Evaluation of methods for prediction of bone mineral density by clinical and biochemical variables in perimenopausal women

Objectives: to predict spinal and femoral bone mineral density (BMD) in perimenopausal women from simple clinical and biochemical variables. Methods: 2016 women 3–24 months past last menstrual bleeding. Mean age 50.1±2.8 years. Age, height, weight, number of full term pregnancies, weekly hours of physical activity, sunbathing habits, use of sun bed, daily intake of calcium and vitamin D, smoking habits, consumption of alcohol, coffee, and tea, history of forearm or femoral neck fractures among the parents, serum osteocalcin (S-OC), serum bone specific isoenzyme of alkaline phosphatase (BSAP), and urine hydroxyproline/creatinine ratio (U-OHP) were used as predictors in three different mathematical models. Lumbar spine (L2–L4) and femoral neck BMD were measured by DEXA. Three mathematical models (multiple regression, logistic regression, and discriminant analysis) were applied. Results: the multiple regression explained 19–21% of the total variation, and the logistic regression and discriminant function had a sensitivity between 53 and 67% with specificity ranging from 67 to 80%. Age, S-OC, serum bone specific alkaline phosphatase, and a maternal history of forearm or femoral neck fractures seemed to be reproducible risk factors for low bone mineral density irrespective of the mathematical model applied. When applied to a separate population, the models performed poorly. Conclusions: Simple clinical and biochemical variables are not useful to predict spinal and femoral BMD in the individual perimenopausal woman.