Advancing trabecular bone score (TBS): clinical performance of TBS version 4.0 with direct correction for soft tissue thickness-the osteolaus study.

Abstract

This study compared TBS v4.0, which uses DXA-derived tissue thickness corrections, with TBS v3, which adjusts using BMI. TBS v4.0 improved soft tissue adjustments and maintained fracture risk prediction equivalence with TBS v3, enhancing applicability across diverse body compositions/phenotypes. Direct tissue thickness adjustment increases TBS's utility in osteoporosis assessment and management.

Purpose: This study aimed to compare trabecular bone score (TBS) version 4.0, which uses direct tissue thickness correction via DXA measurements, with TBS version 3, which adjusts for soft tissues using body mass index (BMI). The objective was to assess the performance of TBS v4.0 compared to v3, for bone health evaluation and fracture risk assessment across diverse body compositions.

Methods: Data from the OsteoLaus cohort were analyzed. Associations between TBS, BMI, DXA-measured tissue thickness, visceral fat (VFAT), and android fat were examined using regression and correlation analyses. Machine learning, including Random Forest (RF) and SHapley Additive exPlanations (SHAP), explored TBS changes between versions. Five-year fracture risk was assessed using FRAX adjustment, and logistic regression.

Results: TBS v3 correlated with BMI (r = 0.110, p < 0 .001), VFAT mass (r =  - 0.162, p < 0 .001), and soft tissue thickness (r =  - 0.165, p < 0.001). TBS v4.0 demonstrated weaker correlations with BMI (r =  - 0.057, p > 0.999), VFAT Mass (r =  - 0.067, p > 0.779), and soft tissue thickness (r =  - 0.114, p = 0.019). Differences between TBS versions were investigated with SHapley Additive exPlanations (SHAP) and explained by BMI, tissue thickness, VFAT, and gynoid fat. Logistic regression and Delong's test revealed no significant differences in vertebral fracture prediction between the two TBS versions (p = 0.564). FRAX adjustments were highly consistent between versions (r = 0.994, p < 0.001), with no evidence of calibration bias (p = 0.241).

Conclusion: TBS v4.0 enhances the adjustment for regional soft tissue effects and results suggest comparable vertebral fracture risk prediction to TBS v3. Explainable AI provided insights into the contributions of BMI, tissue thickness, visceral fat, and gynoid fat to the observed changes between TBS versions. Incorporating direct tissue thickness adjustment improves TBS applicability across diverse body sizes and compositions.