目的：探讨影响康柏西普治疗糖尿病性黄斑水肿(DME)的因素，并构建列线图预测模型。

方法：回顾性分析。纳入2021年1月至2025年10月收治的DME患者，所有患者均进行每月1次康柏西普治疗，连续治疗3 mo，根据完成3次治疗后1 mo黄斑中心凹厚度(CMT)降低幅度将其分为应答良好组(CMT 降幅≥25%)和应答不佳组(CMT 降幅<25%)。收集两组患者临床资料，分析影响康柏西普治疗DME患者应答不佳的因素，构建预测康柏西普治疗DME患者应答不佳的列线图预测模型。

结果：本研究共纳入DME患者400例400眼，应答良好组312例312眼中男200例，女112例，平均年龄64.85±5.13岁，应答不佳组88例88眼中男54例，女34例，平均年龄65.14±4.93岁。两组患者性别、年龄、是否饮酒、是否吸烟、糖尿病病程、水肿分型、疾病分型、低密度脂蛋白、HRF、ELM和/或EZ连续性中断、SRF比较均无差异(均P>0.05)，而DR病程、治疗前BCVA、高密度脂蛋白、糖化血红蛋白、DRIL、VMIA比较均有差异(均P<0.05)。多因素Logistic回归分析，分析结果显示，DR病程≥1 a(OR=2.351，95%CI：1.360-4.065)、治疗前BCVA差(OR=3.432，95%CI：1.581-6.988)、高糖化血红蛋白(OR=1.656，95%CI：1.346-2.038)、存在DRIL(OR=3.115，95%CI：1.532-6.333)、存在VMIA(OR=2.526，95%CI：1.445-4.415)均是影响康柏西普治疗DME患者应答不佳的危险因素，而高密度脂蛋白升高(OR=0.025，95%CI：0.004-0.163)是康柏西普治疗DME患者应答不佳的保护因素(P<0.05)。ROC分析显示，列线图模型预测康柏西普治疗DME患者应答不佳的AUC值为0.817(95%CI：0.776-0.854)，灵敏度为73.86%，特异度为78.21%，当阈值概率>3%时，列线图模型预测康柏西普治疗DME患者应答不佳风险比对所有患者实施方案更有利。

结论：本研究构建的列线图预测模型有助于临床早期识别康柏西普治疗DME患者应答不佳高风险患者，可为康柏西普治疗应答不佳的早期预防提供参考。

METHODS:Retrospective analysis. Patients with DME admitted from January 2021 to October 2025 were selected. All patients received intravitreal conbercept injection once monthly for three consecutive months. Subjects were divided into a good response group and a poor response group according to the reduction in central macular thickness(CMT)at 3 mo after treatment. The clinical data of the patients were collected to analyze factors associated with poor response to conbercept therapy in patients with DME, and to construct a nomogram prediction model for identifying poor responders to conbercept treatment among DME patients.

RESULTS:A total of 400 DME patients(400 eyes)were included in this study. The response group consisted of 312 patients(312 eyes), with 200 males and 112 females, with an average age of 64.85±5.13 y. The non-response group consisted of 88 patients(88 eyes), with 54 males and 34 females, with an average age of 65.14±4.93 y. There were no differences between the two groups in terms of gender, age, alcohol consumption, smoking status, duration of diabetes, edema classification, disease classification, low-density lipoprotein, HRF, ELM and/or EZ continuity interruption, and SRF(all P>0.05). However, there were differences in DR disease duration, pre-treatment BCVA, high-density lipoprotein, glycated hemoglobin, DRIL, and VMIA(all P<0.05). Multivariate Logistic regression analysis showed that DR disease duration ≥1 y(OR=2.351, 95%CI: 1.360-4.065), poor pre-treatment BCVA(OR=3.432, 95%CI: 1.581-6.988), high glycated hemoglobin(OR=1.656, 95%CI: 1.346-2.038), presence of DRIL(OR=3.115, 95%CI: 1.532-6.333), and presence of VMIA(OR=2.526, 95%CI: 1.445-4.415)were all risk factors for poor response to conbercept treatment in DME patients, while elevated high-density lipoprotein(OR=0.025, 95%CI: 0.004-0.163)was a protective factor for poor response to conbercept treatment in DME patients(P<0.05). ROC analysis showed that the area under the curve(AUC)of the nomogram model for predicting poor response to conbercept treatment in DME patients was 0.817(95%CI: 0.776-0.854), with a sensitivity of 73.86% and a specificity of 78.21%. When the threshold probability was >3%, the nomogram model for predicting poor response to conbercept treatment in DME patients was more favorable for all patient treatment plans.

CONCLUSION:The nomogram prediction model constructed in this study is helpful for the clinical early identification of patients with DME who have a high risk of poor response to conbercept treatment, and can provide a reference for the early prevention of poor response to conbercept treatment.