Test–Retest Reliability and Sensitivity to Change of Ultrasound‐Based Methods of Measuring Synovial Inflammation in Knee Osteoarthritis

Abstract

Objectives. To assess test-retest reliability of musculoskeletal ultrasound (US) measures of inflammation in patients with knee osteoarthritis (OA) and to assess the sensitivity to change of US measures of inflammation in patients with knee OA.Methods. To mimic a common clinical scenario, 36 patients (n = 70 knees) with symptomatic knee OA who were in stable condition underwent 2 assessments within 14 days by different operators and different US machines, graded by a single rater. Test-retest reliability was measured using Cohen's kappa coefficient, intraclass correlation coefficient (ICC), and absolute agreement parameters. A total of 51 patients (n = 72 knees) were tested immediately before and 21-28 days after intraarticular glucocorticoid injection to investigate sensitivity to change and longitudinal construct validity. Paired t-tests and standardized response mean (SRM) were used to assess sensitivity to change. Multivariate linear regression was used to investigate longitudinal construct validity of US with Knee Injury and Osteoarthritis Outcome Score (KOOS) pain scores, while adjusting for covariates.Results. US measures of inflammation demonstrated moderate (κ = 0.41, 0.60) to substantial (κ = 0.61, 0.80) agreement. Quantitative measures of synovitis and effusion demonstrated good test-retest reliability (ICC 2,1 0.71, 0.92). US measures of synovitis and effusion demonstrated low-to-moderate sensitivity to change (SRM -0.29, -0.50). The associations between changes in US measures and KOOS pain scores over time were low, and 95% confidence intervals included zero.Conclusion. In a clinical setting, US measures of inflammatory features of knee OA have substantial reliability and low-to-moderate sensitivity to change, whereas measures of structural OA features are less reliable. Longitudinal construct validity of US measures of synovitis and effusion to KOOS pain scores is not strongly supported.