A review of studies mapping (or cross walking) non-preference based measures of health to generic preference-based measures

Abstract

HHe ea al lt th h E Ec co on no om mi ic cs s a an nd d D De ec ci is si io on n S Sc ci ie en nc ce e D Di is sc cu us ss si io on n P Pa ap pe er r S Se er ri ie es s No. 08/02A review of studies mapping (or cross walking) from non-preference based measures of health to generic preference-based measures INTRODUCTIONA common approach to assessing the outcomes of health care is to obtain patient reported descriptions of health status across various dimensions and then to apply a standardised numerical scoring system. There are many different measures of health, including several hundred condition specific measures of health designed for use in specific medical conditions or groups of condition (Spilker et al, 1990), and a number of generic measures designed to cover the core dimensions of health that are relevant across all medical conditions. Health measures can also be distinguished in terms of whether they generate a profile of dimension scores or a single index and if they produce a single index, whether or not the index has been derived using simple summation of item scores or by using preference weights obtained from patients or the general public (known as preference-based measures or multi-attribute utility scales).Patient reported measures of health have become widely used in clinical trials as primary or secondary outcomes. There is little agreement on which specific instruments should be used for this purpose. For assessing clinical efficacy, there is disagreement on whether to use a generic or condition specific measure, and between condition specific measures there is often disagreement amongst clinical researchers on the most appropriate instrument. As a result clinical trials around the world often use different measures for the same patient groups. This presents a substantial barrier to the synthesis of evidence.Preference-based measures of health are necessary to generate the health state utility values required to calculate QALYs for assessing the cost effectiveness of interventions.These are usually based on generic instruments (e.g. EQ-5D) that permit comparisons between patient groups, though there are examples of condition specific preference-based measures . Even for assessing clinical effectiveness, it could be argued that a preference-based index is necessary to deal with trade-offs made between outcomes. There has been a debate amongst health economists about the most appropriate preference-based generic measure to use in cost effectiveness analyses. Whilst the EQ-5D is the most widely applied in recent years (Brooks et al, 1996), the HUI3 (Feeny et al, 3 2002), QWB (Kaplan and Andersen, 1988), SF-6D (Brazier et al, 2002) and others continue to be used. However, different preference-based measures have been shown to generate different values on the same sample of patients (Marra et al, 2005; Feeny et al, 2004;Barton et al, 2004). Furthermore, many key clinical trials on the efficacy of new interventions do not have a generic measure and the recent FDA Guidance on using Patient Rep...

The MAPS Reporting Statement for Studies Mapping onto Generic Preference-Based Outcome Measures: Explanation and Elaboration

“…That review article excluded mapping algorithms targeted at other generic preference-based outcome measures that can generate health utilities, such as the SF-6D [46] and the Health Utilities Index (HUI) [47], which have been the target of numerous other mapping algorithms (e.g. [1,42,[48][49][50][51][52]). Moreover, the popularity of the mapping approach for estimating health utilities is unlikely to wane given the numerous contexts within health economic evaluation where primary data collection is challenging.…”

“…Statistical models used in mapping studies can be categorised into ''direct'' methods (onto index or utility scores) and ''indirect'' or ''response mapping'' methods (onto dimensions responses). There are clear advantages and disadvantages of each method and of different estimators, but clear guidance about which to use in different circumstances is lacking [1]. Therefore, authors should provide sufficient information about their modelling approach(es) so that readers can assess the robustness of their overall estimation strategy.…”

“…mean absolute error (MAE) or mean squared error (MSE)] rather than measures of model fit (e.g. R 2 or information criteria) [1]. The ideal measure of model performance would take account of the distribution of predictions as well as point estimates; since no single measure capturing all aspects of prediction accuracy has yet been developed, many authors select models using multiple measures or criteria (which may include face validity and/or parsimony).…”

Correction: Scale Characteristics and Mapping Accuracy of the US EQ-5D, UK EQ-5D, and SF-6D in Patients with Rheumatoid Arthritis

“…A wide variety of predictor variables have been mapped in different illnesses 34 . In rheumatic diseases, the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) has been mapped to the UK EQ-5D 35,36 and the HUI-3 32 .…”

Health state utility instruments compared: inquiring into nonlinearity across EQ-5D-5L, SF-6D, HUI-3 and 15D

“…Head-to-head comparisons were identified in 392 studies, suggesting low level of agreement, and hence a need for developing algorithms to compare utility scores across the various instruments used. Crosswalks or 'mappings' between a disease specific instrument and a HSU-instruments are commonly used to enable QALY estimations [6,7]. However, while such crosswalks between HSU-instruments are rare [8][9][10][11], previous studies have indicated non-linear relationships [8,9,11].…”