Health State Valuation

Health State Valuation: Concept Architecture

Orientation and learning roadmap

Health state valuation converts preferences about health into numerical values that can support comparison across conditions and interventions. This page separates health description from valuation, explains the anchoring scale and major elicitation methods, and follows the evidence through sampling, modelling, quality control, value-set selection, and QALY calculation. It also makes visible the normative choices about whose preferences count and how states worse than dead are represented.

A health state must be described before it can be valued

Valuation begins with a defined state that respondents can understand and compare. The state may come from a generic preference-based measure, a condition-specific descriptive system, a vignette, or a classification derived from patient-reported outcomes. The wording, dimensions, severity levels, duration, and context determine what respondents are actually valuing.

Measurement and valuation are different stages

Patients often report their own health using a questionnaire, while a separate valuation study assigns preference weights to the states described by that questionnaire. A person’s descriptive profile is therefore not itself a utility value. The value emerges only after applying a relevant value set or a justified direct elicitation method.

The QALY scale needs explicit anchors

For conventional QALY use, full health is commonly assigned 1 and dead is assigned 0. States considered worse than dead can receive negative values, while most standard value sets place full health at the upper anchor. The anchors define the scale but do not imply that preference intervals are free from methodological or normative assumptions.

Values, utilities, and preference weights are related terms

Health state values are often called utilities, preference weights, utility weights, or index scores in applied work. Strictly, “utility” may imply a choice under uncertainty, as in standard gamble, while time trade-off and discrete-choice approaches rely on different behavioral frameworks. Reports should identify the elicitation method rather than assuming all numbers on a 0-to-1 scale are interchangeable.

Time trade-off values health against life duration

Conventional time trade-off asks a respondent to choose between living a longer time in an impaired health state and a shorter time in full health, followed by death in both alternatives. At indifference, the full-health duration (x) relative to the duration (t) in the impaired state gives the value for a state considered better than dead. The basic calculation is:

[ u(h)=\frac{x}{t}. ]

Worked time trade-off example

Suppose a respondent is indifferent between ten years in health state (h) and 7.5 years in full health, with death occurring after each option. Under the conventional linear QALY assumptions, the health state value is 0.75. Three years spent in that state would contribute 2.25 undiscounted QALYs:

[ u(h)=\frac{7.5}{10}=0.75, ]

[ \operatorname{QALYs}=0.75\times3=2.25. ]

Composite time trade-off accommodates worse-than-dead states

Conventional TTO becomes difficult when respondents prefer immediate death to the impaired state. Composite time trade-off combines conventional TTO for states better than dead with a lead-time approach for states worse than dead, using a standardized task and transformation. The exact protocol, lead time, iteration procedure, interviewer training, and quality controls must be reported because they affect comparability.

Standard gamble introduces risk explicitly

Standard gamble asks respondents to compare living in a certain impaired state with a gamble between full health and immediate death. At indifference, the probability of full health provides the state value under expected-utility assumptions and the exact task design. Responses can reflect risk attitude, probability weighting, and difficulty understanding small probabilities as well as preferences for the health state.

Visual analogue scales provide ratings, not choices

A visual analogue scale asks respondents to place states on a line between labeled endpoints. It is comparatively simple and can help describe ordinal severity, but it does not require a trade-off involving duration or risk. VAS ratings should not automatically be treated as choice-based utilities suitable for QALYs without a justified transformation and decision context.

Discrete-choice experiments reveal relative preferences

In a health-state DCE, respondents choose between states described by different attribute levels, allowing a random-utility model to estimate relative preferences. A DCE without duration or another anchor identifies utility on a latent relative scale, not automatically on the dead-to-full-health QALY scale. Duration, dead, hybrid TTO information, or another defensible anchoring strategy is required to generate QALY-compatible values.

Different elicitation methods can produce different values

TTO, standard gamble, VAS, DCE, and hybrid approaches impose different tasks and assumptions. Differences can arise from risk attitude, time preference, scale use, framing, cognitive burden, anchoring, and modelling rather than from random error alone. Combining values across methods requires a clear rationale and sensitivity analysis.

The valuation perspective is a policy choice

Some decision frameworks prefer values from a representative general-population sample because public resources are allocated on behalf of the public. Patients may understand lived experience, adaptation, and treatment burden differently, while caregivers and clinicians bring other perspectives. The relevant perspective should follow the decision context and be stated, not selected after inspecting favorable results.

Experience with illness can change preferences

People living with a condition may adapt, recalibrate expectations, or know aspects of the state that brief descriptions fail to convey. Members of the public may focus more on unfamiliar losses or fear, while patients may incorporate coping and context. Neither perspective is automatically correct for every purpose, and disagreement is substantively informative.

Sampling determines whose values enter the value set

A valuation sample should represent the population whose preferences the decision framework intends to use. Coverage, recruitment, nonresponse, digital access, language, age, education, health literacy, and exclusion criteria can shift the resulting values. Weighting may improve alignment on observed characteristics but cannot recover perspectives that were never measured.

Children and proxy perspectives need special design

Valuing child health states raises questions about respondent age, whose perspective is adopted, family effects, developmental stage, and how duration and dead are presented. Proxy respondents may report or value a child’s health differently from children themselves or the general public. Adult protocols should not be transferred to children without methodological and ethical justification.

Duration and timing assumptions shape the task

TTO and DCE-with-duration methods commonly assume that the value of a health state combines with time in a sufficiently stable way. Preferences can violate constant proportionality, and temporary states may be valued differently depending on timing, sequence, prognosis, and adaptation. Duration should be treated as part of the estimand rather than a neutral survey detail.

Framing and presentation can affect responses

State order, dimension order, labels, examples, color, interviewer behavior, iteration paths, and the presentation of dead can influence valuation. Standardized scripts, interviewer training, practice tasks, randomization where appropriate, and accessible risk communication reduce avoidable variation. Protocol fidelity is essential when values will inform national or jurisdictional decisions.

Quality control should distinguish difficulty from invalidity

Valuation tasks are cognitively demanding, so inconsistent orderings, rapid completion, clustering at anchors, and non-trading responses deserve investigation. These patterns can reflect strong preferences, misunderstanding, protest, fatigue, interviewer effects, or data-quality problems. Exclusion rules should be prespecified, reported, and examined in sensitivity analyses rather than used solely to improve model fit.

Value-set modelling predicts states that were not directly valued

Descriptive systems can define hundreds or thousands of health states, making direct valuation of every state impractical. Statistical models use observed valuations for a designed subset to estimate values for the complete classification. Model specification should reflect the experimental design, repeated responses, censoring or heteroskedasticity, respondent heterogeneity, and logical consistency where relevant.

Hybrid models combine complementary information

Hybrid valuation models can combine cTTO observations, which provide direct anchoring, with DCE choices, which can estimate relative differences efficiently. The data sources operate on different scales and require an explicit linking model. More observations do not guarantee better values if the scale relationship or behavioral assumptions are misspecified.

Logical consistency is important but not purely mechanical

A more severe state would ordinarily be expected to receive no higher value than an otherwise identical milder state. Observed violations may arise from sampling error, model flexibility, interactions, poor comprehension, or genuine non-additive preferences. Constraining a value set can improve monotonicity but may also conceal evidence that the descriptive or statistical model needs revision.

Negative values have a defined interpretation

A negative health state value means the state was valued below dead under the elicitation and modelling framework; it does not mean a person has negative worth. The lower bound varies across protocols and value sets, so equal-looking negative numbers may not be directly comparable. Communication should separate valuation of a health state from valuation of a person’s life.

Value sets are instrument-, version-, and jurisdiction-specific

A value set is tied to a particular descriptive system, version, valuation protocol, population, jurisdiction, and model. A value for one EQ-5D version, country, or elicitation method should not be applied as though it were the native value set for another. Analysts should use the value set required by the relevant decision body or justify the closest alternative transparently.

