3D-DXA-based Finite Element Models Improve Hip Fracture Prediction versus aBMD in the Japanese Population-Based Osteoporosis (JPOS) Cohort Study
In the authors' words
ABSTRACT Summary Hip fractures are among the most serious consequences of osteoporosis, yet many individuals who sustain hip fractures do not meet the diagnostic threshold based on areal bone mineral density (aBMD). Finite element (FE)-based mechanical descriptors may improve hip fracture prediction beyond aBMD. Purpose This study investigated whether FE-based mechanical descriptors derived from dual-energy X-ray absorptiometry (DXA) improve hip fracture discrimination beyond aBMD in the Japanese Population-Based Osteoporosis (JPOS) cohort. Methods The study included 1,636 women, of whom 85 sustained hip fractures during 20 years of follow-up. Subject-specific three-dimensional femoral models were reconstructed from DXA images using statistical shape and appearance modelling, and nonlinear FE analyses were performed to derive mechanical descriptors under simulated sideways-fall loading. Predictive performance was evaluated using discrimination, model fit, reclassification, and hazard ratio analyses. Results Over two-thirds of women who experienced hip fracture were not classified as osteoporotic based on aBMD. Several FE-based descriptors significantly improved fracture prediction relative to total hip and femoral neck aBMD across the evaluation frameworks. The best-performing descriptor, combining femoral strength with residual energy (F_U + W_R), achieved a significantly higher AUROC of 0.797 compared with 0.763 (p = 0.026) and 0.757 (p = 0.040) for total hip and femoral neck aBMD, respectively, and the highest hazard ratio (3.577 per SD). The same descriptors consistently performed best across all evaluation methods. Conclusion These findings demonstrate that FE-based descriptors provide meaningful information beyond aBMD and may improve identification of women at elevated hip fracture risk.
Appeared: Thursday, November 19. medRxiv. Preprint, not yet peer-reviewed.