a 65-year-old medical doctor from Guangdong checks into the 9th floor of the Metropole hotel in Hong Kong

The importance of a susceptible population and hosts


156 closecontactsof HCW

and patients

Index case from

Guangdong

Index case from

Guangdong

Hospital 2Hong Kong

4 HCW +2

Hospital 2Hong Kong

4 HCW +2

Hospital 3Hong Kong

3 HCW

Hospital 3Hong Kong

3 HCW

Hospital 1Hong Kong

99 HCW

Hospital 1Hong Kong

99 HCW

Canada12 HCW +

4

Canada12 HCW +

4

Hotel MHong Kong

IrelandIreland

USAUSA

New YorkNew York

Singapore34 HCW +

37

Singapore34 HCW +

37

Viet Nam37 HCW +

?

Viet Nam37 HCW +

?

BangkokHCW

BangkokHCW

4 otherHong Konghospitals28 HCW

4 otherHong Konghospitals28 HCW

Hospital 4Hong KongHospital 4

Hong Kong

B

I

K

F G

ED

CJ

H

A

GermanyHCW +

2

GermanyHCW +

2

SARS: 8,445 probable cases, 790 deaths


China (5328)

Singapore (206)

Hong Kong (1755)

Viet Nam (63)

Europe:10 countries (38)

Brazil (3)

South Africa (1)

Canada (238)

USA (70)

Colombia (1)

Kuwait (1)

South Africa (1)

Mongolia (9)

India (3)

Australia (5)

New Zealand (1)

Mongolia (9)

Russian Fed. (1)


Epidemic Models

S-I-R Models

Susceptible

Infected Removed

• Have no immunity• Never had the

disease• Have not

been immunized

Recovered and immune

Dead

POPULATION

S-I-R Models – simple setup


dS/dt=-S dI/dt=S-I dR/dt=-I

N = S + I + R

Susceptible

Infected Removed

POPULATION

S-I-R Models – more realistic


dS/dt=N-S-S dI/dt=S-I-I dR/dt=-I-R

N = S + I + R

Susceptible

Infected Removed


1. The prevalence of infection in the population2. How frequently a susceptible individual comes in

contact with an infected individual3. The probability of transmission from infected to

uninfected per contact

What determines infection ()?

Susceptible

Infected

What determines the rate of “removal” ()?


1. The duration of infection2. The disease’s case-fatality rate (of the people who

get infected, how many die)

Infected Removed

What does an epidemic look like?


0

200

400

600

800

1000

1200

0 0.1 0.2 0.3 0.4 0.5

numbe

r

time

St

It

Rt

R0


Called the basic reproduction numberThe average number of secondary cases a typical

infectious individual will cause in a completely susceptible population

Time = 0 Time = 1; R0=2

R0


R0>1

R0<1

Examples of R0


Disease Transmission R0

Measles Airborne 12-18

Pertussis Airborne droplet 12-17

Diphtheria Saliva 6-7

Smallpox Social contact 6-7

Polio Fecal-oral route 5-7

Rubella Airborne droplet 5-7

Mumps Airborne droplet 4-7

HIV/AIDS Sexual contact 2-5

SARS Airborne droplet 2-5Influenza (1918 pandemic strain) Airborne droplet 2-3

How does an epidemic propagate?


R0 only describes the initial epidemic dynamics, and will tell you whether an epidemic is likely to take hold in a population

To find out what will happen to the epidemic, need to look at the effective reproductive rate

Re(t) = R0 * S(t)

It is the number of new infections by a typical individual at a particular time

Other reproduction numbers


Eventually, insufficient susceptibles to maintain chains of transmission

When each infectious person infects less than 1 other (on average), epidemic dies out

Historical example


Implications


What is the effective reproductive rate for an endemic disease? Re(t)=1

Implications for preventing epidemics: Re(t)=1=R0*S(t)

As long as S(t) stays below 1/R0, the epidemic cannot propagate in a population!

1/R0 is the susceptibility threshold


If R0 is 2, then keeping 50% of the population protected will prevent an epidemic

If R0 is 10, then keeping 10% of the population susceptible (90% protected) will prevent an epidemic

What are the implications for vaccines? Measles (R0 is 17)?

Seasonal influenza (R0 is 2)?

This is the elimination criteria for epidemics

Vaccine coverage and disease


1-1/Ro

rubella measles

Traditional view of epidemics


Preventive efforts → reduction in disease transmission Reduced contact (abstinence) Reduced chance of transmission per contact (gloves) Reduced disease transmission → reduced prevalence Reduced prevalence → further

reduction in transmission

Risk of infection is directly proportional to prevalence. Therefore controlling prevalence is highest priority.

Preventive efforts are a public health imperative

Behavioral view of epidemics


Economists’ view: prevention and prevalence affect each other High prevalence disease → people increase personal

preventive measures

Low prevalence disease → people increase risky behaviors

Economic Epidemiology


Mathematical conceptualization of the interplay between economics, human behavior and disease to improve our understanding of the emergence, persistence and spread of infectious agents optimal strategies and policies to control their spread

Prevalence response elasticity Hazard rate into infection of susceptibles is a decreasing

function of prevalence (opposite of epidemiological model predictions)

Application to HIV Application to Measles

Empirical evidence?


Geoffard and Philipson, Int. Econ. Rev., 1996

Implications of behavioral view


Any highly prevalent epidemic is self-limiting because of preventive measures taken by individuals

Any campaign to eradicate a disease will run into trouble when after prevalence decreases HIV in USA


Documents

3/12/2009 Decision and Cost-Effectiveness Analysis Eran Bendavid When Rationality Falters: Limitations and Extensions of Decision Analysis