Scholarlink

<p>Despite advanced diagnostic tools, early detection of rare genetic conditions like Noonan syndrome (NS) remains challenging. We evaluated a deep learning model’s real-world performance in identifying potential NS cases using electronic health record (EHR) data, validated through genetic sequencing and clinical assessment. The model analyzed 92,428 patients, identifying 171 high-risk individuals (score &gt; 0.8) who underwent comprehensive review. Among these, 86 had prior genetic diagnoses, including three NS cases diagnosed during the study period. Genetic sequencing of remaining patients identified two additional NS cases with pathogenic variants. The model achieved 2.92% precision and 99.82% specificity. While precision was lower than prior validation (33.3%), this reflected expected differences in disease prevalence rather than model degradation. NS-associated phenotypes were enriched among high-risk patients, and trajectory analysis showed potential for earlier identification, highlighting both promise and limitations of EHR-based computational screening tools.</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

Sequencing validates deep learning models for EHR-based detection of Noonan syndrome in pediatric patients

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月21日

查看全文
  • Zeyu Yang,
  • Amy Shikany,
  • Ammar Husami,
  • Xinjian Wang,
  • Eneida Mendonca,
  • K. Nicole Weaver,
  • Jing Chen

摘要

Despite advanced diagnostic tools, early detection of rare genetic conditions like Noonan syndrome (NS) remains challenging. We evaluated a deep learning model’s real-world performance in identifying potential NS cases using electronic health record (EHR) data, validated through genetic sequencing and clinical assessment. The model analyzed 92,428 patients, identifying 171 high-risk individuals (score > 0.8) who underwent comprehensive review. Among these, 86 had prior genetic diagnoses, including three NS cases diagnosed during the study period. Genetic sequencing of remaining patients identified two additional NS cases with pathogenic variants. The model achieved 2.92% precision and 99.82% specificity. While precision was lower than prior validation (33.3%), this reflected expected differences in disease prevalence rather than model degradation. NS-associated phenotypes were enriched among high-risk patients, and trajectory analysis showed potential for earlier identification, highlighting both promise and limitations of EHR-based computational screening tools.