Abstract:

Objective　To analyze the correlation between femoral neck anteversion angle and internal fixation failure of intertrochanteric fracture, and to construct a prediction model of internal fixation failure of intertrochanteric fracture based on machine learning algorithm. Methods　A total of 206 patients with intertrochanteric fracture who received internal fixation surgery in Shangluo Central Hospital from January 2022 to September 2023 were retrospectively selected. Among them, there were 103 males and 103 females, aged (69.75±6.90) years, with a body mass index of (22.63±2.31) kg/m², 152 cases of stable fracture and 54 cases of unstable fracture. According to the follow-up results 12 months after surgery, the patients were divided into two groups: an internal fixation failure group (22 cases) and a normal group (184 cases). The clinical data of the patients were collected, factors of internal fixation failure in patients were analyzed by univariate analysis (independent sample t test and χ² test), and machine learning algorithms including logistic regression, categorical regression tree (CRT), and backpropagation neural network (BPNN) were used to construct a prediction model of internal fixation failure of intertrochanteric fracture. The receiver operating characteristic curve (ROC) was used to compare the predictive values of internal fixation failure models constructed by the three methods. Results　Univariate analysis showed that there were statistically significant differences in the body mass index, osteoporosis, fracture reduction quality, lateral wall thickness, and femoral neck anteversion angle between the two groups (t=3.624, P<0.001; χ²=5.016, P=0.025; χ²=7.529, P=0.023; t=4.464, P<0.001; t=7.602, P<0.001). Multivariate logistic regression analysis showed that body mass index, poor fracture reduction quality, lateral wall thickness, and femoral neck anteversion angle were independent influencing factors for internal fixation failure (OR=1.570, P=0.002; OR=22.315, P=0.009; OR=0.589, P=0.001; OR=2.378, P<0.001). The prediction model constructed by CRT method showed that femoral neck anteversion angle, lateral wall thickness, fracture reduction quality, and body mass index were all classification factors for internal fixation failure. The results of BPNN model showed that the most important influencing factors were femoral neck anteversion angle, lateral wall thickness, body mass index, fracture reduction quality, and osteoporosis. The areas under the curves (AUCs) of logistic regression model, CRT model, and BPNN model were 0.952, 0.919, and 0.950, respectively; the results of non-parametric DeLong test showed that there was no statistically significant difference in the prediction performance among the three models (P_{logistic regression model-CRT model} =0.158, P_{logistic regression model-BPNN model} =0.782, P_{CRT model-BPNN model} =0.219). Conclusions　The internal fixation failure of intertrochanteric fracture is correlated with femoral neck anteversion angle, lateral wall thickness, body mass index, and fracture reduction quality, etc. The three prediction models constructed in this study all show good prediction ability and can be further applied.

Key words:

 , Intertrochanteric fracture of femur, Failure of internal fixation, Anteversion angle of femoral neck, Machine learning algorithm

摘要：

目的　分析股骨颈前倾角与股骨转子间骨折内固定失效的相关性，并基于机器学习算法构建股骨转子间骨折内固定失效的预测模型。方法　回顾性选择2022年1月至2023年9月期间在商洛市中心医院接受内固定手术治疗的206例股骨转子间骨折患者，其中男103例、女103例，年龄（69.75±6.90）岁，体重指数（22.63±2.31）kg/m²，稳定型骨折152例、不稳定型骨折54例。根据术后12个月随访结果将患者分为失效组（22例）和正常组（184例）。收集患者临床资料，单因素分析（独立样本t检验、χ²检验）患者出现内固定失效的因素，使用logistic回归、分类回归树（CRT）、反向传播神经网络（BPNN）的机器学习算法构建股骨转子间骨折内固定失效的预测模型，并采用受试者操作特征曲线（ROC）比较3种模型对内固定失效的预测价值。结果　单因素分析结果显示，两组患者的体重指数、骨质疏松、骨折复位质量、外侧壁厚度、股骨颈前倾角比较，差异均有统计学意义（t=3.624，P<0.001；χ²=5.016，P=0.025；χ²=7.529，P=0.023；t=4.464，P<0.001；t=7.602，P<0.001）。多因素logistic回归分析结果显示，体重指数、骨折复位质量差、外侧壁厚度、股骨颈前倾角均是内固定失效的独立危险因素（比值比=1.570，P=0.002；比值比=22.315，P=0.009；比值比=0.589，P=0.001；比值比=2.378，P<0.001）。采用CRT法构建的预测模型显示，股骨颈前倾角、外侧壁厚度、骨折复位质量、体重指数均是内固定失效的分类因素。BPNN模型结果显示，影响因素重要性排序为股骨颈前倾角、外侧壁厚度、体重指数、骨折复位质量、骨质疏松。logistic回归模型、CRT模型、BPNN模型的曲线下面积（AUC）分别为0.952、0.919、0.950；非参数DeLong检验结果显示，3种模型的预测性能比较，差异均无统计学意义（P_{logistic回归模型-CRT模型}=0.158，P_{logistic回归模型-BPNN模型}=0.782，P_{CRT模型-BPNN模型}=0.219）。结论　股骨转子间骨折患者内固定失效与股骨颈前倾角、外侧壁厚度、体重指数、骨折复位质量相关，构建的3种预测模型均展现了良好的预测能力，值得推广应用。

关键词:

股骨转子间骨折, 内固定失效, 股骨颈前倾角, 机器学习算法