Abstract
Existing osteoporosis screening tools have limitations, including using race as a predictor, and development on homogeneous samples. This biases risk assessment of osteoporosis in diverse populations and increases health inequities. We develop a tool that relies on variables easily learned during point-of-care, known by individuals, and with negligible racial bias. Data from the 2012-2016 waves of the population-based cohort Health and Retirement Study (HRS) were used to build a predictive model of osteoporosis diagnosis on a 75 % training sample of adults ages 50-90. The model was validated on a 25 % holdout sample and a cross-sectional sample of American individuals ages 50-80 from the National Health and Nutrition Examination Survey (NHANES). Sensitivity and specificity were compared across sex and race/ethnicity. The model has high sensitivity in the HRS holdout sample (89.9 %), which holds for those identifying as female and across racial/ethnic groups. Specificity is 57.9 %, and area under the curve (AUC) is approximately 0.81. Validation in the NHANES sample using empirically measured osteoporosis produced relatively good values of sensitivity, specificity, and consistency across groups. The model was used to create a publicly-available, open-source tool called the Osteoporosis Health Equality (& Equity) Evaluation (OsteoHEE). The model provided high sensitivity for osteoporosis diagnosis, with consistently high results for those identifying as female, and across racial/ethnic groups. Use of this tool is expected to improve equity in screening and increase access to bone density scans for those at risk of osteoporosis. Validation on alternative samples is encouraged.