目的：基于Logistic回归模型与决策树模型分析原发性闭角型青光眼(PACG)患者术后发生恶性青光眼(MG)的影响因素。

方法：回顾性收集2020年3月至2025年3月期间于邯郸市眼科医院完成手术且术后6 mo发生MG的PACG患者(发生组)及术后6 mo未发生MG的PACG患者(N-发生组)的资料，进行影响因素分析及模型构建。查阅电子病例系统，收集资料。Logistic回归模型及决策树模型分析PACG患者术后发生MG的影响因素； 受试者工作特征(ROC)曲线分析预测效能，评价临床应用价值。

结果：本研究共纳入PACG患者182例182眼，其中发生组91例91眼，N-发生组91例91眼。发生组中男53例，女38例，年龄≥60岁69例，年龄<60岁22例。N-发生组中男47例，女44例，年龄≥60岁33例，年龄<60岁58例。发生组年龄≥60岁、糖尿病、PACG中期、持续高眼压、房角完全关闭、晶状体厚度<4.5 mm、眼轴长度<22 mm、术后严重炎症的比例均高于N-发生组(均P<0.01)。Logistic回归结果显示，年龄\〖OR(95%CI)=2.136(1.401-3.255)\〗、PACG分期\〖OR(95%CI)=2.996(2.044-4.391)\〗、眼压\〖OR(95%CI)=3.527(1.604-7.755)\〗、房角\〖OR(95%CI)=4.826(2.498-9.324)\〗、眼轴长度\〖OR(95%CI)=5.125(1.265-20.771)\〗、术后严重炎症\〖OR(95%CI)=2.338(1.478-3.699)\〗是术后发生MG的影响因素(均P<0.05)。决策树模型筛选出6个解释变量，分别为年龄、PACG分期、眼压、房角、眼轴长度、术后严重炎症，眼轴长度是发生MG最重要的影响因素。Logistic回归模型与决策树模型预测MG的曲线下面积(AUC)分别为0.913(0.863-0.950)、AUC为0.921(0.872-0.956)，二者比较无统计学差异(Z=0.561，P=0.575)。

结论：决策树和Logistic回归模型从不同层面确定PACG患者术后发生MG影响因素为年龄、PACG分期、眼压、房角、眼轴长度、术后严重炎症，其中决策树模型以可视化的形式展现，预测结果更加直观明了，二者均适用于临床工作。

METHODS:A retrospective study was conducted on PACG patients who underwent surgery at Eye Hospital of Handan City from March 2020 to March 2025. Patients were divided into two groups: the MG group, who developed MG within 6 mo postoperatively, and the non-MG group. Data were collected from the electronic medical record system. Univariate analysis was performed, followed by multivariate logistic regression to identify independent risk factors. A classification and regression tree model was constructed to visualize the hierarchical relationships among predictors. The predictive performance of the two models was evaluated and compared using receiver operating characteristic(ROC)curve analysis.

RESULTS:Totally 182 cases(182 eyes)with PACG were enrolled in this study, including 91 cases(91 eyes)in the MG group and 91 cases(91 eyes)in the non-MG group. In the MG group, there were 53 males and 38 females; 69 cases were aged ≥60 y and 22 cases were aged <60 y. In the non-MG group, there were 47 males and 44 females; 33 cases were aged ≥60 y and 58 cases were aged <60 y. The non-MG group comprised 91 patients, including 47 males and 44 females. Among them, 33 cases were aged ≥60 y, and 58 cases were aged<60 y. The MG group had significantly higher proportions of patients aged ≥60 y, diabetes, moderate-stage PACG, persistent high intraocular pressure(IOP), complete anterior chamber angle closure, lens thickness <4.5 mm, axial length <22 mm, and severe postoperative inflammation compared to the non-MG group(all P<0.01). Multivariate Logistic regression identified the following as independent influencing factors for postoperative MG: age \〖OR (95%CI)=2.136(1.401-3.255)\〗, PACG stage \〖OR (95%CI)=2.996(2.044-4.391)\〗, IOP \〖OR (95%CI)=3.527(1.604-7.755)\〗,anterior chamber angle \〖OR (95%CI)=4.826(2.498-9.324)\〗, axial length \〖OR (95%CI)=5.125(1.265-20.771)\〗, and severe postoperative inflammation \〖OR (95%CI)=2.338(1.478-3.699)\〗(all P<0.05). The decision tree model selected six explanatory variables: age, PACG stage, IOP, anterior chamber angle status, axial length, and severe postoperative inflammation. Axial length was the primary splitting factor at the root node. The areas under the ROC curve(AUC)for the logistic regression and decision tree models were 0.913(0.863-0.950)and 0.921(0.872-0.956), respectively, with no significant difference between them(Z=0.561, P=0.575).

CONCLUSION:Both the logistic regression and decision tree models effectively identify key influencing factors for postoperative MG in PACG patients, including age, PACG stage, IOP, anterior chamber angle status, axial length, and severe postoperative inflammation. The decision tree model offers an intuitive, visual representation of risk stratification, facilitating clinical decision-making. Both models are applicable for clinical risk assessment.