Leveraging Large Language Models for Early Detection of Anomaly Work Injury Cases: Data-Driven Approach to...

Area of Science:

• Artificial Intelligence in Healthcare
• Clinical Data Analysis
• Rehabilitation Medicine

Background:

• Large language models (LLMs) show promise for analyzing unstructured clinical data.
• The use of LLMs in rehabilitation therapy for work injuries is not well-established.
• Automating analysis can improve efficiency in managing work injury cases.

Purpose of the Study:

• To assess an LLM-assisted method for identifying anomalous work injury rehabilitation cases.
• To improve the scalability and precision of case management in rehabilitation.
• To evaluate LLM performance in detecting deviations from expected recovery durations.

Main Methods:

• Retrospective analysis of 110,346 deidentified work injury cases (2001-2024).
• Utilized LLMs to predict recovery duration from free-text injury descriptions.
• Classified cases as anomalous if sick leave days exceeded LLM-predicted maximums.

Main Results:

• GPT-4o achieved over 73% accuracy in nonanomalous classification and 79% in overall detection.
• The LLM-assisted method demonstrated high accuracy across various subgroups.
• Reliable extraction of information from unstructured clinical notes was confirmed.

Conclusions:

• LLMs can effectively automate anomaly detection in large-scale rehabilitation datasets.
• The approach shows potential for tailoring age-specific rehabilitation strategies.
• Future research should address generalizability across different healthcare systems and regions.