Comparison of semiquantitative visual and quantitative morphometric assessment of prevalent and incident vertebral fractures in osteoporosis

The assessment of radiographs for vertebral fractures is important in the clinical evaluation of patients with suspected osteoporosis, in the epidemiological evaluation of elderly populations, and in clinical trials of osteotrophic drugs. The purpose of this study is to compare visual semiquantitative (SQ) approaches and quantitative morphometric approaches for assessing prevalent and incident vertebral fractures in postmenopausal osteoporosis. We analyzed lateral thoracolumbar spine radiographs (baseline and approximately 3.5 year follow-up) of 503 women (age > or = 65) randomly selected from the Study of Osteoporotic Fractures (SOF) population. SQ assessment by an experienced radiologist graded vertebral fractures from 0 (normal) to 3 (severe). Incident fractures by SQ were defined as an increase of > or = 1 grade on follow-up radiographs. Trained research assistants visually triaged women as normal, uncertain, or probably fractured and visually flagged vertebrae with moderate/severe (grade > or = 2) prevalent fractures or with any (grade > or = 1 change) incident fracture. The radiographs were also digitized by research assistants, and quantitative morphometry (QM) was used to classify vertebral deformities at several cut-offs based on standard deviation (SD) reductions in height ratios from normal means, e.g., QM > or = 3 SD. Incident fractures by QM were defined as a decrease in height of more than 15% (QM15) on follow-up radiographs. Finally, a combination of these methods was used to detect moderate/severe prevalent fractures and any grade of incident fractures. In the overall analysis, the prevalence of fractures varied from 14 to 33% and the incidence from 5 to 10% by woman, depending upon the method and cut-off criteria. In the detailed analysis, considering visually triaged uncertain as abnormal, triage by research assistants detected 97.0% (163/168) of women with SQ grade > or = 1 fractures and 100% (70/70) with SQ grade > or = 2 fractures. Visual flagging by research assistants detected 88.5% (108/122) of SQ > or = 2 prevalent fractures (kappa score, kappa = 0.82) and 85.2% (52/61) of SQ incident fractures (kappa = 0.79). QM > or = 3 SD detected 37.9% (141/372) of SQ > or = 1 prevalent fractures (kappa = 0.51) and 79.5% (97/122) of SQ > or = 2 prevalent fractures (kappa = 0.68), plus 18 vertebrae without SQ fractures. QM 15 detected 59% (36/61) of SQ incident fractures (kappa = 0.70), plus five vertebrae without SQ incident fractures. The combination assessment detected 92% (112/122) of SQ > or = 2 prevalent fractures (kappa = 0.76) and 84% (51/61) of SQ incident fractures (kappa = 0.91). The precision errors of QM vertebral height measurements (baseline versus follow-up) ranged from 2.71 to 2.92%. Nevertheless, excluding the 5719 vertebrae that were clearly normal by morphometry, i.e., within 2 SD of the normal means at both baseline and follow-up, two-thirds (358/556) of the remaining vertebrae changed classification by at least 1 SD category. Visual triage and visual flagging by research assistants appear to be highly effective methods for vertebral fracture assessment in osteoporosis, potentially reducing the number of false-positive and false-negative fractures detected by QM, at least relative to SQ by the radiologists. There is higher concordance among the visual approaches studied than between the visual SQ and quantitative morphometric approaches, with QM having limited ability to detect mild fractures but good ability to detect moderate/severe fractures, as classified by SQ. Use of a combination of sensitive qualitative and quantitative criteria, with adjudication by an experienced radiologist, is feasible and draws upon the relative strengths of each of the methods. Quantitative morphometry should not be performed in isolation, particularly when applying highly sensitive morphometric criteria at low threshold levels, without visual assessment to confirm the detected prevalent or incident vertebral defor