Background/Purpose: The LupusQoL is a measure of health-related quality of life (HRQoL) developed specifically to assess the impact of systemic lupus erythematosus (SLE) and its treatment. Whilst producing a profile of scores, it is not preference-based, and therefore does not provide utility values. Such utility estimates are necessary to calculate quality-adjusted life years in economic evaluations, which are increasingly needed to guide decisions on how best to allocate health care resources.

The SF-6D (derived from the SF-36/SF-12) is a generic preference-based HRQoL measure which does provide utility scores. Despite evidence to support the validity of the SF-6D in SLE, and growing need for utility data, many trials fail to include a preference-based measure. One solution is empirical mapping; estimating a statistical relationship between a non-preference based and a preference-based measure using a dataset in which both measures have been administered to the same patients.

We aim to derive and test a mapping algorithm to predict SF-6D utility scores from the LupusQoL.

Methods: LupusQoL and SF-6D data were collected from 320 people with SLE at rheumatology outpatient clinics at 7 UK centres. Ordinary least squares regression was used to estimate models of increasing complexity to predict individuals’ SF-6D scores from their LupusQoL responses. Model performance was judged on predictive ability through the size and pattern of prediction errors generated, using mean absolute error (MAE) and root mean squared error (RMSE) statistics. Performance of the selected model was externally validated on an independent data set of 113 females with SLE who had completed both the LupusQoL and SF-36. All patients met 4 or more of the ACR criteria for SLE.

Results: Mean age and disease duration of the estimation sample was 44.8 (SD 13.6) and 10.5 (SD 8.7) years respectively. The sample cover the full range of SF-6D scores, mean 0.615 (range 0.296-1.00). Figure 1 shows mean LupusQoL domain scores. Four LupusQoL domains (physical health, pain, emotional health, fatigue) were selected in the final model. The validation dataset were slightly older (mean age 48.6 (SD 9.3) years) with longer disease duration (mean 12.6 (SD 9.7) years), but had comparable range of SF-6D (0.327 – 1.00) and mean LupusQoL scores (Figure 1). Model fit was good in both the estimation data (R2 0.7219, MAE 0.0557, RMSE 0.0706) and when applied to the validation sample (R2 0.7431, MAE 0.528, RMSE 0.0663). The model predicts mean SF-6D utility in the validation sample extremely well (observed mean 0.624; predicted 0.617).

Conclusion: We provide a method by which utility values can be estimated from patient responses to the non-preference based LupusQoL, generalizable beyond the sample upon which it was estimated. Prediction errors, upon which mapping functions are primarily judged, were lower than those of published studies to date.

Figure 1: