SB

Health State Valuation and Preference-Based Measures (QALYs, DALYs, MAUIs)

Session Aims

  • Introduce approaches for measuring health in evaluations.

  • Define the Quality-Adjusted Life Year (QALY) and common derivation methods.

  • Present generic preference-based measures (Multi-Attribute Utility Instruments, MAUIs).

  • Highlight limitations of QALYs and measurement instruments.

Learning Objectives

  • Understand the assumptions and calculation of QALYs.

  • Distinguish QALYs from Disability-Adjusted Life Years (DALYs).

  • Explain preference-elicitation techniques: TTO, SG, VAS, DCE, BWS.

  • Identify available generic MAUIs, their development, and scoring.

  • Reflect on limitations and inter-instrument differences.

Economic Evaluation: Context

  • Requires measurement of both costs and benefits.

  • Benefit measurement varies by evaluation type; cost-utility analysis uses utility instruments to generate QALYs.

  • Recall 8 stages: Clarify question → Identify consequences → Quantify → Value → Analyse → Interpret → Decision-maker use.

Defining Health & HRQoL

  • WHO (1948): “A state of complete physical, mental and social well-being…”.

  • Distinction between health vs. broader quality-of-life aspects.

  • Health-related quality of life (HRQoL): QoL domains affected by health/treatment.

  • Measures differ in content; some labelled HRQoL but focus only on symptoms.

Measuring Health

  • Discrepancies often observed between clinician vs. patient reports.

  • Preference for self-report (e.g., SF-36); item generation increasingly patient-led.

Identifying Outcomes

  • Surrogate/Intermediate outcomes (e.g., HbA1c, viral load, vaccine uptake).

  • Final outcomes: mortality & morbidity (LYs, QALYs gained, DALYs avoided).

Example Instrument: SF-36

  • 8 domains, 36 items; each domain scored 0\text{–}100.

  • Domains, items, ranges: Physical Functioning (10, 10–30) … General Health (5, 5–25).

  • Two summary scores: Physical (PCS) & Mental (MCS).

  • Equal weighting assumption problematic for economic evaluation (e.g., vigorous vs. moderate activity limitation treated equally).

  • Trial example: mixed domain changes—unclear overall benefit without preference weighting.

Quality-Adjusted Life Years (QALYs)

  • Combine survival (quantity) and HRQoL (quality) into “years in full health”.

  • Anchors: 1=\text{full health}, 0=\text{death}, <0=\text{worse than death}.

  • Calculation: \text{QALY}=\sum{t} Qt\times T_t.

    • Example: 20 years on dialysis, weight 0.8 \Rightarrow 20\times0.8=16\text{ QALYs}.

    • 10 years at uitility weight 0.6 ⇒ 6\text{ QALYs}.

  • Graphical interpretation: area under quality–time curve.

Disability-Adjusted Life Years (DALYs)

  • \text{DALY}=\text{YLL}+\text{YLD}.

  • One DALY = 1 year of full-health lost; weights: perfect health 0, death 1.

  • Goal: minimise DALYs (or maximise DALYs averted).

  • Broad relation: \text{DALYs averted} \approx \text{QALYs gained}.

Expected QALYs Under Uncertainty

  • Treatments have probabilistic future utilities.

    • Example table (1-year horizon):

    • Treatment A: P(\text{cure})=0.70, P(\text{no change})=0.20, P(\text{adverse})=0.10.

    • Combine with utilities (1, 0.8, 0.3) to compute expected QALY.

Core Assumptions Behind QALYs

  1. Independence of quality (Q) and time (T): value unaffected by when/how long.

    • 5\text{ yrs}@0.9 = 6\text{ yrs}@0.75 = 45\text{ yrs}@0.1.

  2. Linearity in probability (risk neutrality).

    • 0.1\times10\text{ QALYs}+0.9\times0=1\text{ QALY}.

  3. Additive separability over time (no state interaction effects).

    • 3\text{ yrs}@0.2 + 5\text{ yrs}@0.8 = 5\text{ yrs}@0.8 + 3\text{ yrs}@0.2.

Preference-Elicitation Techniques

  • Time Trade-Off (TTO)

  • Standard Gamble (SG)

  • Visual Analogue Scale (VAS)

  • Ranking

  • Discrete Choice Experiment (DCE)

  • Best-Worst Scaling (BWS)

  • Person Trade-Off (not covered)

Time Trade-Off (TTO)

  • Sacrifice length for quality: choose between

    1. t years in state i then death.

    2. x<t years in full health then death.

  • Indifference ⇒ utility h_i = x/t.

    • Example: indifference at x=4, t=10 \Rightarrow h_i=0.4.

  • Variants: duration (10 yrs vs. life expectancy), administration mode, titration method, visual aids, context effects, response variable.

  • Advantages: choice-based, explicit Q–T trade-off, easier than SG probabilities.

  • Disadvantages: assumes constant proportional trade-off, ignores time preference, difficulty valuing ‘worse-than-dead’, comprehension issues.

Standard Gamble (SG)

  • Based on Expected Utility Theory (EUT).

  • Choice between certain state h_i and gamble: probability p of full health, 1-p of death.

    • Indifference ⇒ h_i = p.

    • Willingness to accept 10% death risk ⇒ h_i=0.9.

  • States worse than dead: swap positions; utility h_i= -p/(1-p) (bounded to -1 in practice).

  • Advantages: rooted in EUT, incorporates uncertainty, choice-based.

