Rasch analysis of the Oxford Knee Score

Objectives: Use Rasch analysis to examine the psychometric properties of the Oxford Knee Score (OKS), particularly in respect to unidimensionality, and consistency of item functioning before and after total knee replacement and across age and gender groups. Methods: The 12-item OKS was administered to 1,712 patients before the surgery, and 1,322 and 855 patients were administered the instrument repeatedly at the 6-month and 2-year postoperative assessments, respectively. Data were fitted to the Rasch partial credit model with the Winsteps program. Differential item functioning (DIF) analysis was performed, and fit statistics in combination with principal components analysis of the residuals were used to test the unidimensionality assumption. The fit criteria were set at 1.5 and 2.0 for infit mean-square (MNSQ) and outfit MNSQ, respectively. Results: At baseline, item difficulty ranged from -1.86 to 1.78 logits, and person measures had a mean ± SD of -0.01 ± 0.89. Misfit items were "limping" and "night pain" in preoperative data and "limping" and "kneeling" in postoperative data. After removing items limping and kneeling and recoding item night pain, none of the items misfit at each of the time points and there was stability of item difficulty ordering across time. In the modified OKS set, five items displayed DIF by age and three by gender. Conclusion: The original OKS had adequate targeting and good coverage of knee severity levels in preoperative patients. The modified 10-item OKS data fit the Rasch model and had stable item difficulty ordering over time.