RESUMO
The objective of this investigation was the design of two instruments based on clinical risk factors for the presumptive detection of post-menopausal women with spinal BMD<2.5 S.D. below average (LBMD). We investigated the association of 20 risk factors (RF) with LBMD in a series of 131 women. According to current densitometric criteria, subjects were classified as normals (N=33); osteopenics (N=53) and osteoporotics (N=45). Normals and osteopenics were taken as a single group because only 'nulliparity' and 'personal fractures' exhibited significant differences between these groups. A logistic regression attempting to identify which factors were associated with osteopenia showed a poor fit (pseudo R(2)=0.289). Univariate unconditional logistic regression analysis was used to calculate odd ratios (ORs) and their 95% CI for all RF. Those with associated P-values <0.100 were included in a multivariate logistic regression analysis to obtain the odds ratios (OR) adjusted by the effects of the others. The variables with not significant beta coefficients were eliminated, producing a reduced model. BMI (<25 kg/m(2)), calcium intake (<1.2g/day), menopause (>10 years), and the simultaneous occurrence of kyphosis and personal fractures showed significant association with low bone mass at the lumbar spine and their effect was additive. Fitting of the data to the model was assessed with the Hosmer-Lemeshow test (P=0.926) The area under the ROC curve is 0.833 (95% CI=0.757-0.909). The following equation calculates the probability of having low spinal bone mass: The sensitivity, specificity and area under the ROC curve were defined. The point of maximum specificity and sensitivity derived from the ROC curve, has a probability of 0.409. With such a cut-off point, the equation has a sensitivity of 73%, specificity 79%, positive predictive value 65% and negative predictive value 85%. The second instrument associates very low lumbar bone mass with the number of risk factors accumulated per patient. At baseline, all subjects had four RFs: they were, women, white, post-menopausal, and with no previous exposure to estrogens. With six additional RFs the presumptive diagnosis of LBMD has a specificity of 99%, positive predicting value 94% and false positives 6.5%. The area under the curve in a ROC graph was 0.826 (95% CI=0.747-0.914). Comparing present instruments with others in the literature, it is concluded that each population require its own algorithm for the presumptive detection of subjects with low bone mass. The algorithm should be reassessed periodically if the characteristics of the population or its social-economic conditions change.
