Predicting utility values in low vision: An estimation from NEI-VFQ 25

Nalin Payakachat, Purdue University

Abstract

Understanding the relationship between the disease-specific health status measurement (the National Eye Institute Vision Function Questionnaire, NEI-VFQ 25) and the health utility index derived from the EQ-5D will inform researchers of which aspects of vision functioning they should focus on to improve overall quality of life. To date, no model has explained these relationships. The algorithm that can produce utility estimates from the NEI-VFQ 25 that are reasonably consistent with the EQ-5D should be a useful addition to the tool kit of health economists. In this study, different mapping approaches were used to identify which better predict the relative importance of the NEI-VFQ 25 dimensions on the health utility index. Ordinary least squares (OLS), Tobit, and censored least absolute deviation (CLAD) approaches were compared using cross-sectional data (n = 154) at screening from a phase I/II clinical trial in patients with age-related macular degeneration. To evaluate the predictive accuracy of the model, the mean absolute prediction error (MAPE) and other criteria were calculated using in-sample cross validation and out-of-sample validation. Three models were specified: full, short, and reduced models. The full model included all 12 dimensions of the NEI-VFQ 25 while the short model included only the General health and the composite scores. The reduced model was identified using a stepwise regression. The data sets used to conduct in-sample and out-of-sample validation were combined to observe whether the pattern of results remains consistent. The results from in-sample cross validation and out-of-sample validation were consistent in that OLS with heteroscedastic adjustment produced the lowest MAPE when compared to other approaches for the full model, while CLAD performed the best for the short model. CLAD for the short model produces predicted values closest to the observed ones. Results from the combined data sets showed that CLAD produces the lowest MAPE which was consistent with the initial results. The NEI-VFQ 25 Mental Health and Dependency dimension most powerfully impact the health utility index. The short model implies that a 25-point rise in the composite scores equates to a 0.092 rise in the EQ-5D index score. The short model is the best for predicting purposes when only the NEI-VFQ 25 data are available.

Degree

Ph.D.

Advisors

Summers, Purdue University.

Subject Area

Pharmaceutical sciences

COinS