Disutility

Disutility as a Utility Decrement and QALY Loss in Economic Models

In health economic models a disutility is a decrement, the amount by which a utility value falls when a harm is added to a person's health, such as an adverse event, a second chronic condition, an unpleasant treatment or the strain of caring for a sick relative. Decrements are usually how models turn these harms into lost quality-adjusted life years, so the baseline, the source of each value and the rule for combining several decrements can all change the result of a cost-utility analysis.

A decrement measured from a stated baseline

A disutility only has meaning relative to the state it is subtracted from. Models express it as the difference between a baseline utility and the utility with the harm present, applied for as long as the harm lasts.

$$d = U_0 - U_1$$

where $d$ is the disutility, $U_0$ is the utility of the baseline state without the harm, and $U_1$ is the utility with the harm, both from the same instrument and value set. Some studies report decrements as negative numbers.

$$\Delta Q = d \times \tau$$

where $\Delta Q$ is the loss of quality-adjusted life years and $\tau$ is the duration of the harm in years. A decrement measured from full health is a different quantity from one measured from a disease state, and moving it between them assumes the same absolute effect at any level of health.

Four kinds of disutility in a model

Decrements enter models for different reasons, and each kind raises its own questions about duration and evidence. Separating them shows where a model's quality-of-life losses come from.

Adverse events. Short episodes are usually a one-off loss equal to the decrement times the episode duration, weighted by the event probability; the adverse event rate page works through that calculation. Technical Support Document 12 (TSD 12) advises distinguishing acute events from chronic sequelae and treats decrements for grade 3 to 4 adverse events as particularly important to include.

Health states and comorbidities. A chronic condition, or a second condition alongside the main one, lowers utility for as long as the state lasts, which in a Markov model means every cycle spent in the state. People with more than one condition need a combination rule.

Treatment-related disutility. Injections, infusions, frequent dosing or monitoring visits can reduce quality of life in their own right. A systematic review of process utility by Brennan and Dixon found values for different drug delivery methods between 0.02 and 0.27. These decrements last as long as treatment does, so small values can add up over long horizons.

Caregiver and family spillover. Illness can lower the quality of life of carers and relatives. A systematic review by Wittenberg and Prosser found 15 studies, of which 12 found measurable spillover effects and three found none, and described the evidence as limited, pointing to a generally small, negative effect.

Estimating decrements by regression, mapping and vignettes

Decrements are estimated in three main ways, and the method shapes what the number means. A model should say how each value was obtained.

The first is regression analysis of EQ-5D index scores on indicators for each condition or event plus covariates such as age and sex, with the decrement read from each coefficient. The catalogue of EQ-5D scores for the United Kingdom by Sullivan and colleagues took this approach, applying UK preferences to EQ-5D responses from a United States household survey and estimating the marginal disutility of 135 chronic ICD-9 conditions with ordinary least squares, Tobit and censored least absolute deviations regressions.

The second is a mapping function, which predicts utility from another measure when no preference-based data were collected. The third is direct valuation of vignettes, descriptions of a health state with and without the harm, by time trade-off or standard gamble, with the decrement taken as the difference between the two valuations. A methods paper by Matza and colleagues notes that vignettes are often used when no standardised approach is feasible for a condition or treatment, and gives recommendations on developing, valuing and using them.

What the NICE manual asks of utility sources

For appraisals in England, section 4.3 of PMG36 sets out where utility values, and so decrements, should come from. EQ-5D measurement by patients, valued with public preferences, comes first (sections 4.3.1 and 4.3.6), mapping to EQ-5D is a departure from the reference case (section 4.3.9), and vignettes sit well down a hierarchy.

Vignettes appear in figure 4.1, the hierarchy referred to in section 4.3.12. When evidence shows the EQ-5D is not appropriate, the order of preference is another generic preference-based measure, a condition-specific preference-based measure, vignettes developed using the Decision Support Unit's best practice recommendations and valued by the general population with a method such as time trade-off, and then direct valuation of own health. When EQ-5D cannot be obtained from a study, the literature or mapping, the figure allows vignettes on which the general population or people with the condition complete the EQ-5D. A vignette-based decrement therefore needs a stated reason why the routes above it were not possible.

