Objective: To evaluate the effectiveness of Bayesian logistic regression model in the development of a clinical risk score index for screening of osteoporosis in menopausal women.
Materials and Methods: Data of 973 menopausal women attending the menopause clinic at Faculty of Medicine Vajira
Hospital between January 2002 and January 2008 were used as a derivation cohort. Age, body weight, menopausal duration, current estrogen use, previous low impact fracture, and lumbar and total hip bone mineral density [BMD] measurement by dual energy x-ray absorptiometry [DEXA] were used to develop a scoring system under 4 different scenarios. By using the Bayesian logistic regression model, the beta coefficients from the best fitting model of each scenario were transformed into simplified scoring algorithms in the derivation cohort. The diagnostic performance and their 95% confidence intervals [CI] from the best fitting model was determined.
Results: In the derivative cohort under scenario 4 (n = 300), the distribution pattern from all categories of 3 variables (age,
body weight and estrogen use) stabilized distribution pattern within the fitted model. This model the narrowest 95% CI and
smallest Monte Carlo [MC] errors when compared with scenarios 1 to 3 (n = 150, 200, 250). The scoring system was based on 3 variables of age (in year; <50 = 0, 50 to 59 = 0.5, 60 to 69 = 1, >70 = 1.5), body weight (in kilogram; >60 = 0, 50 to 59 = 1, <50 = 2), and current estrogen use (yes = 0, no = 2), showed a good discriminatory performance in identifying risk of osteoporosis in menopausal women. A score of 3.5 or greater yielded an area under the curve of 0.674 (95% CI = 0.604 to 0.744) with sensitivity of 70.6% (95% CI = 65.4 to 75.4), and specificity of 64.3% (95% CI = 58.8 to 69.7).
Conclusion: The Bayesian logistic regression model is an alternative and effective approach to identify postmenopausal
women at risk for osteoporosis.

Keywords: Bayesian model, Logistic regression, Menopause, Osteoporosis screening