摘要
目的比较随机森林模型与logistic回归模型预测老年股骨转子间骨折患者髓内钉固定术后1年死亡的效能。方法回顾性分析2020年1月至2023年6月期间南阳市第二人民医院骨科收治的192例老年股骨转子间骨折患者资料。男54例,女138例;年龄(78.6±8.3)岁;体重指数为(22.4±2.7)kg/m 2;骨折改良Evans分型:Ⅰ~Ⅲ型73例,Ⅳ、Ⅴ型119例,均采用髓内钉固定治疗。根据术后1年是否死亡将所有患者分为两组:存活组(n=167)和死亡组(n=25)。采用单因素和多因素logistic回归分析患者术后1年死亡的危险因素,基于此使用R4.2.1软件绘制列线图模型及随机森林模型,并绘制两种模型的受试者工作特征曲线(ROC),计算并比较两种模型ROC的曲下线面积(AUC)。结果单因素分析结果显示:两组患者的年龄、体重指数、合并症数量、是否合并冠状动脉粥样硬化性心脏病、骨折改良Evans分型、Charlson合并症指数(CCI)、血红蛋白、白蛋白比较差异均有统计学意义(P<0.05)。多因素logistic回归分析结果显示:年龄>85岁(OR=7.650,95%CI:1.504~38.920,P=0.014)、体重指数>24 kg/m^(2)(OR=4.827,95%CI:1.268~18.370,P=0.021)、合并症数量>2种(OR=6.104,95%CI:1.282~29.060,P=0.023)、CCI≥3分(OR=6.479,95%CI:1.848~22.717,P=0.004)为老年股骨转子间骨折患者髓内钉固定术后1年死亡的危险因素。随机森林模型结果显示:重要性排序依次为CCI、合并症数量、血红蛋白、骨折改良Evans分型、体重指数、年龄、冠状动脉粥样硬化性心脏病、白蛋白。随机森林模型预测老年股骨转子间骨折患者髓内钉固定术后1年死亡的AUC为0.906(95%CI:0.834~0.979),显著大于logistic回归模型[0.783(95%CI:0.673~0.893)],差异有统计学意义(P<0.05)。结论随机森林模型对老年股骨转子间骨折患者髓内钉固定术后1年死亡的预测效能优于logistic回归模型。
Objective To compare random forest model and logistic regression model for their efficacy in prediction of 1-year postoperative mortality in elderly patients with intertrochanteric fracture after intramedullary nail fixation.Methods A retrospective study was conducted to analyze the clinical data of 192 elderly patients with intertrochanteric fracture who had been treated at Department of Orthopedics,The Second People's Hospital of Nanyang between January 2020 and June 2023.The cohort included 54 males and 138 females,with an age of(78.6±8.3)years and a body mass index(BMI)of(22.4±2.7)kg/m 2.According to the modified Evans classification,73 cases belonged to typesⅠtoⅢand 119 ones to typesⅣtoⅤ.All patients were fixated by intramedullary nails and divided into a survival group(n=167)and a death group(n=25)according to absence or presence of 1-year postoperative mortality.After univariate and multivariate logistic regression analyses were employed to identify risk factors for 1-year postoperative mortality in the patients,a logistic regression model and a random forest model were constructed respectively using software R4.2.1.The receiver operating characteristic(ROC)curves were plotted for the 2 models and the areas under the curve(AUCs)were calculated and compared between the 2 models.ResultsUnivariate analysis revealed significant differences between the 2 groups in age,BMI,number of comorbidities,coronary artery disease,modified Evans classification,Charlson Comorbidity Index(CCI),hemoglobin,and albumin levels(P<0.05).Multivariate logistic regression identified age>85 years(OR=7.650,95%CI:1.504 to 38.920,P=0.014),BMI>24 kg/m^(2)(OR=4.827,95%CI:1.268 to 18.370,P=0.021),number of comorbidities>2(OR=6.104,95%CI:1.282 to 29.060,P=0.023),and CCI≥3(OR=6.479,95%CI:1.848 to 22.717,P=0.004)as independent risk factors for 1-year postoperative mortality.The random forest model ranked the feature importance in the following order:CCI,number of comorbidities,hemoglobin,modified Evans classification,BMI,age,coronary artery disease,and albumin.The AUC of the random forest model for prediction of 1-year postoperative mortality was 0.906(95%CI:0.834 to 0.979),showing a significant difference from that of the logistic regression model[0.783(95%CI:0.673~0.893)].ConclusionThe predictive power of the random forest model for 1-year postoperative mortality in elderly patients with intertrochanteric fracture after intramedullary nail fixation is larger than that of the logistic regression model.
作者
王荣强
杨柳
吴向坤
尚立林
Wang Rongqiang;Yang Liu;Wu Xiangkun;Shang Lilin(Department of Orthopaedics,The Second People's Hospital of Nanyang,Nanyang City,Henan 473000,China)
出处
《中华创伤骨科杂志》
北大核心
2025年第12期1060-1066,共7页
Chinese Journal of Orthopaedic Trauma
基金
河南省医学科技攻关计划项目(LHGJ20230962)。
