Performance of EQ-5D-3L to EQ-5D-5L Crosswalk and Reverse Crosswalk in Lumbar Spine Surgery Patients.

Abstract

As registries transition from the EQ-5D-3L to the EQ-5D-5L, maintaining continuity of patient-reported outcomes is critical. Crosswalk algorithms were developed to convert between versions, but their performance in spine surgery patients has not been validated. To externally validate the EuroQol crosswalk and reverse crosswalk algorithms in patients scheduled for lumbar spine surgery. Single-center, observational cohort study using prospectively collected registry data from a tertiary spine center in Denmark. A total of 766 adult patients with lumbar spine conditions who completed both EQ-5D-3L and EQ-5D-5L at baseline within three months (≤90 days). Agreement between observed EQ-5D index values and predicted values derived from EuroQol crosswalk algorithms. Paired EQ-5D-3L and EQ-5D-5L responses were compared with predicted crosswalk values. Mean absolute error (MAE), root mean square error (RMSE), Spearman's rho, mean difference, Wilcoxon signed-rank tests, and intraclass correlation coefficients (ICC(A,1)) were used. Agreement and bias were visualized with Bland-Altman plots, scatterplots, residual plots, and histograms. The mean age was 57.8 ± 15.5 years; 53.4% were female. The mean EQ-5D-3L index value was 0.311 ± 0.247. Prediction errors were moderate (MAE 0.23-0.26; RMSE 0.29-0.32) with weak-to-moderate correlations (rho 0.54-0.56). ICCs were low (0.51 for 3L predicted from 5L; 0.32 for 5L predicted from 3L). Visual analyses revealed systematic under- and over-prediction, distributional distortion, and non-linear error patterns. EuroQol crosswalk algorithms showed only moderate predictive performance in lumbar spine patients. While they may be acceptable for broad group-level comparisons or retrospective harmonization, they are unsuitable for individual-level assessment or clinical decision-making. Direct use of the appropriate EQ-5D version is recommended.