Interpretation of bone mineral density measurement and its change.

Bone mineral density (BMD), an important measurable predictor of osteoporotic fractures, is used as a surrogate definition of osteoporosis, and often as an end point in clinical trials. However, BMD is measured with random error and random fluctuation within individuals. This study addresses two specific questions: Given an observed level of BMD for an individual, what is the individual's likely "true" level? To what extent does an observed BMD change reflect a real change? A Bayesian model was formulated to address these questions, using data from the Dubbo Osteoporosis Epidemiology Study, past clinical trials, and a short-term reliability study in individuals ages 60 yr and older. Measurements of lumbar spine and femoral neck BMD by dual X-ray absorptiometry are highly reliable, with coefficients of reliability ranging from 0.90 to 0.99. Consequently, for an individual, there is good agreement between observed and "true" BMD values. However, the 90% confidence interval for the true level in elderly people and those with low measured BMD values is particularly wide. Using the cutoff of 2.5 standard deviations below the young normal mean as a definition of osteoporosis, the rates of false positives and false negatives can be as high as 20% among individuals ages 80 yr and older. In a typical clinical trial with an overall average increase in BMD of 2%, for a subject whose baseline femoral neck BMD is 0.80 g/cm(2), no conclusion of significant change could be drawn until an observed increase of at least 5.5% or an observed decrease of at least 7.5%. If two measurements were taken at baseline and follow-up, the true change could be detected with an observed increase of 3.5% or an observed decrease of 5%. On the other hand, there needs to be an observed increase of 6.2 and 8.5% before one can be 90% certain that a true increase of 2 and 5%, respectively, has occurred in an individual. This analysis suggests that the diagnosis of osteoporosis based on a single measurement of BMD and point estimates of changes in BMD may be inappropriate and unreliable. We propose that the current practice of informing individuals who have BMD measurements about their actual T- and Z-scores be replaced with a system of reporting in which their osteoporosis probability risk category is conveyed. Also, assessment of change in an individual should take into account the overall change in a population.