Mapping is not direct valuation

When a preferred preference-based measure was not collected, mapping may predict its utility values from another outcome measure or clinical variables. Mapping adds model uncertainty and can compress extremes or perform poorly outside the estimation sample. Mapped utilities should be labeled clearly and validated for the target population and decision use.

QALYs combine values with time

In a model divided into intervals, QALYs are accumulated by multiplying the health state value in each interval by the time spent in that state and summing across intervals. Timing corrections and discounting are applied consistently with the model, not embedded arbitrarily in the value set. A discrete approximation is:

[ \operatorname{QALYs}=\sum_{t=1}^{T}u_t\Delta t_t. ]

Uncertainty in values belongs in the economic model

Value sets contain sampling and model uncertainty, while patient-level utility data contain measurement and sampling variation. Probabilistic sensitivity analysis should represent material uncertainty without treating correlated coefficients as independent. Structural uncertainty about instrument, perspective, mapping, or alternative value sets often requires scenario analysis.

Equity concerns extend beyond the average value

Average general-population values can conceal systematic differences by disability, age, culture, socioeconomic position, or lived experience. Using preferences in allocation also raises normative questions about adaptation, discrimination, and which aspects of health are included in the descriptive system. These issues should be discussed alongside, not hidden inside, the numerical value.

Excel can audit value application

A transparent workbook can map health-state codes to the correct value-set table, multiply values by time, and flag unmapped or duplicate codes. If utility is in B2 and years in C2, undiscounted QALYs are =B2*C2; lookup formulas should use exact matching and a visible error check. Excel should not be used to recreate a complex national value-set model from rounded published coefficients without full specification and validation.

A disciplined valuation and application workflow

Reliable health state valuation requires alignment among the descriptive system, respondents, elicitation protocol, model, and intended decision framework. The audit trail should preserve raw responses, exclusions, transformations, and the final scoring algorithm. A practical sequence is:

1. Define the decision use. Specify the jurisdiction, population, perspective, instrument, and required anchors.
2. Describe the states. Confirm dimensions, levels, wording, duration, and accessibility for respondents.
3. Select the elicitation method. Match TTO, cTTO, standard gamble, VAS, DCE, or a hybrid to the estimand.
4. Design the sample. Address representativeness, language, literacy, inclusion, and subgroup requirements.
5. Standardize fieldwork. Train interviewers, provide practice, monitor protocol fidelity, and protect respondent welfare.
6. Apply quality controls. Prespecify flags and evaluate exclusions through sensitivity analysis.
7. Estimate the value set. Account for design, repeated responses, scale, heterogeneity, and prediction uncertainty.
8. Validate the model. Examine prediction, ordering, logical consistency, robustness, and external comparisons.
9. Apply values exactly. Match instrument version, state codes, jurisdiction, time units, and QALY conventions.
10. Report normative choices. State whose preferences were used and which judgments remain contested.

Common valuation and application errors

Most serious errors come from treating all health-related scores as interchangeable utilities. Clear identification of the instrument, value set, anchors, perspective, and elicitation method prevents many mistakes. Common failures include:

• Treating a descriptive health profile as though it were already a utility value.
• Applying a value set from the wrong instrument version or jurisdiction without justification.
• Treating VAS ratings or unanchored DCE coefficients as QALY weights.
• Ignoring values below zero or recoding them to zero without justification.
• Mixing patient and public preferences without stating the decision perspective.
• Assuming full health and best described health are always empirically identical.
• Using mapped utilities without reporting mapping uncertainty.
• Ignoring interviewer, framing, nonresponse, or exclusion effects.
• Multiplying values by time expressed in inconsistent units.
• Interpreting a negative state value as a judgment about a person’s worth.

What a complete valuation record should disclose

A reproducible record identifies the descriptive system and version, sampled states, respondent population, perspective, recruitment, elicitation protocol, duration, anchors, experimental design, fieldwork controls, exclusions, statistical model, uncertainty, validation, and scoring algorithm. Applied analyses should name the exact value set, jurisdiction, publication, mapping method if any, and QALY timing conventions. This evidence allows reviewers to distinguish a defensible preference weight from an unlabeled quality-of-life score.