Section 4.3.7 states that adjustments for age or comorbidities may be needed, that a multiplicative approach is generally preferred and that the methods should be clearly documented. Section 4.3.17 asks evaluations to consider health effects for carers when relevant, with evidence that the condition substantially affects carers' health-related quality of life and of how the technology affects carers.

Combining decrements by additive, multiplicative and minimum rules

Utility data usually come from cohorts with single conditions, but people in a model often have more than one. TSD 12 sets out three conventional rules for estimating the combined state and notes that they can give very different values.

$$U_{AB}^{add} = U_{0} - (U_{0} - U_A) - (U_{0} - U_B)$$

where $U_{AB}$ is the estimated utility for people with both conditions A and B, $U_A$ and $U_B$ are the utilities of people with only one of them, and $U_0$ is the utility of comparable people with neither. The additive rule subtracts both absolute decrements.

$$U_{AB}^{mult} = U_{0} \times \frac{U_A}{U_{0}} \times \frac{U_B}{U_{0}}$$

where the ratios $U_A/U_0$ and $U_B/U_0$ are the proportional effects of each condition. Writing $d_A = U_0 - U_A$ and $d_B = U_0 - U_B$, the result equals $U_0 - d_A - d_B + d_A d_B / U_0$, so with two positive decrements it gives a slightly smaller loss than the additive rule.

$$U_{AB}^{min} = \min(U_{0}, U_A, U_B)$$

where $\min$ takes the lowest of the three values, so the combined state is assumed to be no worse than the worse single condition.

TSD 12, by Ara and Wailoo, reports that the evidence does not unequivocally support any one method and that a newer combination linear model, not yet validated, appeared to give more accurate results; of the three conventional methods, the multiplicative appeared the most accurate overall, although all produced some large errors. Its practical advice is that, when utilities from people with the combined conditions are not available, the multiplicative method should be used, with the multiplier estimated from age-adjusted data as a minimum. It applies the same advice to adverse event decrements that need adjusting, and asks submissions to justify including or excluding them. This is consistent with the preference for a multiplicative approach in section 4.3.7 of PMG36.

Adjusting the baseline for age and sex

The baseline matters as much as the decrements. TSD 12 states that assuming perfect health for people without a given condition is inappropriate, and records agreement that adjusting for age and gender is an absolute minimum. PMG36 section 4.3.7 adds that baselines extrapolated over long horizons should reflect the decline in quality of life seen in the general population and should not exceed general population values at a given age.

One published set of norms is the regression by Ara and Brazier, fitted to EQ-5D data from the 2003 and 2006 Health Survey for England, shown below.

$$U_{gp} = 0.9508566 + 0.0212126 \, male - 0.0002587 \, age - 0.0000332 \, age^2$$

where $U_{gp}$ is mean general population EQ-5D utility, $male$ is 1 for men and 0 for women, and $age$ is in years. The equation is shown for illustration and rests on 2003 and 2006 data. For a woman aged 60, $U_{gp} = 0.9508566 - 0.0002587 \times 60 - 0.0000332 \times 3600 = 0.8158$; at 80, $U_{gp} = 0.9508566 - 0.0002587 \times 80 - 0.0000332 \times 6400 = 0.7177$. A condition that removes 12.5% of baseline utility under the multiplicative rule costs about 0.102 at 60 and 0.090 at 80, whereas a fixed additive decrement costs the same at both ages. PMG36 asks for a recent source, so the norms used in a submission should be current and match the instrument and value set of the other utilities.

Worked example with two comorbidity decrements and three combination rules

The figures below are illustrative. A model follows people aged about 65 with condition A, and an intervention prevents some of them from developing condition B over five years. One EQ-5D dataset gives mean utilities of 0.80 for people of that age with neither condition, 0.70 with A alone and 0.60 with B alone. No data exist for people with both. Each avoided case of B would otherwise have started at the outset and lasted the full five years, and nobody dies.

