CONTEXT: Bone density measurement with dual-energy x-ray absorptiometry (DXA) is widely used for fracture risk assessment. It has not been established that published gradients of fracture risk from study populations can be directly applied to clinical populations.

OBJECTIVE: To assess osteoporotic fracture prediction with DXA in a large clinical cohort.

DESIGN: Historical cohort study (mean observation period 3.2 ± 1.5 yr).

PATIENTS: The study population was drawn from the population-based database of the Manitoba Bone Density Program. Analyses were limited to women age 50 yr or older at baseline (n = 16,505).

MAIN OUTCOME MEASURE: Each subject's longitudinal health service record was assessed for the presence of non-trauma fracture codes (hip, spine, wrist and humerus) after bone density testing. Age-adjusted hazard ratios for fracture were derived from Cox proportional hazards models.

RESULTS: Site-specific and overall fracture rates were significantly associated with each site of bone density measurement (all P < 0.00001). The 95% confidence intervals overlapped those from a widely cited meta-analysis of fracture prediction from different sites. Although fracture prediction was not significantly different between the three hip measurement sites, each hip site was better than the lumbar spine for predicting overall fractures (non-overlapping 95% confidence intervals). The manufacturer SD (equivalent to a unit change in T-score) resulted in a significantly smaller gradient of risk for the spine than when the population SD was used.

CONCLUSIONS: Bone density measurements are effective for predicting fractures in clinical practice. However, hip measurements were superior to the spine in overall osteoporotic fracture prediction.