老年髋部骨折患者术后低蛋白血症预测模型的构建
Establishment of Prediction Model of Postoperative Hypoproteinemia in the Elderly Patients with Hip Fractures

目的：严重低蛋白血症常常发生在老年髋部骨折术后的患者中，而术后低蛋白血症往往伴随着较差的预后。本研究旨在寻找与术后低蛋白血症相关的危险因素，建立评估术后中重度低蛋白血症风险的预后预测模型。以此指导围手术期老年髋部骨折患者的中重度低蛋白血症防治，促进早期的功能康复，改善最终的临床预后。方法：回顾性分析了2017年6月至2022年6月在重庆医科大学第二附属医院住院治疗的年龄 ≥ 65岁髋部骨折患者的临床资料(年龄、性别、骨折类型、术前合并症、受伤至手术的时间、ASA分级、麻醉方式、手术时长、围术期是否输血、实验室检查指标(血红蛋白、血清白蛋白))，使用单因素分析及多因素二元logistic回归筛选出模型的最佳预测因子，应用R软件将多因素回归分析结果带入其中进行Nomogram预测模型的构建。我们采用受试者工作特征(Receiver Operating Characteristic, ROC)曲线下面积(Area Under Curve, AUC)评估模型的区分度，绘制校准曲线并采用Hosmer-Lemeshow拟合优度检验评估模型的校准度，绘制临床决策分析曲线评估模型的临床应用价值。结果：① 多因素二元logistic回归分析表明术前血红蛋白、术前血清白蛋白、手术时间、骨折类型、手术方式、围术期是否输血这6个指标是老年髋部骨折患者术后低蛋白血症的独立危险因素；② 基于上述指标建立Nomogram预测模型，该模型对老年髋部骨折患者术后低蛋白血症有着良好的预测效果，AUC为0.841 (95% CI 0.807~0.876)，校准曲线图表现出了良好的一致性，Hosmer-Lemeshow拟合优度检验表现出了较好的拟合度，临床决策分析曲线表明该模型具有较好的临床应用价值。结论：基于术前血红蛋白、术前血清白蛋白、手术时间、骨折类型、手术方式、围术期是否输血构建的Nomogram预测模型可有效预测老年髋部骨折患者术后低蛋白血症风险，可为临床指导老年髋部骨折术后低蛋白血症的预防和治疗提供重要参考。
Objective: Severe hypoproteinemia often occurs in elderly patients with hip fracture after operation, and postoperative hypoproteinemia is often accompanied by poor prognosis. The purpose of this study is to find out the risk factors related to postoperative hypoproteinemia and establish a prognosis prediction model to evaluate the risk of moderate and severe hypoproteinemia after operation. In order to guide the prevention and treatment of moderate and severe hypoproteinemia in elderly patients with hip fracture during perioperative period, promote early functional rehabilitation and improve the final clinical prognosis. Methods: The clinical data (age, sex, fracture type, preoperative complications, time from injury to operation, ASA classification, anesthesia mode, operation duration, perioperative blood transfusion, laboratory examination indicators (hemoglobin, serum albumin)) of patients with hip fracture who were hospitalized in the Second Affiliated Hospital of Chongqing Medical University from June 2017 to June 2022 were retrospectively analyzed. Single factor analysis and multivariate binary logistic regression are used to screen out the best predictor of the model, and R software is used to bring the results of multivariate regression analysis into it to build the Nomogram prediction model. We use the area under curve (AUC) of receiver operating characteristic (ROC) to evaluate the discrimination of the model, draw a calibration curve and use Hosmer-Lemeshow to test the calibration of the model, and draw a clinical decision analysis curve to evaluate the clinical application value of the model. Results: ① Multivariate logistic regression analysis showed that preoperative hemoglobin, preoperative