Outcomes in Decision Analysis: Utilities, QALYs, and Discounting Aaron B. Caughey, MD, PhD Associate Professor in Residence Director, Center for Clinical and Policy Perinatal Research Department of Obstetrics and Gynecology University of California, San Francisco January 14, 2010

Disclosures No personal financial disclosures Research Funding: NIH/NICHD AHRQ – Elective Induction of Labor Robert Wood Johnson Foundation – Cesarean Delivery: Outcomes, Preferences, Costs Hellman Foundation

Overview  Back to the aneurysm example:  To Clip Or Not To Clip?  Clinical Outcomes  Utilities and utility measurement  Standard Gamble  Time Tradeoff  Calculating quality-adjusted life years  Discounting

Review—Last Lecture Formulated an explicit question Formulated an explicit question “to clip or not to clip” (incidental aneurysm ) “to clip or not to clip” (incidental aneurysm ) Made a simple decision tree Made a simple decision tree Conducted an expected value calculation to determine which course of action would likely yield the highest life expectancy Conducted an expected value calculation to determine which course of action would likely yield the highest life expectancy

To Clip or Not To Clip.865 vs.977 =1.0 =.55 =.9825 =.9921 =.977 Diff = =0

To Clip or not to Clip? Has an impact on life expectancy Has an impact on life expectancy Also actual clinical outcomes: Also actual clinical outcomes: Surgical death Surgical death Aneurysm rupture Aneurysm rupture Death from aneurysm rupture Death from aneurysm rupture Neurologic Injury Neurologic Injury Major Major Minor Minor Fear of aneurysm rupture Fear of aneurysm rupture

Quantifying Health Outcomes Mortality Life Years number of expected years of life Significant Morbidity Paralysis, loss of sight Quality Adjusted Life Years Expected life years adjusted for the valuation of the possible states in each year Financial Valuation of these Outcomes Costs to patient, payor, or society Willingness to pay to avoid outcomes, obtain treatment

Health Outcomes – Mortality Mortality Mortality Death from disease/accident/procedure e.g. If Ms. Brooks undergoes surgery, one of the possible outcomes is mortality Life Years Life Years Calculate an expected value of life years using a probabilistically weighted average of expected life e.g. If Ms. Brooks does not undergo surgery, her life expectancy is less than if she did not have aneurysm, these outcomes are measured in expected life years

Health Outcomes – Morbidity Morbidity Morbidity Some health state that is less than perfect e.g. disability from stroke, chronic pain Comparison of morbidities Comparison of morbidities Difficult – apples and oranges problem e.g. which is worse: Blind v. Deaf Deaf v. Paraplegia Paraplegia v. Blind

To Clip or not to Clip? Clinical outcomes for clinician readers Clinical outcomes for clinician readers Outcomes may affect health-related quality of life: how do we compare? Outcomes may affect health-related quality of life: how do we compare? Neurologic injury can cause mild/moderate disability Neurologic injury can cause mild/moderate disability Not clipping can cause anxiety associated with being at risk of aneurysm rupture Not clipping can cause anxiety associated with being at risk of aneurysm rupture Outcomes may occur at different times Outcomes may occur at different times

How do we incorporate quality-of-life effects into DA? Measure/estimate and apply utilities Measure/estimate and apply utilities Use utilities to quality-adjust life expectancy for decision and cost-effectiveness analysis Use utilities to quality-adjust life expectancy for decision and cost-effectiveness analysis

Preview—Where We Are Going with this Analysis? Recall Ms. Brooks and her incidental aneurysm -- to clip or not to clip? We want to: Determine her utilities Determine her utilities Use them to generate QALYs Use them to generate QALYs Evaluate incremental QALYs and cost (CEA/CUA) Evaluate incremental QALYs and cost (CEA/CUA) Compare incremental cost effectiveness ratios (ICER) to other currently accepted medical interventions Compare incremental cost effectiveness ratios (ICER) to other currently accepted medical interventions

What is a Utility? Utility - Quantitative measure of the strength of an individual’s preference for a particular health state or outcome. Utilities can be obtained for: * Disease states (diabetes, depression) * Treatment effects (cure, symptom management) * Side effects (impotence, dry mouth) * Process (undergoing surgery, prenatal diagnostic procedure)

Utilities Utilities are the currency we use to assign values to outcomes Scaled from 0 to 1 1 = perfect or ideal health or health in the absence of the condition being studied 0 = death

How are utilities measured? Utilities are commonly estimated using comparisons to the 0 and 1 anchors Visual Analog Scale Visual Analog Scale Standard Gamble Standard Gamble Time Trade-off Time Trade-off

BKA vs. AKA Example Patient in the hospital has infection of the leg Two options: 1) BKA BKA –1% mortality risk BKA –1% mortality risk 2)Medical management – 20% chance of infection worsening and needing AKA 2)Medical management – 20% chance of infection worsening and needing AKA AKA – above the knee amputation AKA – above the knee amputation 10% mortality risk 10% mortality risk Let’s draw a decision tree

For which outcomes do we need to measure utilities? Death? Death? Risk of worsening? Risk of worsening? Living with part of a leg (below the knee) missing? Living with part of a leg (below the knee) missing? Living with a bigger part of a leg (above the knee) missing? Living with a bigger part of a leg (above the knee) missing? Others? Others?

