3.2 Associations of ED and osteoporosis and comparison of different models for assessing the prevalence of osteoporosis
As shown in Table 2, we performed logistic regression analysis to explore the associated of ED and osteoporosis. ED can be used as risk factor for prevalence of osteoporosis, the prevalence of osteoporosis was 1.70-fold higher in the ED group compared with the non-ED group (OR: 1.70, 95% CI: 0.99-2.87, P =0.051) after adjustment in total population.
The LASSO regression accurately identified the important candidates for assessing osteoporosis. As shown in Figure 3A , after the LASSO regression analysis, ten features with nonzero coefficients out of 20 clinical parameters (age, education, smoking, drinking, SBP, DBP, TC, LDL-C, HDL-C, TG, BMI, waistline, family history of diabetes, neck circumference, P, WHR, heart rate, PFG, OGTT2h and HbA1c) were selected as important candidates for assessing prevalence of osteoporosis. Of the 10 indices selected by LASSO regression by λ=0.01 included age, education, smoking, drinking, SBP, LDL-C, BMI, neck circumference, heart rate and FPG. Different models were established to assess the ability of diagnostic for prevalence of osteoporosis. Model 1 included age, education, smoking, drinking, SBP, neck circumference, BMI and heart rates, which all of those were non-traumatic parameters. Model 2 further included LDL-C and FPG based on model 1. Model 3 were further included ED based on model 2. We constructed ROC curve to assess the ability of different models to assess the prevalence of osteoporosis (Figure 3B ). The AUC for the ROC curves were 0.70 (95% CI: 0.65-0.80), 0.72 (95% CI: 0.67-0.78) and 0.73 (0.68-0.79) for model 1, model 2 and model 3, respectively. Compared to model 1, model 3 plus LDL-C, FPG and ED were significantly improved the diagnostic capabilities (P=0.017).