1. Find the single-condition decrements. From the baseline, $d_A = 0.80 - 0.70 = 0.10$ and $d_B = 0.80 - 0.60 = 0.20$.

2. Estimate the combined utility by each rule.

$$U_{AB}^{add} = 0.80 - 0.10 - 0.20 = 0.50$$

where both absolute decrements are subtracted from the baseline.

$$U_{AB}^{mult} = 0.80 \times 0.875 \times 0.75 = 0.525$$

where 0.875 is 0.70/0.80 and 0.75 is 0.60/0.80. The minimum rule gives 0.60, the utility with B alone. The multiplicative value is 0.025 above the additive one, equal to $d_A d_B / U_0$, or 0.10 × 0.20 / 0.80.

3. Calculate the QALY gain per case of B avoided. A person who avoids B stays at 0.70. Over five years, undiscounted, the gain is $(0.70 - 0.50) \times 5 = 1.00$ QALY under the additive rule, $(0.70 - 0.525) \times 5 = 0.875$ under the multiplicative rule and $(0.70 - 0.60) \times 5 = 0.50$ under the minimum rule.

4. Show the effect on cost-effectiveness. At an illustrative cost of £20,000 per case avoided, the cost per QALY gained is £20,000 (additive), about £22,857 (multiplicative, 20,000 / 0.875) and £40,000 (minimum). At an illustrative threshold of £25,000 per QALY, the decision changes with the combination rule alone.

5. Check the baseline. Combining the raw utilities as if from full health would give a multiplicative estimate of $U_{AB} = 0.70 \times 0.60 = 0.42$ instead of 0.525, and a gain of $(0.70 - 0.42) \times 5 = 1.40$ QALYs per case avoided. The rule and baseline should therefore be reported, with the alternatives tested in scenario analysis.

Double counting and other errors with decrements

One error is subtracting a loss the base utility already contains. Utilities collected from patients on treatment may reflect the adverse events and treatment burden those patients had, so a separate decrement on top can count the same loss twice.

TSD 12 makes this point for adverse events. Decrements for grade 1 to 2 events need care because the cohort behind the main utilities may already include patients who had them. The same applies to comorbidities, since a general population baseline includes people with common conditions, and to treatment process decrements when the trial questionnaire was completed during treatment. Caregiver decrements need a different check, since carers' losses should stay distinguishable from patients' QALYs so a reviewer can see what was counted for whom.

Other errors are simpler. A decrement measured from full health and applied to a disease state can overstate the loss, as step 5 shows. Decrements taken from one regression are usually correlated, and TSD 12 states that correlations between health state utility values should be characterised in probabilistic sensitivity analysis using multivariate distributions where possible.

Disutility, health state utility and disability weights compared

Neighbouring quantities sit on similar scales but answer different questions. Keeping them apart avoids using a decrement where a level is needed, or the reverse.

A disutility is a utility quantity measured as a difference between two levels, and a disability weight belongs to a separate framework with its own valuation methods.

Sources

• Ara R, Brazier JE. Populating an economic model with health state utility values: moving toward better practice. Value in Health. 2010;13(5):509-518. Coefficients checked in the authors' deposited version, HEDS Discussion Paper 09/11, University of Sheffield.
• Ara R, Wailoo A. NICE DSU Technical Support Document 12: The use of health state utility values in decision models. Sheffield: Decision Support Unit, ScHARR, University of Sheffield; 2011. Executive summary, sections 2, 3.2, 4.2.1 and 5.1 to 5.3, and Appendix.
• Brennan VK, Dixon S. Incorporating process utility into quality adjusted life years: a systematic review of empirical studies. PharmacoEconomics. 2013;31(8):677-691.
• Matza LS, Stewart KD, Lloyd AJ, Rowen D, Brazier JE. Vignette-based utilities: usefulness, limitations, and methodological recommendations. Value in Health. 2021;24(6):812-821.
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; 2022, updated March 2026. Sections 4.3.1, 4.3.6, 4.3.7, 4.3.9, 4.3.12 (figure 4.1) and 4.3.17.
• Sullivan PW, Slejko JF, Sculpher MJ, Ghushchyan V. Catalogue of EQ-5D scores for the United Kingdom. Medical Decision Making. 2011;31(6):800-804.
• Wittenberg E, Prosser LA. Disutility of illness for caregivers and families: a systematic review of the literature. PharmacoEconomics. 2013;31(6):489-500.