Identifying jitter outliers in single fiber electromyography: Comparison of four methods

Little is known about how different outlier estimation methods affect cutoff limits for outliers in single fiber electromyography.

We compared in a prospective fashion the established 18th jitter value (18thjv) method to three, whole-distribution based, outlier detection methods: the interquartile range (IQR), the log-normal, and the Z-score method. The reference limits were probed in a normal cohort and in myasthenia gravis (MG) patients.

Differences in outlier cutoff values between the different methods were in the range of 2 μs. The number of abnormal muscles according to the computed criteria was similar for all four methods in the control group. Classification metrics (sensitivity, specificity, Youden's statistic, and predictive values) were also similar among the different methods. In the MG group, however, the Z-score method failed to identify the abnormal jitter values. Accordingly, Kappa agreement was substantial to perfect (0.658 to 1) between the three methods (18thjv, IQR, and log-normal), but was equivalent to chance between the three methods and the Z-score in the MG group.

The established 18thjv method proved largely robust when compared to whole-distribution based methods, and its use in clinical practice is justified. Simple estimation of outlier limits by adding two SDs to the mean of the data, leads to unacceptable deviations from the true cutoff values. Moreover, in a clinical scenario in which the final electrodiagnosis depends only on the number of outliers, it is meaningful to accept a tolerance zone of about 2 μs, which is the approximate variation range among the different methods.