Use of DXA-derived 3D-modeling, as implemented by 3D-Shaper, for the assessment of fracture risk in a...

Three-dimensional modeling of hip DXA scans has been proposed to partition BMD into cortical surface (csBMD) and trabecular "volumetric" (tvBMD). We assessed in a population-based cohort, whether such partitioning contributes to fracture risk assessment compared to aBMD alone. Participants (N = 4908) from the Rotterdam Study comprised 56% women, with a mean age of 67.4 (SD = 10.1) years. Hip DXA scans were analyzed with enCore (GE Lunar) and 3D-Shaper software. Pearson partial correlation was calculated between aBMD, csBMD, and tvBMD at the total hip, femoral neck, and trochanter. Incident fractures were collected from GP or hospital records. During a mean follow-up of 6.8 (SD = 2.5) years, 171 sustained a hip fracture, and 1019 any-type fractures. Fracture risk estimates were determined as hazard ratios (HR) per SD decrease in BMD, from Cox-regression adjusted for multiple confounders. High correlation (r = >.81; p < .001) was observed between aBMD and the modeled 3D parameters, and between csBMD and tvBMD (r = 0.85; p < .001 at the total hip). Lower aBMD was associated with higher incidence of any-type (HR = 1.60, 95% CI: 1.43-1.78) and hip (HR = 2.46, 95% CI: 1.90-3.17) fracture; lower csBMD with increased risk of any-type (HR = 1.44, 95% CI: 1.30-1.60) and hip (csBMD HR = 1.76, 95% CI: 1.39-2.23) fracture; and tvBMD with increased risk of any-type (HR = 1.62, 95% CI: 1.45-1.80) and hip fracture (HR = 2.44, 95% CI: 1.88-3.15). Inclusion of aBMD in models with tvBMD or csBMD resulted in high multicollinearity and unreliable coefficient parameters (VIF > 5). The effect of csBMD was largely attenuated after inclusion of tvBMD in the same model for both any-type (csBMD HR = 0.96, 95% CI: 0.81-1.13; tvBMD HR = 1.67, 95% CI: 1.41-1.98) and hip (csBMD HR = 0.83, 95% CI: 0.58-1.21; tvBMD HR = 2.82, 95% CI: 1.94-4.10) fractures. Similar results were observed at the neck and intertrochanteric regions. DXA-derived 3D modeled parameters are highly correlated to aBMD and do not provide additional value for the estimation of fracture risk beyond aBMD alone.