Diagnosis of meniscal tears of the knee with MR imaging: effect of observer variation and sample size on sensitivity and specificity.

OBJECTIVE

A wide range in the efficacy of MR imaging for the diagnosis of meniscal tears of the knee has been reported. To evaluate two possible causes for this variation, we studied how sensitivity and specificity are affected when different observers and sample sizes are used.

MATERIALS AND METHODS

Two hundred MR examinations of the knee in patients for whom the results of arthroscopy were available were used for the study. One hundred eight medial meniscal tears and 58 lateral meniscal tears were found at arthroscopy. The sensitivity and specificity for detection of meniscal tears were determined for the original interpretations and retrospective evaluations by three observers. Comparisons were also made between sample sizes of 25 and 100. chi 2 analysis was used for unmatched data sets and McNemar's statistic was used for matched sets.

RESULTS

For the 200 examinations, the sensitivity was 0.89-0.93 for medial meniscal tears and 0.79-0.83 for lateral meniscal tears. The specificity was 0.86 for medial meniscal tears and 0.90-0.92 for lateral meniscal tears. Sensitivity and specificity varied widely among different observers and different sample sizes. However, we found no significant difference between any of the comparisons at the p < .05 level. The largest interobserver variation occurred in the detection of lateral meniscal tears, with a sensitivity of 0.71 for one observer and 0.88 for another observer (p = .16). The largest variation between sets of 100 examinations was a change in sensitivity for detection of lateral meniscal tears from 0.74 to 0.88 for the original interpretations (p = .10). For the sample sets of 25 cases, the variation was even larger, with the sensitivity for detection of lateral meniscal tears varying from 0.5 for one set to 1.0 for another.

CONCLUSION

We conclude that chance variation related to sample size can cause large but not statistically significant variations in sensitivity and specificity in this setting. These variations are of sufficient magnitude to explain many of the differences in reported sensitivity and specificity for MR imaging in the diagnosis of meniscal tears. We found no significant difference in observer performance.