William D Leslie1, Sumit R Majumdar, Suzanne N Morin, Lisa M Lix. 1. University of Manitoba, Department of Medicine (W.D.L.), Winnipeg, Manitoba, Canada R2H 2A6; University of Manitoba, Department of Community Health Services (L.M.L.), Winnipeg, Manitoba, Canada R3E 0W3; University of Alberta, Department of Medicine, (S.R.M.), Edmonton, Alberta, Canada T6G 2B7; and McGill University, Division of General Internal Medicine (S.N.M.), Montreal, Quebec, Canada H3G 1A4.
Abstract
CONTEXT: Intuitively, rapid bone mineral density (BMD) loss should predict fracture risk independently of current BMD, but studies have not confirmed this. We hypothesized that measurement error when characterizing rates of BMD loss might explain this paradox. OBJECTIVE: To examine the importance of measurement error in predicting BMD loss. DESIGN AND SETTING: Retrospective registry study using BMD results for Manitoba, Canada. PATIENTS: Untreated women age 50 years and older with three femoral neck BMD tests. MAIN OUTCOME MEASURES: Correlation in annualized rates of BMD change for interval 1 (first to second scan) versus interval 2 (second to third scan) with confirmatory model-based simulations that varied measurement error and testing intervals. RESULTS: Five hundred forty two women with a mean age of 62 years had BMD measurements separated by a mean of 3.5 years for interval 1 and 3.4 years for interval 2. Mean femoral neck BMD loss was stable (-0.5% per year for interval 1, -0.6% per year for interval 2) with a weak negative correlation between intervals (r = -0.11, P = .01). There were no significant correlations for BMD change at the total hip (r = 0.01, P = .74) or total spine (r = -0.01, P = .77). Simulations showed low explained variation for BMD change between intervals 1 and 2 (<20%). To explain 50% of the variation of BMD change between intervals 1 and 2 required a BMD measurement error ≤ 0.008 g/cm(2) or a BMD testing interval ≥ 5 years. CONCLUSIONS: The low correlation between past and future BMD loss helps explain why the rate of BMD loss is unlikely to be helpful for refining fracture risk.
CONTEXT: Intuitively, rapid bone mineral density (BMD) loss should predict fracture risk independently of current BMD, but studies have not confirmed this. We hypothesized that measurement error when characterizing rates of BMD loss might explain this paradox. OBJECTIVE: To examine the importance of measurement error in predicting BMD loss. DESIGN AND SETTING: Retrospective registry study using BMD results for Manitoba, Canada. PATIENTS: Untreated women age 50 years and older with three femoral neck BMD tests. MAIN OUTCOME MEASURES: Correlation in annualized rates of BMD change for interval 1 (first to second scan) versus interval 2 (second to third scan) with confirmatory model-based simulations that varied measurement error and testing intervals. RESULTS: Five hundred forty two women with a mean age of 62 years had BMD measurements separated by a mean of 3.5 years for interval 1 and 3.4 years for interval 2. Mean femoral neck BMD loss was stable (-0.5% per year for interval 1, -0.6% per year for interval 2) with a weak negative correlation between intervals (r = -0.11, P = .01). There were no significant correlations for BMD change at the total hip (r = 0.01, P = .74) or total spine (r = -0.01, P = .77). Simulations showed low explained variation for BMD change between intervals 1 and 2 (<20%). To explain 50% of the variation of BMD change between intervals 1 and 2 required a BMD measurement error ≤ 0.008 g/cm(2) or a BMD testing interval ≥ 5 years. CONCLUSIONS: The low correlation between past and future BMD loss helps explain why the rate of BMD loss is unlikely to be helpful for refining fracture risk.