|
The accuracy of magnetic resonance imaging in the diagnosis of multiple sclerosis: a diagnostic systematic review carried out according to new Cochrane methods |
Background:
The Cochrane Collaboration has decided to include systematic reviews of diagnostic test accuracy studies. We have completed a review on the accuracy of magnetic resonance imaging (MRI) to diagnose multiple sclerosis (MS).
Objective:
To examine the accuracy of MRI in the diagnosis of MS. We focus on how we applied the Cochrane recommendations at each review stage.
Methods:
We followed the methods of the draft Cochrane diagnostic reviewers’ handbook. We searched electronic databases, bibliographies of included studies, and contacted experts. Diagnostic accuracy studies that compared MRI to a reference standard, or that compared MRI results in patients with and without MS, were eligible. Studies were excluded if they did not report data to construct a 2x2 table of test performance.
Data extraction and quality assessment were done by one reviewer and checked by a second. Data were initially analysed using simple summary statistics (sensitivity, specificity, positive and negative likelihood ratios). We meta-analysed studies that reported results for the same MRI criterion using summary ROC methods from which a typical operating point was identified.
Results:
We identified >6000 references; 29 studies were included. There was heterogeneity between studies in populations, MRI protocol, quality and the criterion used to define a positive result. Many studies published data for multiple criteria. Studies based on inappropriate patient spectra reported greater diagnostic accuracy (Figure 1). Of studies with appropriate spectra, five studies reported positive likelihood ratios (LR+) in excess of five, and one in excess of 10. Only six studies reported negative likelihood ratios (LR-) below 0.1. The Paty criteria were the most frequently evaluated (assessed in seven studies) and showed poor accuracy for ruling in disease (pooled LR+=2.3, considerable heterogeneity), and some potential for ruling out disease (pooled LR-=0.36, no heterogeneity). The Barkhof criteria were evaluated in two independent studies which reported poor accuracy (LR+=2.3, LR- = 0.5, no heterogeneity). The criterion-generating study reported higher accuracy.
Conclusions:
The new Cochrane methods provided a practical framework for a diagnostic systematic review. The accuracy of MRI is not high enough to give likelihood ratios sufficient to rule in or rule out MS with high certainty.