Scholarlink

<p>Children with congenital heart disease (CHD) are at risk of adverse drug events (ADEs) due to developmental pharmacokinetics and polypharmacy. Current pharmacovigilance methods lack specificity and often overlook pediatric populations. This study aimed to develop a machine learning (ML) framework that integrates clinical and pharmacological features to identify ADEs, specifically 29 pre-specified outcomes in children with CHD. Initially, a retrospective dataset of children with CHD cases from the FDA Adverse Event Reporting System (FAERS) was constructed. The dataset included 15 clinically validated high-risk drugs. Each patient/drug record combined clinical features with structured drug descriptors. A total of 18 feature integration strategies were tested across five classifiers. Model performance was evaluated using stratified cross-validation and Area under the receiver operating characteristic curve (AUROC). Random Forest models achieved the highest AUROC scores exceeding 0.95 in identifying key ADEs. The best-performing configuration (Clinical + Target + Enzyme + Pathway) reached an AUROC of 0.966. Drug-specific predictions aligned with known safety profiles, including Amiodarone/bradycardia, Ibuprofen/necrotizing enterocolitis, and Furosemide/hypokalemia.Integrating pharmacological descriptors with clinical data significantly improves identifying ADEs in children with CHD. This ML framework shows strong potential for implementation in clinical decision support systems to guide personalized prescribing in children with CHD.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Identifying adverse drug events in children with congenital heart disease using machine learning techniques and multi-domain descriptors

  • 期刊论文

  • 发表日期：

    2026年7月18日

查看全文
  • Esmaeel Toni,
  • Haleh Ayatollahi,
  • Reza Abbaszadeh,
  • Alireza Fotuhi Siahpirani

摘要

Children with congenital heart disease (CHD) are at risk of adverse drug events (ADEs) due to developmental pharmacokinetics and polypharmacy. Current pharmacovigilance methods lack specificity and often overlook pediatric populations. This study aimed to develop a machine learning (ML) framework that integrates clinical and pharmacological features to identify ADEs, specifically 29 pre-specified outcomes in children with CHD. Initially, a retrospective dataset of children with CHD cases from the FDA Adverse Event Reporting System (FAERS) was constructed. The dataset included 15 clinically validated high-risk drugs. Each patient/drug record combined clinical features with structured drug descriptors. A total of 18 feature integration strategies were tested across five classifiers. Model performance was evaluated using stratified cross-validation and Area under the receiver operating characteristic curve (AUROC). Random Forest models achieved the highest AUROC scores exceeding 0.95 in identifying key ADEs. The best-performing configuration (Clinical + Target + Enzyme + Pathway) reached an AUROC of 0.966. Drug-specific predictions aligned with known safety profiles, including Amiodarone/bradycardia, Ibuprofen/necrotizing enterocolitis, and Furosemide/hypokalemia.Integrating pharmacological descriptors with clinical data significantly improves identifying ADEs in children with CHD. This ML framework shows strong potential for implementation in clinical decision support systems to guide personalized prescribing in children with CHD.