Definition of osteoarthritis on MRI: results of a Delphi exercise.
Hunter DJ., Arden N., Conaghan PG., Eckstein F., Gold G., Grainger A., Guermazi A., Harvey W., Jones G., Hellio Le Graverand MP., Laredo JD., Lo G., Losina E., Mosher TJ., Roemer F., Zhang W.
Despite a growing body of Magnetic Resonance Imaging (MRI) literature in osteoarthritis (OA), there is little uniformity in its diagnostic application. We envisage in the first instance the definition requiring further validation and testing in the research setting before considering implementation/feasibility testing in the clinical setting. The objective of our research was to develop an MRI definition of structural OA.We undertook a multistage process consisting of a number of different steps. The intent was to develop testable definitions of OA (knee, hip and/or hand) on MRI. This was an evidence driven approach with results of a systematic review provided to the group prior to a Delphi exercise. Each participant of the steering group was allowed to submit independently up to five propositions related to key aspects in MRI diagnosis of knee OA. The steering group then participated in a Delphi exercise to reach consensus on which propositions we would recommend for a definition of structural OA on MRI. For each round of voting, ≥60% votes led to include and ≤20% votes led to exclude a proposition. After developing the proposition one of the definitions developed was tested for its validity against radiographic OA in an extant database.For the systematic review we identified 25 studies which met all of our inclusion criteria and contained relevant diagnostic measure and performance data. At the completion of the Delphi voting exercise 11 propositions were accepted for definition of structural OA on MRI. We assessed the diagnostic performance of the tibiofemoral MRI definition against a radiographic reference standard. The diagnostic performance for individual features was: osteophyte C statistic=0.61, for cartilage loss C statistic=0.73, for bone marrow lesions C statistic=0.72 and for meniscus tear in any region C statistic=0.78. The overall composite model for these four features was a C statistic=0.59. We detected good specificity (1) but less optimal sensitivity (0.46) likely due to detection of disease earlier on MRI.We have developed MRI definition of knee OA that requires further formal testing with regards their diagnostic performance (especially in datasets of persons with early disease), before they are more widely used. Our current analysis suggests that further testing should focus on comparisons other than the radiograph, that may capture later stage disease and thus nullify the potential for detecting early disease that MRI may afford. The propositions are not to detract from, nor to discourage the use of traditional means of diagnosing OA.