Mild versus definite osteoporosis: comparison of bone densitometry techniques using different statistical models.

The purpose of this investigation was to determine the ability of three bone densitometry techniques to discriminate subjects with mild vertebral deformities from those with definite compression fractures. We determined bone mineral density (BMD) in 68 postmenopausal women by quantitative computed tomography (QCT) and dual-photon absorptiometry (DPA) of the spine, as well as single-photon absorptiometry (SPA) of the radius. Forty four individuals were classified as having mild deformities of the spine and 24 were considered to have definite vertebral compressions. Several statistical approaches were used to compare these subgroups and to estimate the relative risk of vertebral fracture. Included among these were percent decrements and zeta-scores, ROC curves, odds ratio estimations, and logistic regression analysis. Individuals with definite vertebral fractures had lower bone mineral density at all sites, but measurement of radial compact bone by SPA failed to reach significance. Using ROC analysis to distinguish mild deformities from true compressions, we found that measurement of spinal trabecular bone by QCT to be the most sensitive discriminator; although measurement of spinal integral bone by DPA also gave satisfactory discrimination, whereas assessment of radial compact bone did not adequately differentiate patients with mild deformities from those with definite compressions. Likewise, we found determination of spinal trabecular bone to be the most robust predictor of relative risk of definite fracture using either odds ratios or logistic regression analysis. Measurement of BMD in the peripheral cortical skeleton offered no predictive power for true vertebral fracture. We concluded that direct assessment of the spine, particularly of the trabecular portion, offered the strongest discrimination and relative risk prediction for definite osteoporotic fractures compared with milder forms of this condition.