Accelerometric gait analysis for use in hospital outpatients

To provide clinicians with a quantitative human gait analysis tool suitable for routine use.

We evaluated the reproducibility, sensitivity, and specificity of gait analysis based on measurements of acceleration at a point near the center of gravity of the body. Two accelerometers held over the middle of the low back by a semi-elastic belt were used to record craniocaudal and side-to-side accelerations at a frequency of 50 Hz. Subjects were asked to walk at their normal speed to the end of a straight 40 meter-long hospital corridor and back. A 20-second period of stabilized walking was used to calculate cycle frequency, stride symmetry, and stride regularity. Symmetry and regularity were each derived from an auto-correlation coefficient; to convert their distribution from nonnormal to normal, Fisher's Z transformation was applied to the auto-coefficients for these two variables. Intraobserver reproducibility was evaluated by asking the same observer to test 16 controls on three separate occasions at two-day intervals and interobserver reproducibility by asking four different observers to each test four controls (Latin square). Specificity and sensitivity were determined by testing 139 controls and 63 patients. The 139 controls (70 women and 69 men) were divided into five age groups (third through seventh decades of life). The 63 patients had a noninflammatory musculoskeletal condition predominating on one side. ROC curves were used to determine the best cutoffs for separating normal from abnormal values.

Neither intra- nor interobserver variability was significant (P > 0.05). Cycle frequency was significantly higher in female than in male controls (1.05 +/- 0.06 versus 0.98 +/- 0.05 cycles/s; P < 0.001). Neither symmetry nor regularity were influenced by gender in the controls; both variables were also unaffected by age, although nonsignificant decreases were found in the 61 to 70-year age group, which included only nine subjects. In the ROC curve analysis, the area under the curve was high for all three variables (frequency, 0.81 +/- 0.04; symmetry, 0.85 +/- 0.03; and regularity, 0.88 +/- 0.03), establishing that there was a good compromise between sensitivity and specificity.

Our gait analysis method offers satisfactory reproducibility and is sufficiently sensitive and specific to be used by clinicians in the quantitative evaluation of gait abnormalities.