Reliability and validity of the EQ-5D-5L compared to the EQ-5D-3L in patients with osteoarthritis referred for hip and knee replacement

To assess the test-retest reliability of the EQ-5D-5L (5L) and compare the validity of the 5L and EQ-5D-3L (3L) in osteoarthritis patients referred to an orthopaedic surgeon for total joint replacement. We mailed questionnaires to 306 consecutive patients following referral and a second questionnaire after 2 weeks to assess reliability. Questionnaires included the 5L, EQ-VAS, Short Form-12, Oxford hip and knee scores, pain VAS, and the 3L. We compared the ceiling effect, redistribution properties, convergent and discriminant validity, and discriminatory power of the 5L and 3L. We obtained 176 respondents (response rate 58 %), 60 % female, 64 % knee patients, mean age 65 years (SD 11), with no significant differences between responders versus non-responders. Intraclass correlation coefficients were 0.61-0.77 for the 5L dimensions and 0.87 for the 5L index. For the 3L, most patients used level 2 (some/moderate problems) for mobility (87 %), usual activities (78 %), and pain/discomfort (71 %). In comparison, 5L responses were spread out with only 52, 42, and 50 %, respectively, using the middle level. All convergent validity coefficients were stronger with the 5L (Spearman coefficients 0.51-0.75). Absolute informativity (Shannon's index) showed higher results for all dimensions of the 5L compared with the 3L (average difference 0.74). Relative informativity (Shannon's evenness index) showed an increase from the 3L to the 5L in mobility, usual activities, and pain/discomfort. The 5L provided stronger validity evidence than the 3L, especially for dimensions relevant to this patient population-mobility, usual activities, and pain/discomfort.