Realizing the Potential of Social Determinants Data: A Scoping Review of Approaches for Screening, Linkage, Extraction, Analysis and Interventions

Chenyu Li, Danielle L. Mowery, Xiaomeng Ma, Rui Yang, Uğurcan Vurgun, Sy Hwang, Hayoung Kim Donnelly, Harsh Bandhey, Zohaib Akhtar, Yalini Senathirajah, Eugene M. Sadhu, Emily Getzen, Philip J. Freda, Qi Long, Michael J. BecichmedRxiv 2024 · DOI

Abstract

Despite progress identifying patient social needs, further development of standards, predictive models, and coordinated interventions is critical to fulfill the potential of SDoH-EHR integration. Additional database searches could strengthen this scoping review. Ultimately widespread capture, analysis, and translation of multidimensional SDoH data into clinical care is essential for promoting health equity.

Introduction

It is a well-known "open secret" in medicine that clinical care only accounts for approximately 20% of a person’s health outcomes. The remaining 80% is driven by Social Determinants of Health (SDoH)—the "causes of the causes" such as housing stability, food security, and transportation access. Despite their massive impact, SDoH data is rarely captured in a structured way within Electronic Health Records (EHRs). Instead, this critical information often ends up "trapped" in unstructured clinician notes or is simply never recorded.

This paper presents a comprehensive scoping review of how SDoH are integrated into EHRs. By analyzing 324 studies, the authors map the transition from manual screening to automated Natural Language Processing (NLP)-based extraction, revealing a significant gap between our ability to identify social risks and our capacity to translate that data into standardized, large-scale clinical interventions.

Conceptual Framework and Data Sources

The integration of SDoH into clinical research requires a multi-faceted data ecosystem. Researchers utilize a variety of sources, including community/public health data, clinical data from EHRs, and patient-reported data. These sources feed into a research repository that informs care and generates evidence through a translational research cycle.

Conceptual framework for SDoH data integration, illustrating the flow from data sources to research repositories and the translational research cycle.

Methodology and Domain Categorization

Following PRISMA-ScR guidelines, the authors systematically mapped the literature by searching PubMed for studies at the intersection of "EHR" and "SDoH." From an initial pool of 685 articles, 324 were selected for qualitative synthesis. The research was categorized into five primary SDoH concepts, each encompassing specific domains of social risk.

Table mapping SDoH concepts to specific domains, including economic stability, neighborhood environment, and social context.

The review focused on five non-mutually exclusive areas:

  1. Screening Tools: The use of standardized (e.g., PRAPARE) versus customized questionnaires.
  2. Data Collection: Methods for ingesting external data like US Census or claims data.
  3. NLP for SDoH: Utilizing AI to extract insights from free-text clinical notes.
  4. Health Outcomes: Analyzing correlations between SDoH and diseases like diabetes or COVID-19.
  5. Interventions: Evaluating programs that actually address identified social needs.

Key Insights: The NLP Evolution and the Intervention Gap

A major finding of the review is the rapid evolution of NLP. The field is shifting from simple keyword searches to sophisticated transformer-based models like BERT and RoBERTa. These models achieve precision and recall scores over 80% for extracting nuanced social risks—such as housing instability—that structured checkboxes often miss.

However, the review also highlights a "Home-Grown Trap." Many health systems develop customized screening tools rather than using validated standards. While this facilitates local workflows, it creates significant interoperability challenges, making it difficult to compare social risk data across different regions.

Furthermore, there is a stark imbalance in the literature. While hundreds of papers focus on identifying SDoH or linking them to adverse outcomes (such as increased ED visits), only 19 papers focused on actual interventions. We are currently much more proficient at documenting social risk than we are at modifying clinical workflows to address it.

Future Directions

The authors identify several frontiers for future research:

  • Generative AI: Large Language Models (LLMs) represent the next step for context-aware SDoH recognition.
  • Full-Stack Integration: Moving toward systems where SDoH data automatically triggers clinical decision support, such as automated referrals to food banks.
  • Standardization: Developing domain-specific ontologies is essential to ensure that social risk terminology is consistent across different clinical settings.

Ultimately, the goal is to move beyond "one-off" analyses toward a widespread, standardized capture and translation of multidimensional SDoH data into clinical care to promote health equity.

Summary updated

Abstract

Despite progress identifying patient social needs, further development of standards, predictive models, and coordinated interventions is critical to fulfill the potential of SDoH-EHR integration. Additional database searches could strengthen this scoping review. Ultimately widespread capture, analysis, and translation of multidimensional SDoH data into clinical care is essential for promoting health equity.
BibTeX
@misc{li2024realizing,
  author = {Chenyu Li and Danielle L. Mowery and Xiaomeng Ma and Rui Yang and Uğurcan Vurgun and Sy Hwang and Hayoung Kim Donnelly and Harsh Bandhey and Zohaib Akhtar and Yalini Senathirajah and Eugene M. Sadhu and Emily Getzen and Philip J. Freda and Qi Long and Michael J. Becich},
  title = {{Realizing the Potential of Social Determinants Data: A Scoping Review of Approaches for Screening, Linkage, Extraction, Analysis and Interventions}},
  year = {2024},
  doi = {10.1101/2024.02.04.24302242},
  howpublished = {medRxiv},
}
Figure