Quantitation of incident spinal fractures: comparison of visual detection with quantitative morphometry

The value of quantitative morphometry in detection of new spinal fractures was assessed in serial radiographs from 83 patients with osteoporosis. From vertebral landmarks on lateral spine radiographs, a computer program allowed calculation of vertebral heights and area. By comparing vertebral dimensions in the two sets of films, incident spinal fractures could be quantitated based either on the minimum criteria of 15% reduction in vertebral height (CM2) or a fall in height and area of 20% and 10% (CM1). The results of quantitative morphometry were compared with those of the consensus and individual readings of visual detection by three experienced investigators in the same paired sets of spinal films. For incident new fractures, the visual consensus method (V-C) showed a very good agreement with individual visual detection (kappa of 0.794 to 0.916) as well as with CM1 (kappa of 0.821). However, there was a poor agreement between the results of consensus reading and of detection by CM2 (kappa of 0.341), due to excessive number of fractures identified by CM2, but not by the visual method. For incident recurrent fractures, there was a poor agreement between V-C and individual visual detection, and between V-C and quantitative morphometry (kappa of 0.306 to 0.496). It was due to severe compression fractures at baseline, which caused further changes in vertebral dimensions difficult to measure accurately by either visual or quantitative morphometry. Thus, if the visual detection of fractures by a consensus of experienced investigators is considered as the "gold standard," quantitative morphometry, based on minimum reduction in vertebral height of 20% accompanied by a minimum decline in area of 10%, provides an objective detection of incident new spinal fractures but not of recurrent fractures.