Visual Analog Scaling Full health: intact leg Dead BKA Outcomes rated on a 0-to-100 “feeling thermometer.” AKA

Standard Gamble What chance of immediate death would you be willing to incur to avoid living with the outcome being assessed? Method relies on respondents choosing between: 1) a certain outcome (BKA) 1) a certain outcome (BKA) 2) a gamble between an ideal outcome (intact leg) and the worst outcome (dead) 2) a gamble between an ideal outcome (intact leg) and the worst outcome (dead)

Standard Gamble Question Death Perfect Health

xercise Standard Gamble Exercise Spend the rest of your life with BKA ] immediate death [p]% chance of immediate death 1-[p]% chance of spending the rest of your life with an intact leg Which do you prefer? Choice AChoice B

Standard Gamble Standard gamble measurement involves questioning patients to determine the p at which the two outcomes are equivalent Using expected utilities, the value of p gives the utility Utility (BKA) x Prob (BKA) = Utility(cure) x (p) + Utility(death) x (1-p) The utility of BKA = p: note P(BKA) = 1 Utility (BKA) = [Utility(cure) x (p) + Utility(death) x (1-p)] = [1.0 x p + 0 x (1-p)] = p

Time Tradeoff How many years of your life would you be willing to give up to spend your remaining life without the condition/health state being assessed? Method relies on respondents choosing between: 1) Full life expectancy with the condition/outcome being assessed (BKA) 1) Full life expectancy with the condition/outcome being assessed (BKA) 2) A reduced life expectancy with the ideal outcome (intact leg) 2) A reduced life expectancy with the ideal outcome (intact leg)

Time Tradeoff Preference Elicitation Spend the remaining 40 years of your life with BKA Live 40 more years of life with an intact leg (give up 0 years of life) Which do you prefer? Choice AChoice B

Time Tradeoff Preference Elicitation Spend the remaining 40 years of your life with BKA Live 30 more years of life with an intact leg (give up 10 years of life) Which do you prefer? Choice AChoice B

Utility Measurement – Time Trade-off Time Trade-off involves patients choosing between: quality of life v. length of time alive When patients are equivocal between choice: Time A * Utility A = Time B * Utility B e.g. If you have a life expectancy of 30 years with a BKA; how much time would you give-up to live in your current state? Would you give up 5 years? 3 years? 1 year? 30 years * Utility (BKA) = (30-x) years * 1.0 If you’re willing to give up 3 years, that means: Utility of BKA = [(30-3)*1/ 30] = 27/30 = 0.9

Pros and Cons - VAS Advantage: Easy to understand Disadvantages: Doesn’t require the respondent to: Think about what they’d be willing to give up Think about what they’d be willing to give up Explore risk preference Explore risk preference Values spread over the range Values spread over the range

Pros and Cons – SG Advantages: Requires assessor to give something up, incorporates risk attitude Disadvantages: Choices may be difficult to make Most confusion-prone method Lack of engagement or willingness to participate in exercise Values tend to cluster near 1

Pros and Cons – TTO Advantages: Still asking assessor to give something up Easier choices than SG. Values not so clustered near 1 Disadvantages: Fails to incorporate risk Lack of clarity of when time traded occurs Isn’t something that one can choose to give up. (One can take on a risk of death, but not “pay with life years.”)

Utilities in decision analysis Utilities can adjust life expectancy in DA where outcomes include morbidity/quality- of-life effects. Utilities can adjust life expectancy in DA where outcomes include morbidity/quality- of-life effects. Quality Adjusted Life-Years (QALYs) Quality Adjusted Life-Years (QALYs)

QALYs QALYs are generally considered the standard unit of comparison for outcomes QALYs are generally considered the standard unit of comparison for outcomes QALYs = time (years) x quality (utility) QALYs = time (years) x quality (utility) e.g. 40 years life expectancy after AKA, e.g. 40 years life expectancy after AKA, utility (AKA) = 0.9 utility (AKA) = 0.9 = 40 x 0.9 = 36 QALYs = 40 x 0.9 = 36 QALYs

