Application of large language models as decision support tools in occupational health and safety management: a cohort study of industrial workers

Background Occupational health and safety (OHS) risk assessment is a core preventive process aimed at identifying workplace hazards, estimating risks, and implementing control measures to reduce occupational injuries and diseases. Recent evidence indicates that AI-based systems may assist hazard identification, risk prioritization, and preventive planning, improving efficiency and standardization. Objectives This study compared AI outputs with occupational physician (OP) analyses in risk assessment, health surveillance protocol drafting, and fitness-for-work determinations. Methods This retrospective observational study was conducted in a multinational construction and facility management company with approximately 200 employees. An LLM-based system was evaluated for occupational risk assessment, health surveillance protocol development, and fitness-for-work decisions through structured comparison with an experienced OP. Three objectives were addressed: (1) analysis of the company risk assessment document (RAD); (2) comparison of surveillance protocols for specific tasks; (3) quantitative assessment of agreement in fitness-for-work judgments. The AI system (Perplexity Pro ® , “Deep Research”) was used. Agreement was measured using Cohen’s Kappa. Results AI-generated and OP-generated risk assessments were fully concordant (100%). Risk distribution across job categories was consistent, with high overall concordance (93%). Differences in surveillance protocols reflected regulatory interpretation and contextual exposure assessment rather than omission of clinically relevant elements. Conclusion LLM-based AI can reliably support standardized occupational health decisions when applied to structured data. Despite high concordance in risk assessment, protocol development, and fitness-for-work judgments, regulatory interpretation and contextual clinical evaluation remain dependent on human expertise. AI should therefore be considered a complementary decision-support tool in occupational health practice.