抗PD-1/PD-L1治疗心血管毒性死亡风险因素分析及预测模型构建

Analysis of Risk Factors and Construction of Predictive Models for Cardiovascular Toxic Death in Anti-PD-1/PD-L1 Therapy

摘要: 目的：本研究目的旨在探讨抗PD-1/PD-L1 (程序性细胞死亡蛋白-1，Programmed cell death-1/程序性死亡配体-1，Programmed death-ligand 1)治疗后发生心血管毒性死亡的风险因素，并以此构建预测模型。方法：本研究为单中心回顾性研究，通过系统性回顾方法筛选2018年10月至2023年10月在青岛大学附属医院接受抗PD-1/PD-L1免疫治疗后的2665例实体肿瘤患者的病例资料观察180天，其中发生心血管毒性的33例，根据是否发生抗PD-1/PD-L1治疗心血管毒性死亡为结局，以是否发生死亡分为死亡组和存活组。采用单因素Cox回归分析其风险因素并控制相关混杂因素后构建预测模型绘制列线图。最后，利用受试者工作特征曲线(Receiver Operating Characteristic, ROC)、决策曲线分析法(Decision Curve Analysis, DCA)、校准曲线(Calibration Curve, CC)进行内部评价和内部验证。结果：1) 淋巴细胞计数、单核细胞计数、血小板计数、合并糖尿病、免疫检查点抑制剂种类、彩超室壁运动异常在心血管毒性存活组和死亡组间的差异均有统计学意义(均p < 0.05)。2) 单因素Cox回归分析结果显示合并糖尿病(HR = 6.03, 95% CI为1.67~21.77, p = 0.006)、免疫检查点抑制剂种类(HR = 6.62, 95% CI为1.69~25.89, p = 0.007)、彩超室壁运动异常(HR = 4.61, 95% CI为1.19~17.85, p = 0.027)、单核细胞数(HR = 0.02, 95% CI为0.00~0.49, p = 0.015)是心血管毒性死亡发生的预测因素。3) 多因素Cox回归分析结果显示，单核细胞数(HR = 0.02, 95% CI为0.00~0.77, p = 0.036)为心血管毒性死亡发生的独立保护因素。4) 将单因素Cox回归4个变量(p < 0.05)纳入列线图中构建心血管毒性死亡风险预测模型，AUC为0.88，95% CI为0.75~1.00。结论：1) 合并糖尿病、免疫检查点抑制剂种类、彩超室壁运动异常、单核细胞数是心血管毒性死亡发生的预测因素。2) 单核细胞数是心血管毒性死亡发生的独立保护因素。3) 将单因素Cox回归4个变量纳入列线图中构建心血管毒性死亡风险预测模型，此模型具有良好的精确性、区分度、准确性及临床受益效果。

Abstract: Purpose: This study aimed to investigate risk factors for death from cardiovascular toxicity following anti-PD-1/PD-L1 therapy and develop a predictive model. Methods: This study was a single-center retrospective study, which screened the case data of 2665 patients with solid tumors after receiving anti-PD-1/PD-L1 immunotherapy at the Affiliated Hospital of Qingdao University for 180 days of observation by a systematic retrospective method, 33 cases of cardiovascular toxicity, and the outcome was based on whether or not death from cardiovascular toxicity of anti-PD-1/PD-L1 therapy occurred as the outcome categorized into death and survival groups. A one-way Cox regression analysis was used to analyze the risk factors and control for relevant confounders, and a predictive model was constructed to draw a column-line graph. Finally, Receiver Operating Characteristic (ROC), Decision Curve Analysis (DCA), and Calibration Curve (CC) were used for internal evaluation and internal validation. Results: 1) Lymphocyte, Platelet, Diabetes, Types of immune checkpoint inhibitors, Echocardiographic ventricular wall motion abnormalities and monocyte differences between two groups: those with death from cardiovascular toxicity and those without were statistically significant (all p < 0.05). 2) Univariate Cox regression identified diabetes (HR = 6.03, 95% CI 1.67~21.77, p = 0.006), Types of immune checkpoint inhibitors (HR = 6.62, 95% CI 1.69~25.89, p = 0.007), Echocardiographic ventricular wall motion abnormalities (HR = 4.61, 95% CI 1.19~17.85, p = 0.027), monocyte HR = 0.02, 95% CI 0.00~0.49, p = 0.015) as significant predictors. 3) Multivariate analysis confirmed monocyte (HR = 0.02, 95% CI 0.00~0.77, p = 0.036) as an independent predictor. 4) A predictive model for the risk of death from cardiovascular toxicity was constructed by including 4 variables (p < 0.05) from univariate Cox regression in a nomogram, with an AUC of 0.88 and 95% CI of 0.75 to 1.00. Conclusion: 1) Diabetes, Types of immune checkpoint inhibitors, Echocardiographic ventricular wall motion abnormalities and monocyte were significant predictors of death from cardiovascular toxicity. 2) Monocyte was an independent protective factor, after adjusting for other covariates. 3) A predictive model for the risk of cardiovascular toxic death was constructed by incorporating 4 variables from univariate Cox regression into a nomogram, and this model had good precision, discrimination, accuracy, and clinical benefit effects.