Mapping of utilities is the statistical estimation of preference-based utility values from data collected via a non-preference-based instrument. This procedure is used when a study has not collected a suitable utility questionnaire, but utility values are required for a health economic model. It typically utilizes regression analyses and introduces additional uncertainty into the analysis.
From Study Instrument to Utility Value
In a clinical study, a disease-specific questionnaire can capture symptoms or functional capacity in a highly differentiated manner without providing a utility value on a preference-based scale. For mapping, a statistical relationship is estimated between these observed values and a preference-based target instrument. The resulting model is then applied to the non-preference-based study data to generate approximated utilities.
The estimation requires a suitable data basis in which predictors and target utilities were observed together. Crucial factors include the similarity of the population, disease severity, and measurement conditions between the development data and the application study. A model developed for a different disease or a significantly different level of severity can produce systematic errors, even if its regression metrics appear favorable on the development data.
Limits of Transferability
IQWiG considers mapping to be secondary to the direct elicitation of utilities. Regression cannot restore individual preferences; instead, it provides expected values based on observed questionnaire scores. Particularly at the ends of the utility scale or in states with severe side effects, estimated values may smooth the real distribution or underestimate relevant differences.
For a model, assumptions, coefficients, validation, and the consequences of the estimation must be disclosed. External validation demonstrates whether a mapping functions effectively in independent data. Sensitivity analyses can test how alternative equations or plausible error ranges affect QALYs and incremental results. Without such testing, it remains unclear whether an economic result depends on the chosen conversion formula.
Differentiation: Imputation and General Regression
Mapping does not replace missing individual values. Imputation addresses missing data within a variable, such as an unanswered questionnaire at a specific visit. In mapping, however, values are present but not in the required preference-based format; the regression translates them into a different health economic target metric.
While mapping is a use case for regression, not every regression analysis is a mapping. For example, a regression may investigate predictors for survival, disease progression, or resource consumption. The mapping of utilities has the narrow purpose of deriving approximated values for a utility scale from non-preference-based measurements.
Distributional properties play a special role in model selection. Utilities can have an upper value of 1 and can take on negative values; a linear equation may therefore generate predictions outside the permissible range. Alternative models or bounding procedures can address this problem but create new assumptions. Therefore, it is not sufficient to provide only a mean error value. What is relevant is whether the estimated values remain plausible in clinically important states.
Mapping can also distort differences between treatment groups if the non-preference-based scale weights certain symptoms differently than the target instrument. For an economic decision, it is precisely the incremental difference that counts, not just the quality of the prediction across all observations. Group-wise validations and transparent sensitivities help identify whether a seemingly minor prognostic problem significantly alters the QALY difference.
The report should clearly indicate which utilities were directly observed and which were estimated. Only then can readers assess the origin of the QALYs and the additional potential for error in the model calculation.
Directly elicited and mapped values should also not be mixed uncritically within a time series. A change in method can appear like a change in health status, even though it results only from a different valuation logic.
Relevance for clinical trials
The most reliable way to obtain robust utilities is the prospective collection of data using a suitable preference-based instrument. If mapping is nevertheless expected, the study instrument, concurrent measurements, and analysis populations should be planned so that the bridge to the target instrument can be reconstructed. The database must contain sufficient variation in symptoms and health states; otherwise, the equation extrapolates precisely where the economic decision may be particularly sensitive.
Full-service CROs like Mediconomics provide support in selecting instruments directly capable of providing utility values and in specifying a mapping analysis plan if direct measurement is not possible. Biostatistics can evaluate development and validation data, while data management ensures the harmonization of questionnaire variables, visit times, and analysis formats for a transparent model.
Frequently Asked Questions (FAQ)
Is mapping the preferred method for determining utilities?
No. IQWiG ranks mapping as secondary to the direct elicitation of preference-based utilities due to the additional uncertainty.
Does mapping fix missing answers in an EQ-5D questionnaire?
No. That would be a problem of missing data and thus the subject of an imputation strategy, not mapping.
Why must a mapping be externally validated?
This is the only way to check whether the relationship estimated in a development sample is sufficiently transferable to an independent target population.
Regulatory References
- IQWiG, General Methods Version 8.0 – evaluates mapping as secondary to direct elicitation.
- Section 35b SGB V – forms the legal framework for cost-benefit assessments.
- Rules of Procedure of the G-BA, Chapter 5 – regulates procedural aspects of drug assessment.