Abstract
Purpose To develop a risk index by item reduction from multiple variable regression, which can identify male Taiwanese patients at risk of developing osteoporosis.
Methods To develop the model, a risk index was identified by item reduction from multivariate regression analysis. Using receiver operating characteristic (ROC) curve analysis and their sensitivity/specificity, MOSTAi was validated in a separate cohort of Taiwanese men and its performance with compared with the National Osteoporosis Foundation recommendations (NOF 2013).
Results Between 2008 and 2011 a total of 4,323 males were enrolled for bone mineral density (BMD) measurements. Univariate analysis identified four major risk factors for osteoporosis, including age, body weight (BW), previous fracture and body height. The ROC analysis showed the area under the curve (AUC) for the model based on the three-variable, two-variable (age and BW), and one-variable models (BW), was 0.701 (p<0.001, 95% confidence interval [CI] 0.658-0.744), 0.700 (p<0.001, 95% CI, 0.656-0.742), and 0.690 (p<0.001, 95% CI, 0.646-0.734), respectively. Using the optimal cutoff value (−2) for the OSTA, the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) in the validation cohort were 64.0%, 65.7%, 26.9% and 90.2%, respectively. The ROC curves for predicting osteoporosis by MOSTAi, OSTA and NOF 2013 and the AUC for MOSTAi, OSTA and NOF 2013 was 0.706 (p<0.001, 95% CI: 0.664-0.748) and 0.697(p<0.001, 95% CI: 0.657-0.738), respectively.
Conclusion The results showed that MOSTAi could be a more precise model than OSTA and NOF 2013, for identifying men in Taiwan with osteoporosis who require referrals for DXA scans. It was demonstrated that MOSTAi is a simple tool with fair sensitivity/specificity and PPV, and high NPV. MOSTAi could also be a more accurate model than OSTA for identifying men in Taiwan at risk of osteoporosis. In comparison with NOF 2013, MOSTAi is a more accurate and simpler tool for the referral of Taiwanese men for DXA scans.
