Evaluation of Lumbar Spine Motion With Dynamic X-ray–A Reliability Analysis

In Brief Study Design. Radiologic evaluation of lumbar range of motion (ROM) with dynamic radiograph. Objectives. To calculate 95% confidence intervals (CIs) for the measurement error accompanying different methods, different observers, and different levels of training when measuring sagittal plane segmental ROM in lumbar spine. In addition, to compare the 95% CI with frequently common statistical methods of reliability analysis. Summary of Background Data. Dynamic radiographs are commonly used for ROM calculation of the lumbar spine. Yet, the reliability of different measurement methods still remains unclear. Methods. In 24 patients, levels L4–L5 and L5–S1 were measured with the Cobb and superimposition methods on flexion-extension radiographs. There were 2 experienced and 1 inexperienced observer that performed the measurements. The 95% CIs were compared with the corresponding Pearson correlation coefficient and P value (t test). Results. The 95% CI of the superimposition method was ±4.0° for the experienced and ±4.7° for the inexperienced observer. The corresponding values for the Cobb method was ±4.2° for the experienced and ±6.8° for the inexperienced observer. The 95% CI for the measurement error became even worse when different methods or observers were compared, whereas a method constancy revealed superior reliability than observer constancy in experienced observers. Conclusions. For lumbar ROM measurement with dynamic radiograph, the superimposition method seems to be more reliable than the Cobb method. Study protocols dealing with ROM measurement have to calculate the 95% CI of the measurement method used because clinically valid conclusions can only be drawn with respect to these intervals. Even in the ideal setting (i.e., same observer, same method), range of motion measurements with dynamic radiograph has a measurement error of nearly ±4°. Clinically valid conclusions regarding range of motion measurement can only be drawn with respect to the 95% confidence interval for the measurement error and, with limitation, by correlation coefficients. P values are insufficient for this purpose.