Comparison of semiquantitative and quantitative techniques for the assessment of prevalent and incident vertebral fractures.

To compare visual semiquantitative (SQ) reading and quantitative morphometry (QM) for assessing prevalent and incident vertebral fractures, radiographs of the thoracolumbar spine were evaluated in 400 (only baseline films) and 335 (baseline and follow-up films) postmenopausal women with osteopenia as defined by a T-score of less than 2 SD below young normals. QM was performed using different cutoff thresholds, and the SQ reading was performed by three radiologists independently. A joint consensus reading of the radiographs by all participating radiologists was used as a reference standard. Our results indicate that the concordance within SQ methods is excellent, and it is higher than that between SQ and QM regardless of the cutoff criteria used. This finding was consistent for the diagnosis of prevalent as well as incident vertebral fractures. For prevalent fractures the use of the 2.5 SD cutoff criterion provided the highest concordance with the consensus reading and with the radiologists' reading, thereby providing high sensitivity (70.23%) with moderate specificity (98.76%) relative to the consensus reading when compared with the 3 SD or 4 SD cutoff criteria. For the diagnosis of incident vertebral fractures the best agreement between our consensus reading and QM was found for an absolute reduction of 6 mm and for a combination of relative and absolute reduction of 15% and 3 mm in vertebral height. The respective sensitivities and specificities for the two criteria were 51%/99.9% and 75.4%/98.9%, respectively. Even though the consensus reading may favor the reader's evaluation over QM, the assessment of prevalent and incident fractures using QM alone may not be sufficiently reliable for detection of vertebral fractures in epidemiological studies and clinical trials. It should be performed principally in conjunction with a trained radiologist or a highly experienced clinician.