OBJECTIVES
To review and reassess the published diagnostic performance statistics for MRI of the menisci and cruciate ligaments. To illustrate the potential sources and effects of bias in the evaluation of this widely accepted diagnostic technique.
METHODS
Published evaluations of knee MRI were identified from the literature. Criteria for inclusion in the review were a total sample size > or = 35, arthroscopic correlation of MRI findings and presentation of complete results. Diagnostic performance statistics were then recalculated for each published study.
RESULTS
Twenty-two studies were identified with sample sizes between 35 and 1014. The overall sensitivity for MRI of the menisci and cruciates was 0.88 (95% confidence interval 0.86-0.90). The overall specificity was 0.94 (0.93-0.94). Sampling error varied widely amongst studies and was rarely quantified.
CONCLUSIONS
Diagnostic performance statistics are widely used. It is still not well appreciated that these are subject to sampling error. Such errors make meaningful comparisons between published studies more difficult. Nevertheless, the results for meniscal and cruciate lesions are consistently high and support the use of MRI for these common problems. The diagnostic performance of other applications of MRI should be subjected to similar critical review.