Data Sheet 1_Trabecular texture and paraspinal muscle characteristics for prediction of first vertebral fracture: a QCT analysis from the AGES cohort.docx

IntroductionVertebral fractures (VFs) significantly increase risk of subsequent fractures. Areal bone mineral density (BMD) assessed by DXA and volumetric BMD by QCT, are strong predictors of VF. Nevertheless, risk prediction should be further improved. This study used data from the Age, Gene/Environment Susceptibility Reykjavik (AGES-Reykjavik) cohort to evaluate whether trabecular texture and paraspinal muscle assessments improve the prediction of the first incident VF.
MethodsCT scans of the L1 and L2 vertebrae of 843 elderly subjects; including 167 subjects with incident, VFs occurring within a 5-year period and 676 controls without fractures. Image analysis included measurement of BMD, cortical thickness and of parameters characterizing trabecular architecture and the autochthonous muscles. Fifty variables were used as predictors, including a BMD, a trabecular texture and a muscle subset. Each included age, BMI and corresponding parameters of the QCT analysis. The number of variables in each subset was reduced using stepwise logistic regression to create multivariable fracture prediction models. Model accuracy was assessed using the likelihood ratio test (LRT) and the area under the curve (AUC) criteria. Bootstrap analyses were performed to assess the stability of the model selection process.
Results96 women and 78 men with prior VF were excluded. Of 50 initial predictors, 17 were significant for women and 11 for men. Bone and texture models showed significantly better fracture prediction in women (p<0.001) and men (p<0.01) than the combination of age and BMI. The muscle model showed better fracture prediction in men only (p<0.03). Compared to the BMD model alone, LRT showed a significantly improved VF prediction of the combinations of BMD with texture (women and men) (p<0.05) or with muscle models (men only) (p=0.03) but no significant increases in AUC values (AUC women: Age&BMI: 0.57, BMD: 0.69, combined model: 0.69; AUC men: Age&BMI: 0.63, BMD: 0.71, combined models 0.73-0.77)
DiscussionTrabecular texture and muscle parameters significantly improved prediction of first VF over age and BMI, but improvements were small compared to BMD, which remained the primary predictor for both sexes. Although muscle measures showed some predictive power, particularly in men, their clinical significance was marginal. Integral BMD should remain the focus for fracture risk assessment in clinical practice.

引言
椎体骨折（Vertebral Fractures, VFs）会显著提升后续骨折的发生风险。通过双能X线吸收法（DXA）检测的面积骨密度（Areal Bone Mineral Density, BMD）以及定量CT（QCT）检测的体积骨密度，均为椎体骨折的强预测因子。然而，骨折风险预测模型仍有待进一步优化。本研究借助年龄、基因/环境易感性雷克雅未克（AGES-Reykjavik）队列的数据，评估骨小梁纹理与椎旁肌肉评估是否可提升首次发生椎体骨折的预测效能。

方法
本研究纳入843名老年受试者的腰椎1（L1）与腰椎2（L2）CT扫描影像，其中167名受试者在5年内发生了首次椎体骨折（病例组），剩余676名受试者无骨折史（对照组）。影像分析内容包括骨密度、皮质厚度以及表征骨小梁结构与椎旁固有肌的参数。研究共选取50个变量作为预测因子，涵盖骨密度、骨小梁纹理、肌肉特征三个子集，每个子集均包含年龄、体质量指数（BMI）以及对应的定量CT分析参数。通过逐步logistic回归缩减各子集的变量数量，以构建多变量骨折预测模型。模型的预测效能采用似然比检验（Likelihood Ratio Test, LRT）与曲线下面积（Area Under the Curve, AUC）进行评估。同时开展Bootstrap分析，以评估模型选择过程的稳定性。

结果
本研究排除了96名女性与78名男性既往存在椎体骨折的受试者。在初始的50个预测因子中，女性有17个具有统计学意义，男性则有11个。相较于仅纳入年龄与BMI的模型，骨密度联合骨小梁纹理的模型对女性（p<0.001）与男性（p<0.01）的骨折预测效能均显著更优。仅在男性群体中，肌肉特征模型的骨折预测效能更佳（p<0.03）。相较于单独使用骨密度的模型，联合骨密度与骨小梁纹理的模型（男女群体均适用，p<0.05）或联合骨密度与肌肉特征的模型（仅适用于男性，p=0.03）的椎体骨折预测效能经似然比检验显示显著提升，但曲线下面积未出现显著变化（女性群体：年龄&BMI模型AUC=0.57，骨密度模型AUC=0.69，联合模型AUC=0.69；男性群体：年龄&BMI模型AUC=0.63，骨密度模型AUC=0.71，联合模型AUC=0.73~0.77）。

讨论
骨小梁纹理与肌肉参数相较于年龄和BMI，显著提升了首次椎体骨折的预测效能，但相较于作为两性骨折风险首要预测因子的骨密度，其提升幅度较小。尽管肌肉测量指标具备一定预测价值，尤其在男性群体中，但其临床意义较为有限。临床实践中，骨密度仍应作为骨折风险评估的核心指标。