目的 分析脊髓损伤患者并发神经源性膀胱功能障碍（neurogenic bladder dysfunction， NB）的危险因素，另通过决策树算法建立脊髓损伤患者并发NB的风险预测模型。 方法 回顾性分析2022年4月至2024年7月收治的176例脊髓损伤患者的临床资料，根据脊髓损伤患者是否并发NB将患者分为障碍组和非障碍组，采用多因素logistic回归分析筛选NB的危险因素，运用Modeler软件构建脊髓损伤患者并发NB的决策树模型，采用5折交叉验证法对模型进行内部验证，并对比模型预测效能。 结果 176例脊髓损伤患者中，有42例患者并发NB，发生率为23.86%；logistic回归分析显示，脊髓损伤平面（T₁₀—L₂）、脊髓损伤程度（完全损伤）、病程（≥ 6个月）、膀胱顺应（异常）、泌尿系统感染（有）、逼尿肌括约肌失调（是）均是脊髓损伤患者并发NB的独立危险因素（P < 0.05）；概率预测模型P = 1/［1 + e-^{（-6.008+0.791*X1+3.117*X2+1.492*X3+1.270*X4+1.516*X5+2.158*X6）}］，模型预测总体正确性为80.5%；经5折交叉验证显示，模型预测正确率为71.7%；决策树模型显示，脊髓损伤程度对脊髓损伤患者并发NB的影响最大，信息增益为0.46；ROC结果显示，两种模型预测NB的AUC值接近（0.873 vs. 0.852，Z = 0.875，P = 0.469）。 结论 脊髓损伤平面、脊髓损伤程度、病程、膀胱顺应、泌尿系统感染、逼尿肌括约肌失调均可预测NB发生风险，该研究构建的决策树模型可有效预测脊髓损伤患者并发NB的风险概率，医务人员可根据上述因素制定针对性干预方案，以降低NB发生风险。

Objective To analyze the risk factors of neurogenic bladder dysfunction （NB） in patients with spinal cord injury， and establish a risk prediction model of NB in patients with spinal cord injury by decision tree algorithm. Method Clinical data of 176 patients with spinal cord injury admitted from April 2022 to July 2024 were retrospectively analyzed. Patients with spinal cord injury were divided into disorder group and non-disorder group according to whether they were complicated by NB. Multivariate Logistic regression analysis was used to screen the risk factors of NB. Modeler software was used to construct the decision tree model of spinal cord injury patients with concurrent NB， and the 5-fold cross-validation method was used to internally verify the model， and the prediction efficiency of the model was compared. Results Among 176 patients with spinal cord injury， 42 patients had concurrent NB， the incidence of NB was 23.86%. Logistic regression analysis showed that the level of spinal cord injury （T₁₀—L₂）， degree of spinal cord injury （complete injury）， course of disease （≥ 6 months）， bladder compliance （abnormal）， urinary system infection （yes） and detrusor sphincter disorder （yes） were all independent risk factors for NB in patients with spinal cord injury （P < 0.05）. Probability forecasting model P = 1/［1 + e- ^{（-6.008+0.791*X1+3.117*X2+1.492*X3+1.270*X4+1.516*X5+2.158*X6）}］， models to predict the overall accuracy is 80.5%； The prediction accuracy of the model is 71.7% through the cross-verification of 5 fold. Decision tree model showed that the degree of spinal cord injury had the greatest effect on the complication of NB in patients with spinal cord injury， and the information gain was 0.46. ROC results showed that the AUC values of NB predicted by the two models were close （0.873 vs. 0.852， Z = 0.875， P = 0.469）. Conclusion The level of spinal cord injury， degree of spinal cord injury， course of disease， bladder compliance， urinary system infection， detrusor sphincter disorder can all predict the risk of NB. The decision tree model constructed in this study can effectively predict the risk probability of NB in patients with spinal cord injury， and medical staff can make targeted plans according to the above factors to reduce the risk of NB.