Deep learning for occupation recognition and knowledge discovery in rheumatology clinical notes.

Area of Science:

• Rheumatology
• Digital Health
• Occupational Medicine

Background:

• Occupational data is a key social determinant of health, impacting healthcare.
• Electronic health records (EHR) often lack structured occupational information.
• Understanding occupation's role in rheumatology is crucial for diagnosis and treatment.

Purpose of the Study:

• To assess the collection and use of occupation data in rheumatology clinical narratives.
• To identify factors influencing occupation data collection in EHRs.
• To analyze the association between patient occupations and diagnoses.

Main Methods:

• A pre-trained Spanish language model was fine-tuned on biomedical texts.
• The model identified occupation mentions in 35,586 rheumatic patient EHRs.
• Occupation data was normalized using the European Skills, Competences, Qualifications, and Occupations (ESCO) classification.

Main Results:

• The automated model achieved an F1-score of 0.73, identifying occupations in 10% of patients.
• Mechanical pathologies were linked to higher occupation data collection rates.
• Specific occupations were associated with distinct rheumatic conditions, including autoimmune diseases.

Conclusions:

• Automated occupation recognition in EHRs is feasible and valuable for rheumatology.
• Occupational data is a significant social determinant of health in rheumatic diseases.
• Integrating occupation data can enhance targeted prevention and treatment strategies.