Model uncertainty and mapping from stroke handicap scales to the EQ-5D

Mapping or crosswalking involves the development of mathematical algorithms to predict healthrelated utility values using responses to some other measure of health. In the absence of
prospectively collected generic preference-based measures of health-related quality of life
(HRQoL), mapping methods may be used to predict preference based utility values from
disease-specific scales to inform resource allocation decisions. Simple linear or generalised
linear regressions used to devise mapping algorithms are often used with little or no
consideration of model uncertainty. Reporting a single “best” or “appropriate” model is typical of
this literature as is the ad hoc inclusion of potential confounding variables in a single model.

We examine how Bayesian variable selection and model averaging can be used to incorporate
model uncertainty into inferences about parameters and prediction when assessing the association (if any) between the EQ-5D and the Modified Rankin Scale (mRS), a simple and
widely used means of grading patient handicap following stroke. We address “adjustment
uncertainty” or uncertainty about which variables should be included in the mapping model
using a novel approach called Bayesian adjustment for confounding (BAC).

External validation of an existing mapping algorithm from the mRS to EQ-5D will be presented.
We will compare the performance of traditional mapping studies that condition inference on a
single model with alternative specifications using the BAC approach. The potential for
generating and applying misleading mapping algorithms to predict EQ-5D scores for individuals
following a stroke given their degree of handicap is illustrated.

For further information please contact l.k.watson@exeter.ac.uk

Parking at St Lukes is extremely limited and restricted to University of Exeter permit holders. Parking attendants patrol regularly. Where possible, please consider alternative options when planning your visit.