Measurement Properties of the Exercise Dependence Scale-Revised (EDS-R): A Systematic Review Using the COSMIN Methodology.

The Exercise Dependence Scale-Revised (EDS-R) is widely utilized internationally, yet empirical evaluations of its measurement properties exhibit pronounced inconsistencies. This systematic review and meta-analysis evaluated the full-spectrum measurement properties of the EDS-R by integrating the COSMIN methodology, modified GRADE approach, and a quantitative Reliability Generalization (RG) meta-analysis. Systematic searches were conducted in PubMed, Web of Science, EMBASE, PsycInfo, and CINAHL up to October 24, 2025, without language restrictions. Methodological quality was appraised via the COSMIN Risk of Bias checklist. Twenty-three studies (N = 14,779) across competitive athletes, sports academy students, gym members, and general populations were included. Content validity and hypothesis testing for construct validity (88.9% alignment, 24/27) were rated as sufficient (moderate-quality evidence). Structural validity was supported for the original first-order seven-factor model (20/22, 90.9%; moderate-quality evidence), as well as higher-order and bifactor configurations (sufficient, high-quality evidence); in contrast, variant 5- and 6-factor models showed mixed fit, and a unidimensional model was rejected (insufficient, high-quality evidence). Demographic and athletic subgroup measurement invariance (gender, age, competition level, training duration) was established with high-quality evidence; however, cross-cultural validity across languages was supported by limited evidence (1 study, insufficient). In the RG meta-analysis, pooled subscale reliabilities ranged from 0.724 to 0.886; notably, the "Reduction in Other Activities" subscale demonstrated marked vulnerability in athletic cohorts (pooled α = 0.561, 95% CI [0.522, 0.597]) compared to non-athletic cohorts (α = 0.760, p = 0.005). Egger's test indicated no significant publication bias across subscales (all p ≥ 0.059). Test-retest reliability was rated as sufficient but supported only by very low-quality evidence due to unstandardized retest windows (7-28 days) and methodological reliance on Pearson's r. Longitudinal responsiveness and measurement error were completely uninvestigated. In conclusion, the EDS-R, particularly in its original seven-factor configuration, appears psychometrically sound for cross-sectional screening and multidimensional assessment. However, caution is warranted when interpreting the "Reduction in Other Activities" subscale in competitive athletes and conducting cross-national score comparisons. Furthermore, until longitudinal responsiveness and measurement error are empirically established, considerable caution is warranted when using the EDS-R to evaluate intervention efficacy or track individual clinical change.