Applications of Natural Language Processing and Large Language Models for Social Determinants of Health: Systematic...

Area of Science:

• Computational linguistics
• Health informatics
• Public health research

Background:

• Social determinants of health (SDOH) significantly impact health outcomes but are often found in unstructured text.
• Natural language processing (NLP) offers methods to extract SDOH information from sources like electronic health records and social media.
• Existing research on NLP for SDOH is fragmented, making synthesis and comparison difficult.

Purpose of the Study:

• To systematically review the application of NLP, including large language models (LLMs), in social determinants of health research.
• To identify current trends, methodological variations, and research gaps in this field.
• To highlight future research directions for advancing SDOH understanding and application.

Main Methods:

• Conducted a systematic review following PRISMA guidelines, searching 7 databases for publications from 2014 to November 2025.
• Included studies applying NLP to identify, classify, extract, or predict SDOH from text.
• Data extraction and risk-of-bias assessment performed by multiple reviewers, with influential studies identified via citation counts.

Main Results:

• 142 studies met inclusion criteria, with significant growth from 2023-2025.
• Most studies used electronic health records and private datasets; housing instability, employment, and financial conditions were common SDOH domains.
• NLP models showed strong performance (median F1-scores 0.75-0.85), but code sharing and reproducibility details were limited. LLMs appeared in 19.7% of studies.

Conclusions:

• The review synthesizes NLP and LLM applications in SDOH research, addressing a critical knowledge gap.
• Identified limitations include the lack of a unified SDOH framework, scarce public benchmarks, and insufficient real-world deployment evaluation.
• Future work should focus on transparent dataset development and implementation-focused evaluation to ensure equitable health outcomes.