Scholarlink

Medication discontinuation is common among post-breast cancer patients prescribed hormone therapy. However, identifying the reasons for discontinuation through conventional methods, such as structured interviews have been limited by small sample sizes, high costs and response bias. This study evaluates the accuracy of GPT-4 in extracting and classifying treatment-related information from patient-generated posts. We applied GPT-4 to analyze 6,877 original posts about hormone therapy drugs in an online breast cancer community, without additional training or fine-tuning. In a validation test using 250 randomly sampled posts, GPT-4 achieved over 80% accuracy across four formulated questions on hormone therapy adherence, side effects, and discontinuation status. GPT-4’s analysis of all original posts revealed that 35.3% inferred discontinuation of at least one drug. Tamoxifen, anastrozole and letrozole showed the highest discontinuation prevalence, with pain and discomfort being common contributing factors. These findings highlight the effectiveness of GPT-4 on real-world medical text analysis and offered an approach to uncover trends in treatment adherence.

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

Using Large Language Model for Efficient Extraction of Treatment Discontinuation Information

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Qingyuan Song,
  • Jessie Yang,
  • Ndidiamaka Obi,
  • Congning Ni,
  • Jeremy L. Warner,
  • Qingxia Chen,
  • Lijun Song,
  • S. Trent Rosenbloom,
  • Bradley A. Malin,
  • Zhijun Yin

摘要

Medication discontinuation is common among post-breast cancer patients prescribed hormone therapy. However, identifying the reasons for discontinuation through conventional methods, such as structured interviews have been limited by small sample sizes, high costs and response bias. This study evaluates the accuracy of GPT-4 in extracting and classifying treatment-related information from patient-generated posts. We applied GPT-4 to analyze 6,877 original posts about hormone therapy drugs in an online breast cancer community, without additional training or fine-tuning. In a validation test using 250 randomly sampled posts, GPT-4 achieved over 80% accuracy across four formulated questions on hormone therapy adherence, side effects, and discontinuation status. GPT-4’s analysis of all original posts revealed that 35.3% inferred discontinuation of at least one drug. Tamoxifen, anastrozole and letrozole showed the highest discontinuation prevalence, with pain and discomfort being common contributing factors. These findings highlight the effectiveness of GPT-4 on real-world medical text analysis and offered an approach to uncover trends in treatment adherence.