External Validation of A Mapping Algorithm To Estimate Eq-5d-3l Utilities From Oxford Knee Score Responses

• Published in: Value in health Vol. 20; no. 9; p. A537
• Main Authors: Trigg, A, Woodcock, F
• Format: Journal Article
• Language: English
• Published: Lawrenceville Elsevier Science Ltd 01.10.2017
• Subjects: Algorithms; Clinical outcomes; Cost analysis; Joint replacement surgery; Knee; Knee replacement; Mapping; Questionnaires; Regression analysis; Sensitivity analysis; Surgery; Validity
• ISSN: 1098-3015; 1524-4733; 1524-4733
• DOI: 10.1016/j.jval.2017.08.784

OBJECTIVES: A key strength of any model is its generalisability to samples other than the one from which it was developed. This analysis evaluated the external validity of a mapping algorithm for predicting EQ-5D-3L utilities from Oxford Knee Score (OKS) responses in a large sample of patients receiving knee replacement surgery. METHODS: Data were obtained from the UK Patient Reported Outcome Measures dataset (April 2011 - March 2015), yielding 332,235 complete, concurrent observations of EQ-5D and OKS from 177,132 patients. A response mapping algorithm developed by Dakin et al (2013) was evaluated, employing multinomial logistic regression to predict EQ-5D-3L item responses based on OKS responses, and estimating utilities based on an expected value approach. Predictive accuracy was assessed through calculating the mean squared error (MSE), mean absolute error (MAE) and the proportion of absolute errors >0.1. Performance across the range of observed OKS scores was also assessed. Additionally, an algorithm based on linear regression was evaluated for comparison. RESULTS: Pre- and post-operative EQ-5D-3L utilities respectively exhibited a bimodal and trimodal distribution (Mean[SD]=0.407[0.312] and 0.721(0.263]). The MSE and MAE of predicted versus observed EQ-5D-3L utilities were 0.033 and 0.129, respectively. Notably, these values were lower than those previously observed in the original development and validation of the algorithm, indicating improved predictive accuracy. 56.9% of predicted EQ-5D-3L utilities were within 0.1 of their true observed values. Reduced performance was observed in patients with OKS scores between 10 and 19. As expected, the algorithm based on linear regression had lower predictive accuracy (MSE=0.041, MAE=0.159). CONCLUSIONS: The findings of this study support the validity of this algorithm, outperforming the original estimation and validation results in terms of predictive accuracy. This algorithm can be used to estimate utilities and thus QALYs for cost-effectiveness analyses where only the OKS was administered, or within sensitivity analyses where both questionnaires were administered.