  • Disadvantages: probability comprehension, risk attitudes & loss aversion contaminate values, death framing concerns.

Visual Analogue Scale (VAS)

  • Rating thermometer 0-100 anchored at “worst/best imaginable health”.

  • For QALYs must anchor to death (0) and full health (100); recalibrate if necessary.

    • Calculation example: health state at 80, death at 20 ⇒ \frac{80-20}{100-20}=0.75.

  • Advantages: simple, cheap, high response.

  • Disadvantages: no sacrifice notion, end-aversion, spreading & context biases, yields ordinal data; mapping to TTO/SG possible but debated.

Discrete Choice Experiment (DCE) & Best-Worst Scaling (BWS)

  • DCE: respondents choose preferred state from pairs; modelled via Random Utility Theory (logit/probit), producing marginal utilities.

  • BWS: select best & worst attribute within a profile; extension of DCE.

  • Anchoring to 0–1 scale options: worst-state =0, hybrid with TTO/SG, include ‘dead’, mapping, logistic transformation with duration attribute.

  • DCE with duration overcomes ‘dead’ issue; closer to TTO concept.

  • Pros: potentially less cognitively demanding, online feasible.

  • Cons: no explicit sacrifice, possible heuristic use, modelling complexity, large samples.

Comparison of Valuation Methods

Criterion

VAS

TTO

SG

DCE

Choice-based

✗

✔

✔

✔ (if duration)

Cardinal scale

?

✔

✔

✔ (some models)

Economic theory

✗

✗

✔ (EUT)

✔ (RUT)

Includes uncertainty

✗

✗

✔

✗

Ease/Cost

High

Medium

Low

High (online=M)

  • Typical pattern: \text{VAS}<\text{TTO}<\text{SG} values.

Choosing a Technique

  • VAS often ruled out for utility generation.

  • SG & TTO historically dominant but interview-intensive.

  • Ordinal methods (DCE/BWS) gaining popularity; cost & practicality advantages.

Multi-Attribute Utility Instruments (MAUIs)

  • Provide off-the-shelf scorings; two components:

    1. Classifier (health description).

    2. Tariff (pre-scored utilities from general-population valuations).

  • Advantages: cheap, comparable across studies, accepted by NICE & others.

  • Disadvantages: may lack sensitivity/relevance; differing domains, levels, valuation methods cause divergent utilities.

Common Generic MAUIs

  • EQ-5D-3L / EQ-5D-5L

  • SF-6D (from SF-36)

  • Health Utility Index (HUI 1-3)

  • AQoL-8D

  • QWB, PROMIS-29, 15D (Scandinavia)

EQ-5D Structure

  • 5 dimensions: Mobility, Self-care, Usual Activities, Pain/Discomfort, Anxiety/Depression.

  • 3L: 3 severity levels ⇒ 3^5=243 states.

  • 5L: 5 levels ⇒ 5^5=3125 states.

EQ-5D Valuation Example (UK 3L tariff)

  • Regression on 3,395 general-population TTO interviews.

  • Utility for state 11223:

    • Constants & coefficients: -0.081, -0.036, -0.123, -0.236, -0.269 (N3 interaction).

    • U=1-0.081-0.036-0.123-0.236-0.269=0.255.

Other Instruments: Key Facts

Instrument

Valuation

Extrapolation

Sample

EQ-5D-3L

TTO+VAS

Regression

GP (many countries)

EQ-5D-5L

TTO+DCE

Regression

GP (growing)

SF-6D

SG

Regression

GP (UK, Japan…)

HUI3

VAS→SG

MAUT

GP (Canada, France)

AQoL-8D

VAS→TTO

MAUT

GP & patients (Australia, etc.)

Differences Between Instruments

  • Coverage: e.g., vision in HUI, not EQ-5D.

  • Severity range (floor/ceiling): SF-6D narrow range at worst health.

  • Sensitivity: number of levels (EQ-5D-3L coarse at top end).

  • Scaling: divergent bottom values; not perfectly comparable.

Child-Specific Utilities

  • CHU9D, HUI2, EQ-5D-Y (Australian 3L value set 2025).

  • Adjusted descriptors, age targets, valuation challenges.

Whose Values? Public vs. Patients vs. Experts

  • NICE, ICER, WHO favour general-population values (taxpayer perspective, veil of ignorance).

  • Patients: firsthand experience but potential adaptation/response-shift; sometimes unable (children, severely ill).

  • Empirical finding: Public values ↓ for physical states, ↑ for mental health compared with patients.

  • Explanations: different reference frames, adaptation, focusing illusion.

Instrument Choice Considerations

  • Suitability: domains, sensitivity, target population, intervention effects.

  • Practicalities: length, cognitive burden, licensing cost, administration mode.

  • Cross-cultural response variation observed (e.g., Asians report fewer problems).

Health State Measurement & Valuation Pathways

  1. Non-preference instruments (e.g., SF-36) → map to utilities.

  2. Direct elicitation (TTO/SG/DCE) → utility scores.

  3. MAUIs (generic or condition-specific) directly yield utilities.

Summary

  • QALYs enable comparison across diseases/interventions by combining quantity and quality of life.

  • Utility estimates derived via direct elicitation or pre-scored MAUIs; each method/instrument has strengths, weaknesses, and assumptions.

  • Core QALY assumptions (independence, linearity, additivity) are pragmatic simplifications enabling practical modelling.

  • Selection of valuation technique and instrument should consider theoretical soundness, cognitive burden, sensitivity, cost, and decision-maker guidelines.