Educational level and osteoporosis risk in postmenopausal Moroccan women: a classification tree analysis

The objectives of this study are (1) to evaluate whether the prevalence of osteoporosis and peripheral fractures might be influenced by the educational level and (2) to develop a simple algorithm using a tree-based approach with education level and other easily collected clinical data that allow clinicians to classify women into varying levels of osteoporosis risk. A total number of 356 women with a mean age of 58.9 ± 7.7 years were included in this study. Patients were separated into four groups according to school educational level; group 1, no education (n = 98 patients); group 2, elementary level (n = 57 patients); group 3, secondary level (n = 138 patients) and group 4, university level (n = 66 patients). We observed dose–response linear relations between educational level and mean bone mineral density (BMD). The mean BMDs of education group 1 (10.39% (lumbar spine), 10.8% (trochanter), 16.8% (wrist), and 8.8% (femoral neck)) were lower compared with those of group IV (p < 0.05). Twelve percent of patient had peripheral fractures. The prevalence of peripheral fractures increased with lowered educational levels. Logistic regression analysis revealed a significant independent increase in the risk of peripheral fracture in patients with no formal education (odds ratio, 5.68; 95% , 1.16–27.64) after adjustment for age, BMI and spine BMD. Using the classification tree, four predictors were identified as the most important determinant for osteoporosis risk: the level of education, physical activity, age >62 years and BMI <30 kg/m². This algorithm correctly classified 74% of the women with osteoporosis. Based on the area under the receiver–operator characteristic curves, the accuracy of the Classification and Regression Tree (CART) model was 0.79. Our findings suggested that a lower level of education was associated with significantly lower BMDs at the lumbar spine and the hip sites, and with higher prevalence of osteoporosis at these sites in a dose–response manner, even after controlling for the strong confounders. On the other hand, our CART algorithm based on four clinical variables may help to estimate the risk of osteoporosis in a health care system with limited resources.