目的：探究糖尿病视网膜病变(DR)患者抗血管内皮生长因子(VEGF)治疗反应不佳的影响因素，并基于影响因素构建预测模型，为临床个体化治疗提供参考依据。

方法：回顾性分析2022年7月至2025年8月于医院接受抗VEGF治疗的DR患者临床资料，按照7：3的比例随机划分成训练集和验证集。训练集患者根据3次抗VEGF治疗后1 mo的治疗反应分为反应不佳组与反应良好组。收集患者基本资料，通过单因素和多因素Logistic回归分析DR患者抗VEGF治疗反应不佳的影响因素，基于影响因素构建列线图预测模型，绘制校正曲线、受试者工作特征(ROC)曲线对模型进行验证评估，并采用决策曲线评估列线图模型的临床实际获益情况。

结果：本研究共纳入DR患者1 250例1 250眼，训练集875例875眼(年龄60.82±10.54岁，男262例，女613例)和验证集375例375眼(年龄59.70±10.61岁，男100例，女275例)。训练集患者中反应不佳组266例266眼(年龄61.33±9.92岁，男82例，女184例)与反应良好组609例609眼(年龄60.59±10.80岁，男180例，女429例)。训练集与验证集患者一般资料、治疗反应不佳率比较无差异(均P>0.05)。多因素Logistic回归分析结果显示，糖尿病黄斑水肿(DME)分型-浆液性视网膜脱离、治疗前黄斑中心凹视网膜厚度(CMT)、治疗前最佳矫正视力(BCVA)、椭圆体带(EZ)破坏、治疗前糖化血红蛋白(HbA1c)、治疗前中性粒细胞计数均为DR患者抗VEGF治疗反应不佳的危险因素(均OR>1，P<0.05)。基于危险因素绘制列线图风险模型，训练集预测抗VEGF治疗反应不佳的C-index为0.880(95%CI：0.855-0.904)，验证集C-index为0.867(95%CI：0.828-0.906)； 绘制ROC曲线，训练集、验证集预测模型曲线下面积(AUC)分别为0.884(95%CI：0.859-0.908)、0.880(95%CI：0.841-0.919)，提示模型区分度良好； 决策曲线显示，训练集、验证集阈值在0.06-0.99范围内的净受益率大于0，在该阈值概率范围内，使用本模型指导临床决策可获得正向净获益。

结论：DME分型-浆液性视网膜脱离、治疗前CMT、治疗前BCVA、EZ破坏、治疗前HbA1c、治疗前中性粒细胞计数水平均为DR患者抗VEGF治疗反应不佳的危险因素，基于此构建的列线图风险预测模型，具有较高的预测效能，能够为临床早期制定针对性干预对策提供参考依据。

METHODS: A retrospective analysis was conducted on the clinical data of DR patients who received anti-VEGF treatment in the hospital from July 2022 to August 2025. The patients were randomly divided into a training set and a validation set in a 7:3 ratio. Patients in the training set were divided into a poor response group(n=266)and a good response group(n=609)based on their treatment response 1 mo after 3 sessions of anti-VEGF therapy. The basic information of patients was collected. The influencing factors of poor response to anti-VEGF treatment in DR patients were analyzed through univariate and multivariate Logistic regression. A nomogram prediction model was constructed based on the influencing factors. Based on the identified influencing factors, a nomogram prediction model was constructed. The model was validated and evaluated by calibration curves and receiver operating characteristic(ROC)curves. Decision curve analysis was adopted to assess the clinical net benefit of the nomogram model.

RESULTS:This study included 1 250 DR patients(1 250 eyes), 875 training subjects(age 60.82±10.54 y, 262 males and 613 females), and 375 validation subjects(age 59.70±10.61 y, 100 males and 275 females). Among the patients in the training set, there were 266 cases(266 eyes, age 61.33±9.92 y, 82 males, 184 females)with poor response and 609 cases(609 eyes, age 60.59±10.80 y, 180 males, 429 females)with good response. No statistically significant differences were observed in baseline patient characteristics or treatment response rate between the training set and validation set(P>0.05). Multivariate Logistic regression analysis showed that the classification of diabetic macular edema(DME)-serous retinal detachment, central macular thickness(CMT)before treatment, best corrected visual acuity(BCVA)before treatment, disruption of ellipsoidal zone(EZ), glycosylated hemoglobin(HbA1c)before treatment, and neutrophil count before treatment were all risk factors for poor response to anti-VEGF treatment in DR patients(all OR>1, P<0.05). A nomogram risk model was plotted based on risk factors. The C-index of the training set for predicting poor response to anti-VEGF treatment was 0.880(95%CI: 0.855-0.904), and that of the validation set was 0.867(95%CI: 0.828-0.906). The ROC curves were plotted. The area under the curve(AUC)of the prediction model in the training set and validation set was 0.884(95%CI: 0.859-0.908)and 0.880(95%CI: 0.841-0.919), respectively, suggesting that the model had good discrimination. The decision curve showed that the net benefit rate of the training set and the verification set threshold in the range of 0.06-0.99 was greater than 0. Within the threshold probability range, this model for clinical decision-making can obtain positive net benefits.

CONCLUSION: DME serous retinal detachment subtype, pre-treatment CMT, pre-treatment BCVA, EZ disruption, pre-treatment HbA1c, and pre-treatment neutrophil count levels are all risk factors for poor anti-VEGF treatment response in DR patients. The nomogram risk prediction model constructed based on it has high predictive power and can provide a reference for the early development of targeted intervention strategies in clinical practice.