Back to aneurysm

.865 vs.977 =1.0 =.55 =.9825 =0 =.9921 =.977 Diff =

Now we want to add utilities for intermediate outcomes Normal survival 1.0 Worry about possibility of aneurysm rupture 0.95 Stroke (clipping complication or aneurysm rupture) ( )/2=0.5 Early death 0.5 Immediate death 0.0

Including utility for early death and stroke=0.5

Adding utility for worry =.95

“Men often, from infirmity of character, make their election for the nearer good, though they know it to be the less valuable”* *Mill JS. Utilitarianism. London: Routledge, 1871 Outcomes - Discounting

Aneurysm Example Aneurysm Example We said since life expectancy is reduced by 2/3, so instead of 35, it is = 35 *.333 = We said since life expectancy is reduced by 2/3, so instead of 35, it is = 35 *.333 = However, are all years considered equal? However, are all years considered equal? Consider: Favorite Meal Consider: Favorite Meal Extreme Pain Lifetime Income

Outcomes - Discounting Generally, present > future Generally, present > future One common way to value the different times is discounting One common way to value the different times is discounting Essentially this year is worth δ more than next year Essentially this year is worth δ more than next year δ is commonly set at 0.03 or 3% δ is commonly set at 0.03 or 3% In order to compare values of all future times, a calculation, net present value, is often used In order to compare values of all future times, a calculation, net present value, is often used NPV = 1 / (1 + δ) t Where t is number of years in the future NPV = 1 / (1 + δ) t Where t is number of years in the future

Outcomes - Discounting Aneurysm Example Aneurysm Example If utility is 0.6 and life expectancy is 3 years If utility is 0.6 and life expectancy is 3 years NPV would be:  Utility / (1 + δ) t NPV would be:  Utility / (1 + δ) t However, when is year 1? However, when is year 1? Often, since events in year one occur on average half way through, we use 0.5 for year 1 NPV = 0.6 / (1.03) / (1.03) / (1.03) 2.5 NPV = 0.6 / (1.03) / (1.03) / (1.03) 2.5 NPV = 0.6 * {(1.03) (1.03) (1.03) -2.5 } NPV = 0.6 * {(1.03) (1.03) (1.03) -2.5 }

Outcomes - Discounting

Exponential Discounting Exponential discounting first described in 1937* Mathematically easy to manipulate Assumed discounting in “simple regular fashion” Does not differentiate difference between: Today vs. tomorrow Ten years vs. ten years plus one day *Samuelson PA. A Note on Measurement of Utility. Rev Econ Stud 1937;4:155-61

Discounting – Special Topic Think about your favorite dessert. Think about your favorite dessert. How much would you pay to have now? How much would you pay to have now? How much would pay to have tonight? How much would pay to have tonight? How much would you pay to have in 1 yr? How much would you pay to have in 1 yr? How much would you pay in 1 yr and 1 day? How much would you pay in 1 yr and 1 day?

Exponential Discounting Problems with the Model Discounting unlikely to be constant Anticipal effect is not demonstrated Difference in valuations appears greater when closer Discount reversal effects not incorporated Far future, prefer A to B Near future, prefer B to A

Discounting – Special Topic Solutions: Measure discount rates through life Could model with present-biased preferences Essentially, “today” versus all other time periods is valued higher for many outcomes Difference in future outcomes is likely similar

Present-Biased Preferences Described by: Phelps and Pollack in 1968* O’Donoghue and Rabin in 1999** Two parameter model***: β – the difference between today and “tomorrow” δ – the difference between all future time intervals Model accounts for Discount reversal effects Component of anticipal effects *Phelps ES, Pollack RA. On Second-Best National Saving and Game-Equilibrium Growth. Rev Econ Studies 1968;35: **O’Donoghue T, Rabin M. Doing it Now or Later. Amer Econ Rev 1999;89: *** Laibson D. Golden Eggs and Hyperbolic Discounting. QJE 1997;112:443-77

Exponential vs. PBP Exponential: U T = U P (outcome) + Σn δ n U P (outcome) Present-biased preferences: U T = U P (outcome) + β[Σn δ n U P (outcome)] U T is the total NPV utility U P is the moment to moment utility β gives difference between immediate and all other time periods, while δ is difference in the future

Discounting: Prescriptive vs. Descriptive We discount We discount But, should we But, should we Example - perceived time Example - perceived time

Overall Review Outcomes OutcomesMortalityMorbidity Measuring Utilities Measuring Utilities Visual Analog Standard Gamble Time Trade-off Quality Adjusted Life Years (QALYs) Quality Adjusted Life Years (QALYs) QALYs = time (years) x quality (utils) QALYs = time (years) x quality (utils) Discounting Discounting NPV =  Utility / (1 + δ) t