Generating EQ-5D-3L Utility Scores from the Dermatology Life Quality Index: A Mapping Study in Patients with Psoriasis

• Published in: Value in health 2018-08, Vol.21 (8), p.1010-1018
• Main Authors: Davison, Niall J., Thompson, Alexander J., Turner, Alexander J., Longworth, Louise, McElhone, Kathleen, Griffiths, Christopher E.M., Payne, Katherine
• Format: Article
• Language: English
• Subjects: Algorithms; Anxiety; Dermatology; Discomfort; Discriminant analysis; DLQI; EQ-5D; Exploratory factor analysis; Factor analysis; Interaction terms; Intervention; Mapping; Mental depression; Pain; Pharmacovigilance; Psoriasis; Quality of life; Questionnaires; Random effects; Same day; utility; Validity
• ISSN: 1098-3015; 1524-4733
• DOI: 10.1016/j.jval.2017.10.024

†List of members: Jonathan Barker, MD, FRCP; Marilyn Benham; David Burden, MD, FRCP; Ian Evans; Chris Griffiths, MD, FRCP; Sagair Hussain, PhD; Brian Kirby, MRCP; Linda Lawson; Kathy McElhone, PhD; Ruth Murphy, MB ChB, PhD, FRCP; Anthony Ormerod, MB, MD, FRCP; Caroline Owen, MBChB, FRCP; Nick Reynolds, MD, FRCP; Catherine Smith, MD, FRCP.To develop an algorithm to predict the three-level EuroQol five-dimensional questionnaire (EQ-5D-3L) utility scores from the Dermatology Life Quality Index (DLQI) in psoriasis. This mapping study used data from the British Association of Dermatologists Biologic Interventions Register—a pharmacovigilance register comprising patients with moderate to severe psoriasis on systemic therapies. Conceptual overlap between the EQ-5D-3L and DLQI was assessed using Spearman rank correlation coefficients and exploratory factor analysis. Six regression methods to predict the EQ-5D-3L index (direct mapping) and two regression methods to predict EQ-5D-3L domain responses (response mapping) were tested. Random effects models were explored to account for repeated observations from the same individual. Estimated and actual EQ-5D-3L utility scores were compared using 10-fold cross-validation (in-sample) to evaluate predictive performance. Final models were selected using root mean squared error, mean absolute error, and mean error. The data set comprised 22,085 observations for which DLQI and EQ-5D-3L were recorded on the same day. A moderate correlation was found between the measures (r = −0.47). Exploratory factor analysis showed that two EQ-5D-3L domains (pain/discomfort and depression/anxiety) were associated with all six DLQI domains. The best-performing model used ordinary least squares with DLQI items, age, and sex as explanatory variables (with squared, cubic, and interaction terms). A tool was produced to allow users to map their data to the EQ-5D-3L, and includes algorithms that require fewer variables (e.g., total DLQI scores). This study produced mapping algorithms that can generate EQ-5D-3L utility scores from DLQI data for economic evaluations of health interventions for patients with psoriasis.