Body height changes with hyperextension

Objective. To automatize the lumbar physical examination with an acceptable rate of error. Design. An external skin marker method for automatizing the physical examination was developed and its ability to discriminate between normal and abnormal subjects tested in a blind clinical trial. Background. The low reproducibility of clinical findings, even among experienced doctors, has been well documented. This is of particular concern and may explain why there is such a wide variation in surgical rates across the USA (tenfold for disc herniation). Inconsistencies among physicians in the evaluation of benign low back conditions make standardization desirable. Methods. A computerized physical examination was used to evaluate patients with low back pain and compare their results with a normative database obtained from a selection of healthy subjects. A high-resolution motion analysis system tracked the movement of skin markers placed on the midline and pelvis. Surface EMG electrodes placed above L5 collected data from multifidus. From the kinematics of skin markers during flexion-extension with lifts up to 32 kg, and lateral bending with lifts up to 10 kg, the following parameters were estimated: lumbosacral angle and elongation, contribution of each lumbar segment to the lordosis reduction, relative pelvic/spine motion, and trunk velocity. First the average normal value for each estimated parameter was determined using 40 normal subjects. For each subject the difference between his parameter and the normal was processed by an expert system generating a normality index varying from zero (perfect abnormal) to one (perfect normal). To develop the expert systemʹs rules, a preliminary group of 20 very abnormal subjects was used, such that the normality index separated them from the normals. For validation, a set of 29 back-sprain patients and another set of 42 discogram-positive patients were selected. Each subject was tested and his computerized normality index calculated without any clinicianʹs input, then compared with the clinicianʹs evaluation, which was taken to be the gold standard. The receiver operating characteristic technique was used to quantify the discrepancies. Results. The expert system could detect clinically abnormal subjects with accuracy (sensitivity 83–91% and specificity >90%) while providing quantitative information on workersʹ functional capacities. Conclusions. Once a reference normative database is agreed upon, each patient can be compared with that reference according to the same rules, with the resulting machine classification being independent of the clinician. This eliminates the inter- and intra-clinician variability in patient follow-up. Because of the severity of the selection criteria, this study is based upon a relatively restricted number of subjects, as well as a limited normative database of 40 subjects.