Extracting Social Determinants of Health from Dental Clinical Notes

Area of Science:

• Dentistry and Public Health
• Natural Language Processing (NLP) in Healthcare

Background:

• Social determinants of health (SDoH) are crucial for understanding patient well-being.
• Electronic Dental Records (EDRs) may contain valuable SDoH information within clinical notes.
• Current methods for extracting SDoH from dental notes are limited.

Purpose of the Study:

• To assess the availability of SDoH data in dental clinical notes.
• To evaluate and compare NLP methods for extracting SDoH from these notes.

Main Methods:

• A dataset of 1,000 dental clinical notes was manually annotated for SDoH categories.
• Fine-tuned pre-trained language models (PLMs) and a rule-based approach were used for SDoH extraction.
• Performance was measured using F1-scores.

Main Results:

• PLMs achieved higher F1-scores for sugar (0.75), alcohol (0.67), and housing (0.42) compared to the rule-based method.
• Both methods struggled with employment and methamphetamine extraction, yielding F1-scores of 0.
• SDoH were infrequently recorded, with sugar (9.1%) and tobacco (3.9%) being most common.

Conclusions:

• Extracting SDoH from dental notes is challenging due to low recording frequency, brevity, and inconsistency.
• NLP methods, particularly PLMs, show promise but require further refinement.
• Standardizing SDoH reporting in dental clinical notes is essential for improved surveillance.