中国药物警戒 ›› 2023, Vol. 20 ›› Issue (6): 634-638.
DOI: 10.19803/j.1672-8629.20230106

• 上市后药品不良反应监测定量统计学方法研究专栏 • 上一篇    下一篇

基于主动监测系统的免疫检查点抑制剂心肌炎不良反应探索

郑轶1, 罗枭1, 张朋朋2, 刘永梅2, 叶小飞1, 郭晓晶1, 陈霞3#, 贺佳1,*   

  1. 1海军军医大学卫勤系军队卫生统计学教研室,上海 200433;
    2联勤保障部队天津康复疗养中心,天津 300191;
    3重庆大学附属肿瘤医院,重庆 400030
  • 收稿日期:2023-02-27 出版日期:2023-06-15 发布日期:2023-06-15
  • 通讯作者: * 贺佳,女,博士,教授·博导,药物流行病学及新药评价。E-mail: hejia63@yeah.net。#为共同通信作者。
  • 作者简介:郑轶,男,硕士,主治医师,药物流行病学。
  • 基金资助:
    国家自然科学基金资助项目（82073671）; 中国药学会药物临床评价研究专业委员会研究课题（CPA-CDCER- 2021-001）

Adverse reactions of immune checkpoint inhibitor myocarditis based on active monitoring system

ZHENG Yi1, LUO Xiao1, ZHANG Pengpeng2, LIU Yongmei2, YE Xiaofei1, GUO Xiaojing1, CHEN Xia3#, HE Jia1,*   

  1. 1Department of Health Statistics, Naval Military Medical University, Shanghai 200433, China;
    2Tianjin Rehabilitation Center of Joint Logistic Support Force, Tianjin 300191, China;
    3Chongqing University Affiliated Cancer Hospital, Chongqing 400030, China
  • Received:2023-02-27 Online:2023-06-15 Published:2023-06-15

PDF (PC)

128

摘要： 目的 以免疫检查点抑制剂心肌炎不良反应为例,初步探索基于中国医院药物警戒系统开展药品不良反应主动监测的可行性,为相关研究提供参考。方法 基于2018年6月1日至2022年6月1日重庆某哨点医院的中国医院药物警戒系统数据,提取应用过免疫检查点抑制剂的肿瘤患者信息,以及未使用过免疫检查点抑制剂的肿瘤患者信息作为对照,并采用倾向性评分匹配1∶4的方法将2组间已观测到的混杂因素进行控制。基于几种常用的机器学习算法和Logistic回归构建心肌炎的预测模型,选择预测效能最佳的模型作为心肌炎的预测模型,对患者是否患有心肌炎进行识别,随后将2组进行对比,探索免疫检查点抑制剂是否会增加心肌炎的发生风险。结果 共纳入15 589名患者,其中免疫检查点抑制剂组3 496名,对照组12 083名。构建的心肌炎预测模型中,随机森林的预测效能最佳（AUC=0.948, ACC=0.988, 精准率=1.000, 召回率=0.545, F1分数=0.706）,将其作为心肌炎的预测模型。基于该模型对纳入研究的患者是否发生心肌炎进行识别,其中免疫检查点抑制剂组发生心肌炎64名（1.83%）,对照组有160名（1.32%）,2组间的发生率差异P<0.05,有统计学意义。结论 免疫检查点抑制剂的应用会增加心肌炎的发生风险,临床医师在给患者使用免疫检查点抑制剂时应注意心肌炎的发生,确保患者用药安全。

关键词: 免疫检查点抑制剂, 心肌炎, 肿瘤, 中国医院药物警戒系统, 主动监测, 机器学习, 药品不良反应

Abstract: Objective To explore the feasibility of active surveillance of adverse drug reactions based on the China Hospital Pharmacovigilance System, taking adverse reactions to immune checkpoint inhibitors in myocarditis as an example, and to provide a corresponding basis for related studies. Methods Based on data from the China Hospital Pharmacovigilance System at a sentinel hospital in Chongqing between June 1, 2018 to June 1, 2022, information on tumor patients who had applied immune checkpoint inhibitors and tumor patients who had not used immune checkpoint inhibitors were extracted as controls, and the observed confounding factors between the two groups were controlled by using propensity score matching 1 : 4. Predictive models for myocarditis were constructed based on several commonly used machine learning algorithms and logistic regression, and the model with the best predictive efficacy was selected as the predictive model for myocarditis to identify whether patients had myocarditis, and the two groups were subsequently compared to explore whether immune checkpoint inhibitors increase the risk of myocarditis. Results A total of 15 589 patients were included in this study, including 3 496 in the immune checkpoint inhibitor group and 12 083 in the control group. The best predictive efficacy of random forest (AUC=0.948, ACC=0.988, precision=1.000, recall=0.545, F1 score=0.706) was used as a predictive model for myocarditis. Based on this model to identify whether myocarditis occurred in the patients included in the study, 64 (1.83%) of the immune checkpoint inhibitor group had myocarditis and 160 (1.32%) of the control group, the difference in incidence between the two groups was P<0.05, and the difference in incidence between the two groups was statistically significant. Conclusion The application of immune checkpoint inhibitors increases the risk of myocarditis, and clinicians should pay attention to the occurrence of myocarditis when applying immune checkpoint inhibitors to patients to ensure the safety of patients' medication.

Key words: immune checkpoint inhibitors, myocarditis, tumor, CHPS, active surveillance, machine learning, adverse drug reactions

中图分类号: