High Throughput Phenotyping of Physician Notes with Large Language and Hybrid NLP Models

Area of Science:

• Medical Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Deep phenotyping, the detailed description of patient signs and symptoms using ontologies, traditionally requires manual review of clinical notes.
• High-throughput methods are essential for analyzing the vast amount of data in electronic health records (EHRs).
• Significant progress has been made over the past 30 years to enable high-throughput phenotyping.

Purpose of the Study:

• To evaluate the accuracy and feasibility of using a large language model (LLM) and a hybrid Natural Language Processing (NLP) model for high-throughput deep phenotyping of physician notes.
• To compare the performance of LLMs and hybrid NLP models against traditional methods for extracting clinical information.

Main Methods:

• Development and application of a large language model for analyzing physician notes.
• Implementation of a hybrid NLP model, combining word vectors and a machine learning classifier.
• Assessment of model performance on accuracy and throughput for deep phenotyping tasks.

Main Results:

• Both the large language model and the hybrid NLP model demonstrated high accuracy in performing high-throughput phenotyping on physician notes.
• LLMs showed particular promise for efficient and accurate deep phenotyping of clinical documentation.
• The study confirms the feasibility of automated phenotyping for large volumes of EHR data.

Conclusions:

• Large language models are poised to become the preferred method for high-throughput deep phenotyping of physician notes.
• These advanced NLP techniques will significantly enhance the ability to extract and analyze patient signs and symptoms from EHRs.
• The findings have significant clinical relevance for improving patient care through efficient data analysis.