Chinese Journal of Pharmacovigilance ›› 2025, Vol. 22 ›› Issue (1): 16-22.
DOI: 10.19803/j.1672-8629.20240807

Previous Articles     Next Articles

Pediatric Drug-Induced Immune Thrombocytopenia Based on Real World Data and Quantitative Structure-Activity Relationship

NIE Xiaolu1,2,3, ZHAO Houyu4, HUO Donghui5,6, YAN Aixia6, SUN Feng2,3, PENG Xiaoxia1,3, NI Xin7#, ZHAN Siyan2,*   

  1. 1Center for Clinical Epidemiology, Evidence-Based Medicine, Beijing Children's Hospital, Capital Medical University, National Center for Children's Health, Beijing 100045, China;
    2Department of Epidemiology and Biostatistics, School of Public Health, Peking University, Beijing 100091, China;
    3Hainan Institute of Real World Data, Qionghai Hainan 571437, China;
    4School of Medicine, Chongqing University, Chongqing 400030, China;
    5SINOPEC (Dalian) Research Institute of Petroleum and Petrochemicals Co., Ltd., Dalian Liaoning 116045, China;
    6College of Life Science and Technology, Beijing University of Chemical Technology, Beijing 100029, China;
    7Office of Beijing Children's Hospital, Capital Medical University, National Center for Children's Health, Beijing 100045, China
  • Received:2024-10-21 Online:2025-01-15 Published:2025-01-22

PDF (PC)

46

Abstract: Objective To conduct risk signal mining and toxicity evaluation related to drug-induced immune thrombocytopenia (DIIT) . Methods A real-world study was conducted using the electronic medical record data from Beijing Children's Hospital spanning from 2009 to 2020. Potential DIIT signals in children were mined using a modified laboratory extreme abnormal proportion imbalance method. Based on multi-source information, a DIIT benchmark database was constructed. Three machine learning algorithms-support vector machine (SVM), random forest (RF), and XGBoost-and drug molecular descriptors (ECFP4 and CORINA) were employed to establish quantitative structure-activity relationship (QSAR) models. Toxicity evaluation research on DIIT was conducted based on the physicochemical properties and molecular fingerprints of drugs. Results Eighteen positive signal drugs for pediatric drug-induced thrombocytopenia (DITP) were identified from the electronic medical record data. Among them, nystatin and latamoxef sodium emerged as two new positive DITP signals in both children and adults (DITP risks of OR: 1.75, 95%CI: 1.37 to 2.22 and OR: 1.61, 95%CI: 1.38 to 1.88, respectively). Additionally, six new positive signal drugs not previously reported in children were discovered: imipenem, teicoplanin, fusidic acid, cefotaxime sodium, ceftazidime, and cefepime. Among the nine different QSAR models constructed based on 1 319 compounds, the top three models with optimal performance were the SVM-ECFP4+CORINA model, the RF-ECFP4+CORINA model, and the XGBoost-ECFP4 model, with the area under the curve (AUC) values on the external validation set of 0.747, 0.732, and 0.712, respectively. A consensus model composed of these three optimal models accurately predicted the seven potential positive signals identified in the first step. Conclusion Combining real-world signal mining methods with QSAR models can enhance the methodological framework for post-marketing drug safety evaluation in China.

Key words: Drug-Induced Immune Thrombocytopenia (DIIT), Pediatric, Adverse Drug Reactions, Safety Evaluation, Signal Detection, Quantitative Structure-Activity Relationship Study, Real-World, Nystatin, Latamoxef Sodium

CLC Number: