Scholarlink

Smoking status and pack-per-year are essential factors for cardiovascular and cancer research, yet in Denmark, structured documentation is typically limited to severe or chronic cases, leaving valuable data hidden in unstructured electronic health records. Manual extraction of this information is impractical for large-scale studies, necessitating automated solutions. This study presents an explainable AI model tailored to the linguistic complexities of Danish medical text, achieving accuracies of 93.51% for detecting smoking status, 86.77% for predicting pack-per-year, and 82.75% in a combined multi-task scenario. The model still faces challenges with specific categories such as E-Cigarette use, passive smoking, and differentiating between low and medium pack-per-year classes. Future efforts will refine these aspects and explore advanced NLP methods such as transformers to enhance clinical applicability.

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

Explainable AI for Smoking Behaviour Detection: A Danish Medical Records Study

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amir Sorayaie Azar,
  • Margrethe Bang Høstgaard Henriksen,
  • Ole Hilberg,
  • Amin Naemi,
  • Jamshid Bagherzadeh Mohasefi,
  • Uffe Kock Will,
  • Ali Ebrahimi

摘要

Smoking status and pack-per-year are essential factors for cardiovascular and cancer research, yet in Denmark, structured documentation is typically limited to severe or chronic cases, leaving valuable data hidden in unstructured electronic health records. Manual extraction of this information is impractical for large-scale studies, necessitating automated solutions. This study presents an explainable AI model tailored to the linguistic complexities of Danish medical text, achieving accuracies of 93.51% for detecting smoking status, 86.77% for predicting pack-per-year, and 82.75% in a combined multi-task scenario. The model still faces challenges with specific categories such as E-Cigarette use, passive smoking, and differentiating between low and medium pack-per-year classes. Future efforts will refine these aspects and explore advanced NLP methods such as transformers to enhance clinical applicability.