Effect of statistical methodology on normal limits in nerve conduction studies

Abstract

Mean ± 2 standard deviations (SD), which relies on a Gaussian distribution, has traditionally been used to derive normal limits for nerve conduction studies. Our purpose was to examine skew in nerve conduction study (NCS) parameters, and to compare normal limits derived by several alternative methods. We examined 22 NCS parameters from 75 asymptomatic, nondiabetic men (controls). The coefficient of skewness (g₁) was significantly positive (P < 0.10, two-tailed test) in 5 of 8 amplitude and 6 of 8 latency measurements. Transformation reduced g₁ in 19 of 22 parameters, and was optimal when g₁ was closest to zero. For each measurement, ideal normal limits were defined as mean ± 2 SD of the optimally transformed data of the control subjects. The percentage of 66 diabetic subjects classified as abnormal by the raw data, but normal by the ideal normal limits, was the positive misclassification rate; while the percentage considered normal by the raw data, but abnormal by the ideal normal limits, was the negative misclassification rate. Mean ± 2 SD of the raw data produced up to 11% positive misclassifications and 12% negative misclassifications. When the range of observed values was used, up to 6% positive misclassifications and 13% negative misclassifications were found, while the 2.5 or 97.5 percentile values produced up to 10% positive misclassifications and 13% negative misclassifications. We conclude that analyses using the raw data to derive normal limits result in an unacceptable rate of misclassification. Normal limits should be derived from the mean ± 2 SD of the optimally transformed data.