economics writing
lilianjie
ECON317Lecture10S2020.pdf
ECON 317 The Economics of Canadian Health Care
Lecture 10: Measuring Health Outcomes via QALY
January 28th, 2020
Version 1
Required Reading
• Section 3.5 (pages 17-25) of
• Torrance, G. W. (1986). Measurement of health state utilities for economic appraisal: a review. Journal of Health Economics, 5(1), 1-30. Retrieved from https://doi-org.ezproxy.library.uvic.ca/10.1016/0167-6296(86)90020-2
• The standard source on QALY measurement.
• Nord, E. Daniels, N. & Kamlet, M. (2009). QALYs: Some Challenges. Value in Health, 12(Supplement 1), 510-515. Retrieved from https://doi.org/10.1111/j.1524-4733.2009.00516.x
• A good discussion of Ex Ante vs Ex Post health, concerns for fairness and the variation in health state utilities depending on the valuation method used.
2
Optional Readings
• Soares, M. O. (2012). Is the QALY blind, deaf and dumb to equity? NICE's considerations over equity. British Medical Bulletin, 101(1), 17-31. Retrieved from https://doi.org/10.1093/bmb/lds003
• An excellent summary of how QALYs work, followed by a theoretical and practical discussion of the equity/efficiency trade-off in the context of the UK's National Health Service.
• Round, J. (2012). Is a QALY still a QALY at the end of life? Journal of Health Economics, 31, 521-527. Retrieved from https://doi.org/10.1016/j.jhealeco.2012.01.006
• The source for Lecture 10's discussion of end-of-life care, including the diagram.
• Menzel, P. T. (2014). Utilities for Health States: Whom to Ask in A. J. Culyer (Ed.), Encyclopedia of Health Economics, 417-423. Retrieved from https://doi.org/10.1016/B978-0-12-375678-7.00508-3
• Richardson, J., McKie, J. & Bariola, E. (2014). Multiattribute Utility Instruments and Their Use in A. J. Culyer (Ed.), Encyclopedia of Health Economics, 341-357. Retrieved from https://doi.org/10.1016/B978-0-12- 375678-7.00505-8
• Torrance, G. W., Feeny, D. & Furlong, W. (2001). Visual Analog Scales: Do They Have a Role in the Measurement of Preferences for Health States? Medical Decision Making, 21(4), 329-334. Retrieved from https://doi.org/10.1177/0272989X0102100408
• Discusses the use (and abuse) of Visual Analog Scales in detail.
3
Learning Objectives
• Understand the use and calculation of QALY.
• Understand the main benefits and limitations of each approach.
• Understand the choices and trade-offs to be made in deciding who to ask about the utility of health states, and how to ask about the same.
• Understand the difference between Ex Ante and Ex Post utility of health states, and the implications for the valuation of health outcomes.
4
How should we value health outcomes?
• Why do we want health care? • Because it helps people ‘get better’ (quality of life; utility measure) • Because it saves lives (years of life)
• Ideally, we’d like a way to attach a value to health outcomes that…
• Takes both time AND quality into account
• Plays nicely with math/statistical methods
• Can be compared across all kinds of treatments and illnesses
• Can be used by social planners to set priorities/allocate resources efficiently
• Health professionals have come up with a number of (imperfect) measures. Today we’ll look at the most common: QALY
5
Quality-Adjusted Life Year (QALY) • A measure of individual health that takes both lifetime and quality into account.
• 1 QALY = 1 year in full health. A year spent dead has a value of 0 QALY. Fates worse than death have QALY < 0.
• 𝑄𝐴𝐿𝑌 = σ𝑖=1 𝑡 𝑞𝑖, where t=time periods (years), and 𝑞𝑖 is the quality weight corresponding
to the state of health experienced in period i.
• Weights are obtained from individual preferences (more below).
• Independence: Quality of life is assumed to be path-independent, so health state qualities can be added as in the equation above.
• Interval Scale: a change of 0.1 (say) on the 0-1 scale has the same meaning no matter where it is. 0.1 0.2 = 0.6 0.7
• “A QALY is a QALY is a QALY.” (attributed to NICE) When aggregating population-wide, a QALY gained by one individual is equal to a QALY gained by anyone else.
• Depending on the study, QALY may or may not be time-discounted (so future gains are less valuable than immediate gains, similar to how interest rates and the time value of money work.)
6
QUALY and End-of-Life
• What about palliative care? Won’t a QALY measure of ‘worth’ automatically mean non-life-extending care is useless?
• No. QALY can increase as long as quality of life increases, and it’s possible a shorter lifespan can lead to increased QALY (see graph below)
• It is well known that as known death approaches, the marginal value of (life) time increases. QALY is therefore not appropriate, because ‘A QALY is a QALY is a QALY’. Even the UK’s NICE, which require the use of QALY, have implemented an end-of-life exception:
• In the past, the UK health system has not considered treatments that cost more than £30,000 per additional QALY cost effective (worth paying for).
