Advances in Personalized Medicine: Analyzing Unstructured Electronic Health Records with Large Language Models

Strategic Insights -

Investigators from the Department of Artificial Intelligence in Biomedical Engineering have made significant strides in personalized medicine by employing large language models (LLMs) to analyze unstructured electronic health records (EHRs) in patients with dysphagia. The study, published in the IEEE Journal of Translational Engineering in Health and Medicine, demonstrates the effectiveness of LLMs in identifying distinct groups of patients sharing similar pathophysiological swallowing dysfunctions. This breakthrough has the potential to enhance understanding and treatment of dysphagia, a complex and common disorder that complicates diagnoses and treatment.

Key Takeaways:

  • Researchers from the Department of Artificial Intelligence in Biomedical Engineering have developed a method to analyze unstructured clinical narratives and extract diagnostic information from EHRs using NLP techniques and LLMs.
  • The study included a dataset of 486 patients with diverse dysphagic conditions and utilized clustering algorithms to identify distinct groups of patients sharing similar pathophysiological swallowing dysfunctions.
  • Basic NLP techniques provided limited insights due to the high variability of the data, but LLMs helped bridge the gap in understanding nuanced medical information about dysphagia and related conditions.
  • The study demonstrated that levering closed-source models can effectively cluster different categories of dysphagia and provide evidence that LLMs are highly promising in future dysphagia research.
  • The clinical impact of this research is to provide evidence that analyzing large volumes of EHRs can help clarify the causes of dysphagia and identify contributing factors, ultimately improving individualized care.
  • The study aligns with clinical research, enhancing diagnostic speed and enabling personalized treatment.

Statistics:

  • 486 patients were included in the dataset, representing a group with diverse dysphagic conditions.
  • The study utilized clustering algorithms on the extracted diagnostic features to identify distinct groups of patients sharing similar pathophysiological swallowing dysfunctions.
  • The publisher for IEEE Journal of Translational Engineering in Health and Medicine is IEEE.

Sources:

  • Unstructured Electronic Health Records of Dysphagic Patients Analyzed by Large Language Models. IEEE Journal of Translational Engineering in Health and Medicine, 2025, 13(): 237-245. (IEEE Journal of Translational Engineering in Health and Medicine - http://ieeexplore.ieee.org/xpl/aboutJournal.jsp?punumber=6221039)
  • NewsRx. Department of Artificial Intelligence in Biomedical Engineering Researchers Target Personalized Medicine (Unstructured Electronic Health Records of Dysphagic Patients Analyzed by Large Language Models). Health & Medicine Week. June 20, 2025; p 1098.