Background: The McDonald International Panel accepted the Barkhof/Tintoré criteria for providing MRI evidence of dissemination in space to allow a diagnosis of multiple sclerosis in patients with clinically isolated syndromes (CIS). We applied these criteria in a large cohort of patients with CIS, representative of those seen in a general diagnostic setting, to assess their accuracy in predicting conversion to definite multiple sclerosis and to identify factors that affect this risk.
Methods: In a collaborative study of seven centres, baseline MRI and clinical follow-up data for 532 patients with CIS were studied, with the development of a second clinical event used as the main outcome. All scans were scored for lesion counts and spatial lesion distribution to assess the fulfilment--ie, at least three out of four--of the Barkhof/Tintoré criteria. We used survival analysis and 2x2 tables to assess the test characteristics of the criteria at baseline.
Findings: Overall conversion rate was 32.5% with a median survival time of 85.3 months. Fulfilment of the criteria at baseline showed, after a survival time of 2 years, a conversion rate of about 45% (95% CI 37-53) versus about 10% (6-16) in those with no asymptomatic lesions at baseline (p<0.0001). For patients with a follow-up of at least 2 years, the fulfilment of the MRI criteria showed an accuracy of 68% (sensitivity 49%, specificity 79%) for predicting conversion and an increase in risk of nearly four times for conversion compared with those not fulfilling the criteria (odds ratio 3.7, 95% CI 2.3-5.9; p<0.0001). Cox proportional hazards regression analysis accorded with this increased risk. No effects were recorded on the performance of the criteria by sex, presenting symptoms, or centre. Age at baseline did have a small but significant effect as predictor (hazard ratio 0.97, 0.95-0.99; p=0.002), but did not affect the prognostic value of the MRI criteria.
Interpretation: MRI abnormalities have important prognostic value. The cut-off, based on the Barkhof/Tintoré criteria, as incorporated in the McDonald diagnostic scheme yields acceptable specificity, but could have lower sensitivity than previously reported.