Secondary Logo

Journal Logo

Institutional members access full text with Ovid®

Classification of Low Back Pain With the Use of Spectral Electromyogram Parameters

Peach, John P., MSc; McGill, Stuart M., PhD


Study Design. An electromyogram procedure using spectral parameters to distinguish subjects with low back pain from those without.

Objectives. To add to the growing database on this procedure, to assess the possible overfitting of data in the classification model, to determine whether a model based on a contraction level of 60% of maximum voluntary contraction can produce concordance rates similar to those in models based on 40% and 80% of maximum voluntary contraction, and to develop a classification model to distinguish subjects with low back pain from those without.

Summary of Background Data. Other investigators have published a series of models in which spectral parameters measured during fatiguing contractions from the paraspinal muscles have been able to classify a subject into a low back pain or non-low back pain group with a more than 80% concordance rate.

Methods. Subjects with chronic low back pain (N = 21) and without (N = 18) performed a series of isometric, fatiguing back extensor contractions in which the median power frequency was measured bilaterally from T9, L3, and L5. A Student's t test was used to determine which parameters would be entered into the classification models. Discriminant analysis and logistic regression procedures were used to develop models to classify subjects and were compared for overfitting of data based on the number of input parameters. The logistic regression method used a holdout group (N = 6) for validation.

Results. The discriminant analysis selected all 10 input parameters and was believed to overfit the data. Logistic regression selected two parameters and had a concordance rate of 92.4%. Five of the six subjects in the holdout group were correctly classified.

Conclusions. The use of spectral parameters to classify subjects with low back pain from those without appears to have merit. Compared with discriminant analysis, logistic regression provided an equally powerful method for classifying these two groups but did not overfit the data. Models based on 60% of maximum voluntary contraction demonstrated results comparable with those of previous research using 40% and 80% of maximum voluntary contraction.

From the Occupational Biomechanics and Safety Laboratories, Faculty of Applied Health Sciences, Department of Kinesiology, University of Waterloo, Ontario, Canada.

Supported, in part, by the Natural Sciences and Engineering Council, Canada.

Acknowledgment date: July 15, 1997.

Acceptance date: October 27, 1997.

Device status category: 1.

Address reprint requests to: Professor Stuart M. McGill; Occupational Biomechanics and Safety Laboratories; Department of Kinesiology; University of Waterloo; 200 University Ave.; Waterloo, Ontario, N2L 3G1; Canada; E-mail:

© Lippincott-Raven Publishers.