AI-Induced Occupational Health Assessment

Area of Science:

• Occupational Health
• Epidemiology
• Computational Linguistics

Background:

• Work is vital for well-being but poses health risks.
• Job-Exposure Matrices (JEMs) link job histories to health risk assessments.
• Standardizing job descriptions for JEMs is manual, costly, and time-consuming.

Purpose of the Study:

• To provide an overview of occupational health assessment and Automatic Occupational Coding (AOC).
• To explore the potential of Large Language Models (LLMs) in AOC.
• To address limitations in current AOC methods, including generalizability.

Main Methods:

• Review of JEMs and occupational classification systems.
• Analysis of existing (semi-)automatic coding and Decision Support Systems (DSS).
• Exploration of Large Language Models (LLMs) for AOC applications.

Main Results:

• Current fully automatic coding systems struggle with out-of-distribution data, limiting real-world use.
• Decision Support Systems (DSS) improve reliability through expert correction.
• LLMs offer potential for enhanced accuracy and generalizability in AOC.

Conclusions:

• LLM application in AOC is currently unexplored.
• LLMs could significantly improve the accuracy and generalizability of AOC.
• Future research should focus on LLM integration into AOC systems.