Introduction: Approaches for recognizing vertebral fractures remain controversial.
Methods: An age-stratified population sample of 512 postmenopausal women was followed with serial radiographs for up to 12 years (4455 person-years).
Results: 112 women experienced a new vertebral fracture (20% reduction in any vertebral height from baseline) within this study period, for an annual age-adjusted (to US white women > or =50 years of age in 2000) incidence of 23 per 1000. Depending on the morphometric definition used, the prevalence of vertebral deformities at baseline ranged from 3 to 90%. A recent method to standardize vertebral heights produced the best agreement with a qualitative clinical reading of the films [kappa (kappa), 0.53]. Almost all of the different baseline definitions predicted future vertebral fractures, but most of the predictive power was attributable to the severe (e.g., 4 SD) deformities included within more generous (e.g., 3 SD) classifications. Whereas the generous definitions were more sensitive, and the restrictive ones more specific, their overall abilities to predict a new vertebral fracture were roughly comparable as evaluated by the c-index (analogous to the area under an ROC curve).
Conclusion: This result suggests that the choice of a morphometry definition depends on the particular application and, in particular, on whether it is more important to maximize sensitivity or specificity.