Implementation and evaluation of a statistical framework for nerve conduction study reference range calculation

Nerve conduction studies (NCS) play a central role in the clinical evaluation of neuropathies. Their clinical utilization depends on reference ranges that define the expected parameter values in disease-free individuals. In this paper, a statistical framework is proposed and described in detail for deriving NCS parameter reference ranges. The bootstrap technique is used to identify demographic and physiologic covariates that influence the NCS measurements. Multi-variate linear regression is used to improve the accuracy and effectiveness of NCS interpretation by reducing parameter variance. Non-linear mappings are used to transform parameters into a Gaussian distribution in order to minimize the influence of outliers. Modeling of heteroscedasticity observed in this and other studies leads to more sensible normal limits for several parameters. The proposed reference range method is automated using the MATLAB programming language. Data from a large sample of healthy subjects are used to establish reference ranges for 24 commonly measured NCS parameters. All but three parameters follow Gaussian distributions in their respective transformed domains. Excluding the distal motor latency difference between median and ulnar nerves, the reduction of the parameter variance as a result of regression in the transform domain is greater than 50% for all F-wave latency parameters and at least 10% for all other NCS parameters. Subject age is found to influence normal limits of all but one parameter and height has a statistically significant impact on all but three parameters. These reference range specifications provide clinicians with an alternative to developing their own reference ranges as long as their NCS techniques are consistent with those described in this paper. The proposed method should also be applicable to reference range development for other NCS techniques and physiological measurements.