Developing reference data for nerve conduction studies: an application of quantile regression

Abstract

Introduction: The interpretation of nerve conduction studies, as with any diagnostic test, requires an ability to differentiate normal from abnormal. To this end there have been many efforts to establish reference or normative data, but they have been hampered by a variety of methodological shortcomings. The goal of this article is to introduce a statistical method known as quantile regression, which we contend is better suited than existing methods to generate reference data, especially when there is a need to adjust for covariates.

Methods: Statistical methods previously used for generation of reference data are reviewed. Quantile regression is presented and used to estimate the lower percentiles for response amplitudes of the radial sensory and tibial motor nerves.

Results: Using data from 190 subjects, it is possible to estimate as low as the 2nd percentile for the radial nerve. Using data from 99 subjects it is possible to estimate as low as the 4th percentile for the tibial nerve. Percentile estimation for both nerves required adjustment for age, but no other covariates.

Discussion: Quantile regression is well suited to the estimation of extreme percentiles, the very percentiles that are most relevant to reference data. It is also less dependent on data distribution and permits covariate adjustment, even for continuous variables such as age, which are clinically important determinants of reference data for nerve conduction studies. We recommend the use of quantile regression for future studies of reference data.