摘要目的:构建并验证中老年健康体检人群骨质疏松（OP）风险预测模型。方法:本研究为横断面研究，选取2020年1月至2022年12月在广西壮族自治区人民医院健康管理中心进行骨密度检测的18 030例中老年人群为研究对象，收集了受试者的一般资料、体格检查指标及生化血液指标。采用简单随机抽样方法按7∶3比例分为训练集（12 621例）和验证集（5 409例），用最小LASSO回归联合logistic回归对变量进行筛选并建立相应的中老年健康体检人群骨质疏松症发病风险列线图预测模型。采用受试者工作特征（ROC）曲线下面积（AUC）、特异度、灵敏度、校准曲线（CAL）及决策曲线（DCA）分析评估该列线图模型的效能。结果:训练集LASSO回归联合多因素logistic回归结果显示，性别、年龄、体重指数（BMI）、臀围、腰围、收缩压、总胆固醇（TC）、γ-谷氨酰转移酶（GGT）和白蛋白/球蛋白比值（A/G）是筛选出的9个中老年健康体检人群OP风险独立最佳预测因子（均 P<0.05）。训练集AUC值为0.895（95% CI：0.886~0.904），其灵敏度为87.25%，特异度为85.01%。验证集AUC值为0.892（95% CI：0.886~0.898），灵敏度为83.74%，特异度为82.46%。CAL显示训练集C-index值为0.790，验证集C-index值为0.784，CAL均显示出偏差校正和与理想线相似的明显曲线。DCA显示训练集OP风险阈值概率在45%~93%时，验证集OP风险阈值概率在45%~92%时，列线图模型预测中老年体检人群发生OP风险的效能更好，二者结果仍较为一致。CAL和DCA均表现出良好性能。 结论:本研究构建了一种实用性强的中老年人群发生OP风险预测模型，能为中老年人群及时发现OP风险提供预警。

abstractsObjective:To construct and validate the risk prediction model of osteoporosis (OP) in the middle-aged and elderly healthy physical examination population.Methods:In this cross-sectional study, 18 030 middle-aged and elderly people with bone mineral density tested in Health Management Center of Guangxi Zhuang Autonomous Region Hospital from January 2020 to December 2022 were selected. The general data, physical examination index and biochemical blood index were collected. The subjects were divided into training set (12 621 cases) and validation set (5 409 cases) in a ratio of 7∶3 with the simple random sampling method. The variables were screened with minimum LASSO regression and logistic regression and the corresponding nomogram prediction model for the risk of osteoporosis in the middle-aged and elderly health examination population was established. The performance of the nomogram model was evaluated with the area under the receiver operating characteristic curve (ROC AUC), specificity, sensitivity, calibration curve (CAL), and decision curve (DCA).Results:The results of LASSO regression and multivariate logistic regression in training set showed that gender, age, body mass index, hip circumference, waist circumference, systolic blood pressure, total cholesterol, glutamyl transpeptidase and albumin/globulin ratio were the independent best predictors of OP risk in the middle-aged and elderly health examination population (all P<0.05). The ROC AUC-value of the training set was 0.895 (95% CI: 0.886-0.904), with a sensitivity of 87.25% and a specificity of 85.01%. The ROC AUC value of the validation set was 0.892 (95% CI: 0.886-0.898), with a sensitivity of 83.74% and a specificity of 82.46%. The CAL showed a C-index value of 0.790 in the training set and a C-index value of 0.784 in validation set. The CALs all showed deviation correction and obvious curves similar to the ideal line. DCA showed that when the OP risk threshold probability of the training set was 45%-93%, and the OP risk threshold probability of the validation set was 45%-92%, the nomogram model had better efficacy in predicting OP risk in the middle-aged and elderly physical examination population, and the two results were still relatively consistent. Both CAL and DCA showed good performance. Conclusion:This study establishes a practical prediction model for osteoporosis risk in the middle-aged and elderly population, it can provide an early warning for the timely detection of OP risk for the middle-aged and elderly people.