• IF life expectancy < 2 years AND a treatment increases life by > 3 (expected) months THEN the usual £30,000/QALY cost threshold can be waived.
7
Lower lifetime, higher QALY? (Adapted from Round, 2012)
Treatment A
Treatment B
X
Y
Z
tA tB Time
Quality of Life QALY A = Y + X
QALY B = Y + Z
X > Z, so QALY A > QALY B
8
Some problems with the QALY
• Disutility of a health state is the same, no matter who the health state happens to: vision/mobility loss is just as bad in a teenager who could have been cured, as in a 90-year-old with no hope of cure.
• Different methods of obtaining weights lead to values that vary systematically, and it is not clear the weights truly represent utility.
• Society values equity (‘rule of rescue’, favouring more severe illnesses, ‘fair innings’, etc.). Using QALY as the only guide ignores this, and focuses only on utilitarian efficiency. (But note that equity weights can be obtained/added.)
• The QALY Trap: Life-years gained by the healthy are valued more than life-years gained by the chronically ill. If instead we say all life years gained are equal, then the value of curing a chronic illness must be zero.
• Ex-Ante ‘decision’ utilities are used for weights, but valuation of health states is different Ex-Post, when obtained from someone in the health state (more below).
9
Who to ask?
• Two groups to ask about health state utility: those who have experienced it, and those who have not.
• ‘Ex Ante’ or ‘Decision’ utility: asks non-patients to imagine what the health state would be like, and rank it. More representative of ‘the general public’, and used as standard practice.
• Convenient/efficient (can ask the same people about many health states at the same time), and ‘fair’ (no vested interest).
• ‘Ex Post’ or ‘Experience’ utility: asks patients what it is/was like.
• Experience utility is almost always higher than decision utility.
• Using Ex Post utility may reduce the value to the social planner of (say) finding a cure for paraplegia.
• The main reason for the difference is adaptation.
10
Adaptation (Menzel 2014)
• “Patients adapt to their condition.” A review found Ex Post ratings of 0.83 QALY for health states with Ex Ante ratings of 0.65 QALY. Why?
• Skill enhancement (e.g. learning to use a wheelchair)
• Adjusted choice of activities (jogging reading)
• Revision of life goals (busy career fulfilling home life)
• Heightened stoicism (increased patience/tolerance)
• Lowered expectations (ill is the new baseline normal)
• Altered conception of health (forget exactly how full, full health is)
• Changed frame of reference/status quo (consistent with prospect theory)
11
How to ask: Visual Analogue Scale • One of several approaches we’ll look at.
• Sometimes called the ‘thermometer’ approach.
• Take a ruler. Mark the top of the ruler ‘best health state imaginable’, and the bottom of the ruler ‘death’ or ‘worst state imaginable’ (states worse than death are possible).
• Ask people to draw a line on the scale corresponding to their rating of a health state.
• Benefits: Easy to administer.
• Drawbacks: people avoid the ends of the scale, ‘snap’ to calibration marks, tend to space multiple valuations evenly, etc.
• Also: ‘best imaginable’ is not comparable between individuals, not based on tradeoffs, difficult to deal with time discounting (later).
12
A simple (and flawed) visual analog scale
• Please indicate on the scale above how you feel today.
• Your turn: what are some problems with this scale?
13
Worst Health Imaginable
Dead Best Health Imaginable
How to Ask: Standard Gamble (SG)
• Let H be the health state we’re interested in evaluating. • Perfect Health = 1 and Death = 0. • Option 1: 1 with chance p and 0 (Instant Death) with chance (1 – p) • Option 2: H, with certainty, for life. • When the interviewee is indifferent between A and B, p = H. • p x U(Perfect Health) + (1 – p) x U(Death) = U(Health State) • p x 1 + (1 – p) x 0 = H p = H • Problem: SG ‘mixes’ health state utilities with risk aversion. • A risk-averse person may report p = 1 (or very near 1) • Possibly susceptible to framing, since risk aversion (or seeking) depends on
frame of reference and Gain or Loss domain.
14
Chronic condition better than death
15
Option 2
Outcome: H Chance: 100% Expected Utility: 1 x H
Option 1
Outcome: Full Health Chance: p Expected Utility: p x 1
Outcome: Instant Death Chance: (1 – p) Expected Utility: (1 – p) x 0
If both options are equal: Then their expected values are equal, so H = p
Chronic condition worse than death
16
Option 2
Outcome: Instant Death Chance: 100% Expected utility: 1 x 0
Option 1
Outcome: Full Health Chance: p Expected utility: p x 1
Outcome: H Chance: (1 – p) Expected utility: (1 – p) x H
If both options are equal: Then (1 – p)H + p = 0 H = p/(p - 1)
This can get very negative, so in practice most researches enforce a lower bound of H = -1.
