Abstract
Despite the profound prevalence and fracture risk of osteoporosis, access to the gold standard DXA scans remains limited, especially in rural communities. Rho is an artificial intelligence software that can identify individuals at risk of low BMD and osteoporosis using radiographs. This study will independently validate Rho software by evaluating its performance against DXA in a retrospective cohort of patients. We conducted a retrospective study of 4878 patients (mean age 70 ± 10, 80% female), with DXA acquired within 1 yr of a radiograph. The area under the curve (AUC) was calculated to evaluate the performance of Rho in identifying patients at risk for low BMD (T-Score < -1) and osteoporosis (T-Score ≤ -2.5). Further subgroup analyses were performed based on radiograph location, sex, and rural vs urban populations. The overall AUC for predicting low BMD was 0.840 (95% CI: 0.831-0.848), with an optimal Rho score threshold of 6. For osteoporosis prediction, the AUC was 0.815 (95% CI: 0.806-0.824), with an optimal Rho score threshold of 7. Rural and urban populations have strong AUCs for low BMD (AUC = 0.873; 0.873) and osteoporosis (AUC = 0.865; 0.812). Likewise, Rho demonstrated strong, comparable (p > .25) performance in both men and women for prediction of low BMD. Although, optimal cutoffs differed between females and males for both low BMD and osteoporosis. Rho demonstrated high effectiveness in identifying patients at risk for low BMD and osteoporosis. The findings support Rho as an opportunistic screening tool and may fill a clinical gap in communities lacking access to DXA.