Patient selection and study design
We performed a subgroup study within the HONEUR knee cohort, a prospective observational cohort study in general practice, in which patients were included if they consulted their general practitioner for the first time because of acute or chronic knee complaints. The study design of this general cohort, in which a total of 1068 patients were included, has been described in detail by Heintjes et al. . The subgroup under investigation in the present study consisted of consecutively included patients aged 18 to 65 years who were consulting their general practitioners because of an acute knee trauma within the preceding 5 weeks from March 2002 until October 2003 . Patients were excluded in the case of severe injury requiring immediate hospital referral, if a fracture was demonstrated in those referred for conventional radiography, or if there were contra-indications for MRI. The study was approved by the institutional review board of the Maasstad Ziekenhuis, Rotterdam, The Netherlands, and written informed consent was obtained from each patient.
MRI technique and interpretation
Magnetic resonance imaging was performed within 3 to 6 weeks of trauma and repeated after approximately 1 year follow-up. For both examinations, we applied the same technique and pulse sequences, using a 1.0 Tesla whole-body MRI unit and a dedicated knee coil. The imaging protocol consisted of sagittal T1-, T2-, and proton density-weighted fast spin-echo sequences, coronal T2*-weighted gradient echo and fat-suppressed T2-weighted fast spin-echo sequences, and an axial proton density-weighted fast spin-echo sequence.
The initial and follow-up MRI examinations were evaluated for the presence of degenerative abnormalities of the femorotibial joint. For this purpose, we used the items of the Knee Osteoarthritis Scoring System (KOSS) . To grade the severity of OA we used the Kellgren and Lawrence scoring system, which was originally developed for grading osteoarthritic change on conventional radiography  (Table 1). In addition to degenerative abnormalities including meniscal lesions and bone marrow oedema, we assessed the presence of ligamentous lesions on the initial MRI. Findings were documented on a standardised report form.
Two independent readers (one senior radiology resident (SB) and one experienced musculoskeletal radiologist (DV)) evaluated the initial MRI examination. In the case of discrepancies, consensus was reached through discussion. The follow-up MRI examinations were assessed by three independent readers (two senior radiology residents (EHGO and IMK) and one experienced musculoskeletal radiologist (JHH)). In the case of discrepancies the majority opinion was used for analysis.
To reflect clinical practice, both the reports and the images of the initial MRI examination were available on follow-up MRI evaluation. In this observational cohort study, the treating general practitioner was not informed of the findings on the initial MRI unless findings required immediate treatment. Hence, the treatment strategy was not influenced by the MRI findings. If a patient was referred to a medical specialist, the initial MRI report was provided on request, so as to avoid unnecessary repetition of an MRI examination.
Self-reported questionnaires were completed at baseline and 12 months after trauma. On the questionnaires, we recorded a pain score measured on an 11-point numeric rating scale ranging from 0 (no pain) to 10 (unbearable pain) , the Lysholm knee function score , recurrence, level of sports activities, referral to a medical specialist, and whether an operation had been performed. At 12 months we also scored perceived recovery as rated by the patient and measured on a 7-point Likert scale, ranging from ‘completely recovered’ to ‘worse than ever’. Both the initial and follow-up MRI readers were blinded to the reported scores on the questionnaires.
Univariate and multivariate binary logistic regression analysis was performed using SPSS 11.0 for Windows. As the binary outcome, we defined any increase on the Kellgren and Lawrence scale on follow-up MRI compared with initial MRI. Therefore, this could either be new degenerative change in a previously unaffected knee or deterioration of pre-existing osteoarthritis.
We analysed the following independent variables: age (both continuous and dichotomised at age 50 years), sex, physique as measured by body weight and body mass index (both continuous and dichotomised at 25 and 30 kg/m2), sports injury as the setting of knee trauma, baseline pain score measured by the 11-point numeric rating scale (continuous and dichotomised at 6 points or higher), Lysholm knee function score (continuous and dichotomised at 50 points). The prognostic value of the following knee lesions on initial MRI was also evaluated: meniscal lesion (degenerative, tear, or combined; lateral and medial meniscus analysed separately or in combination), anterior cruciate ligament rupture (partial, total, or combined), bone marrow oedema in the distal femur or proximal tibia (medial and lateral femorotibial joint compartments analysed separately and combined). As there were relatively few total PCL and collateral ligament ruptures, we combined these lesions as a total ACL rupture, resulting in a combined variable indicating any total rupture of either cruciate of the collateral ligament. In the univariate analysis, we regarded variables as statistically significant with a p value of less than 0.10. In the multivariate analysis, we used a statistically significant threshold of a p value of 0.05.
We missed data on sports injury in 6 patients. As the association with a sports injury was strongly correlated with age and sex, we used these variables in a logistic regression analysis to impute the missing data on sports injury. Data on baseline pain score and Lysholm knee function score were missing in 3 patients, the pain score and perceived recovery data after 12 months’ follow-up were lacking in 7 and 1 patients respectively. We imputed these data with the mean values of the other patients, as it has been demonstrated that imputation of missing data reduces the risk of bias and is preferable over complete case analysis [22, 23].
Sixteen patients underwent surgery during follow-up. As this study was not aimed at therapy, the exact type of surgical or arthroscopic procedures was not known in every operated patient. To explore the possible effect of surgery on the result of the analysis, we performed a sensitivity analysis from which the operated patients were excluded.
Perceived recovery measured on the self-reported questionnaire at 12 months’ follow-up was regarded as an indicator of clinical outcome. The expected course of recovery one year after knee trauma was complete recovery or at least strong improvement. Following dichotomisation of this variable into complete recovery or strong improvement versus some improvement, unchanged or deteriorated, we analysed the association between OA progression on the Kellgren and Lawrence score and the clinical outcome using Fisher’s exact test.