Temporary health condition
17
Option 2
Outcome: H Chance: 100% Expected Utility: H
Option 1
Outcome: Full Health Chance: p Expected Utility: p x 1
Outcome: S Chance: (1 – p) Expected Utility: (1 – p) x S
If both options are equally preferred: Then p + S x (1 – p) = H
‘Instant Death’ is overkill when talking about toothache, so we use a temporary state with ‘known’ utility, S, instead.
How to Ask: Time Trade-Off (TTO) • H, Perfect Health = 1, Death = 0 as in the discussion of the SG.
• Option 1: Survive for B<A years in perfect health, followed by death.
• Option 2: Survive for A years in health state H, followed by death.
• A is fixed, B is moved around to find indifference. When Alt 1 and Alt 2 are equally preferred…
• A x U(Health State) + U(Death) = B x U(Perfect Health) + U(Death)
• A x H + 0 = B x 1 + 0 H = B/A
• If H is considered worse than death, there’s a slight modification:
• B x U(Health State) + (A – B) x U(Perfect Health) = U(Death)
• A is kept fixed, and B is moved around to find indifference.
• BH + (A – B) = 0 H = 1 – A/B
• TTO has similar assumptions to QALY, and can be vulnerable to time discounting effects.
• (Mixes time discounting with health state utility, just as SG mixes risk aversion/seeking.) 18
Chronic condition better than death
19
Valuation
TimeA
H
B
Full Health = 1
Death = 0
Area = B x 1 = B
Area = A x H = AH
If both choices are equally preferred, then B = AH H = B/A
Option 1: B years of full health Option 2: A years of state H
Chronic condition worse than death
20
Valuation
Time
A
H
B
Full Health = 1
Death = 0
Area = B x 1 = B
Area = (A – B) x H
New choices are needed:
Option 1: Death Option 2: B years of full health, followed by (A – B) years of state H. (Sometimes the order is flipped)
If both choices are equally preferred… B + (A – B) x H = 0 H = B/(B – A) < 0
As before, this is usually capped at -1.
Temporary Health State
21
Valuation
TimeA
H
B
S
Death = 0
Area = A x S
Area = (A – B) x 1
If both choices are equally preferred, then B = AH H = B/A
As before, S is a temporary state with known utility.
Now we assume B periods of H are followed by (A – B) of full health.
Full Health = 1
Area = B x H
If both choices are equally preferred:
A – B + BH = AS H = 1 - A/B x (1 – S)
Option 1: A years of S Option 2: B years of H, and (A – B) years of full health
Multiattribute Approaches • Multiattribute Utilities (MAU) allow us to approximate a calculation of
utilities for more complicated health states, in which more than one attribute is important.
• These utilities, appropriately scaled, can be used as QALY weights. • Each MAU consists of a health questionnaire, and a formula that maps
each possible combination of answers (health state) into an utility value. • The mapping formula may, for example, use weights from SG or TTO
evaluations of key health states. Linear regression is used to find weights for individual attributes.
• Why not use TTO/SG for all health states? Some are not plausible (e.g. no trouble moving, confined to bed), while the number of possible states in some MAU may be unwieldy.
• The EQ-5D, one of the most popular MAU, has 243 possible health states.
22
The EQ-5D (Required in UK, common elsewhere)
• Rates 5 Dimensions of Health: Mobility (MOB), Self-Care (CARE), Usual Activities (ACT), Pain/Discomfort (PAIN) and Anxiety/Depression (DEP).
• Disutility of each can be rated is none/moderate/severe (1/2/3).
• 243 possible health states, written as 5-digit sets, e.g. (1,1,3,1,2)
• Each of the 10 possible non-’none’ answers (PAIN2,CARE3,MOB2, etc.) is assigned a weight. (Weight for all ‘none’ answers is 0.)
• Weights for key health states (42 or 17 of 243) are obtained via TTO. Weights for each answer are derived via linear regression.
• To obtain the total utility score for a health state, the weights for each answer are added together and subtracted from 1 (perfect health).
• e.g. U(1,1,3,1,2) = Weight for ACT3 + Weight for DEP2
23
FYI: The ED-5D Equation (Richardson/McKie/Bariola)
• You do not need to memorize this formula.
• In the formula, a variable such as MOB2 is equal to 1 if the answer is included in the health state, and 0 if it is not.
• ANY(A) = 1 if any answer =/= 1, 0 otherwise.
• ANY(B) = 1 if any answer = 3, 0 otherwise.
• Utility = 1 – [MOB + CARE + ACT + PAIN + DEP + ANY]
• Utility = 1 – [(.069MOB2 + .314MOB3) + (.104CARE2 + .214CARE3) + (.036ACT2 + .094ACT3) + (.123PAIN2 + .386PAIN3) + (.071DEP2 + .236DEP3) + (.081ANY(A) + .269ANY(B)]
24
