Mapping FACT-Melanoma Quality-of-Life Scores to EQ-5D Health Utility Weights

• Published in: Value in health 2011-09, Vol.14 (6), p.900-906
• Main Authors: Askew, Robert L., MPH, Swartz, Richard J., PhD, Xing, Yan, MD, PhD, MS, Cantor, Scott B., PhD, Ross, Merrick I., MD, Gershenwald, Jeffrey E., MD, Palmer, J. Lynn, PhD, Lee, Jeffrey E., MD, Cormier, Janice N., MD, MPH
• Format: Article
• Language: English
• Subjects: Adult; Aged; EQ-5D; Female; functional assessment of cancer therapy; health-state utility; Humans; Internal Medicine; Male; melanoma; Melanoma - psychology; Melanoma - therapy; Middle Aged; Neoplasm Staging; Prospective Studies; Quality of Life; questionnaires; Reproducibility of Results; Socioeconomic Factors; Surveys and Questionnaires
• ISSN: 1098-3015; 1524-4733
• DOI: 10.1016/j.jval.2011.04.003

Abstract Objectives We sought to develop a mapping function from functional assessment of cancer therapy-melanoma (FACT-M) quality of life scores to the EuroQol-5D (EQ-5D) utility scores. Methods FACT-M and EQ-5D scores were collected during a prospective study of melanoma-related quality of life at a tertiary cancer care center in the United States. The study sample was divided into development and validation datasets with equal distributions by cancer stage and treatment status. Censored Least Absolute Deviation (CLAD) and Ordinary Least Squares (OLS) regression analyses were performed using the developmental dataset to derive mapping functions, and model performance was examined through comparisons of residuals and measures of fit in the validation dataset. Exploratory analyses examined the predictive ability of clinical factors and individual subscales. Results Of 273 patients, 75 were undergoing treatment with 198 in follow-up surveillance. Relatively even distributions were observed by melanoma stage: I/II (n = 102), III (n = 100), and IV (n = 71). OLS regression resulted in a mapping function of EQ-5D = 0.0037*FACT-M+0.2238 with an R2 0.499. CLAD regression resulted in a mapping function of EQ-5D = 0.0042*FACT-M+0.1648 with pseudo R2 0.328. When applied to the validation dataset, correlations between observed and predicted values resulted in identical coefficients (r = 0.824, P < 0.001). Though the mapping functions were similar, residuals were smaller at the 20th, 40th, and 60th percentiles using the OLS model. The CLAD derived mapping function resulted in smaller residuals only for patients whose EQ-5D = 1. Conclusions The OLS mapping function demonstrated better predictive ability and will facilitate the derivation of utilities when direct population preference measures are not available.