Objective:

To develop and evaluate methods for identifying social determinants of health (SDoH) from clinical text using rule-based and large language model (LLM) techniques.

Key Findings:

• LLM-based methods showed potential for effective SDoH extraction with minimal annotated data.
• Rule-based approaches had lower sensitivity due to reliance on fixed rules.
• Advanced reasoning models demonstrated improved performance in identifying less-explored SDoH factors.

Interpretation:

The study highlights the importance of leveraging advanced NLP techniques to extract critical SDoH information from clinical notes, which can enhance population health research and clinical decision-making.

Limitations:

• The study focused on a limited set of SDoH domains.
• Performance evaluation was conducted on specific models without extensive fine-tuning.

Conclusion:

This research provides a cost-efficient framework for future clinical NLP applications, addressing key gaps in SDoH research and emphasizing the need for comprehensive data extraction methods.