TL;DR
This paper introduces SDoHO, a comprehensive ontology designed to standardize and represent Social Determinants of Health (SDoH) factors. By formally modeling SDoH concepts and their relationships, SDoHO aims to bridge the gap in current terminology and facilitate better integration of SDoH data into biomedical informatics for improved health equity.
What Problem Does This Paper Address?
Social Determinants of Health (SDoH) – the conditions in the environments where people are born, live, learn, work, play, worship, and age – have a profound impact on health outcomes, accounting for a significant portion of health disparities. However, current research and data representation of SDoH suffer from heterogeneity in definitions, categorizations, and a lack of standardization. This makes it challenging to comprehensively understand the complex associations between SDoH and health, hindering efforts to reduce healthcare inequalities and promote health equity. Existing ontologies and terminologies often cover only specific aspects of SDoH or lack the necessary detail and structure for broad application.
Methods
The authors developed SDoHO using a systematic, ontology-driven approach. Key methods include:
- Data Source Integration: Extensive literature review and utilization of various data sources, including official websites (WHO, CDC, Healthy People), standardized medical vocabularies (LOINC, SNOMED-CT, UMLS), and biomedical literature.
- Ontology Design Principles: The design was guided by three core principles:
- Interoperability: Reusing existing standards and vocabularies to ensure seamless integration with other resources.
- Applicability: Creating flexible classes and properties to support downstream applications like semantic reasoning and Natural Language Processing (NLP).
- Scalability: Designing for future growth and expansion.
- Hierarchical Structure: Combining classifications from sources like Healthy People 2030 and LOINC, and accommodating definitions of socio-behavioral determinants of health (SBDH).
- Formal Modeling: Defining classes, object properties (relationships), data properties (attributes), and annotation properties using Web Ontology Language (OWL2).
- Semantic Evaluation: Employing the Hootation tool to transform ontological hierarchies into natural language phrases for expert review, followed by iterative refinement based on evaluator feedback.
- Coverage Evaluation: Assessing the ontology's ability to represent real-world SDoH factors using clinical notes and survey data.
Datasets
The study utilized the following datasets for evaluation:
- Psychiatric Patient Clinical Notes: A subset of psychosocial assessment notes from the Harris County Psychiatric Center (HCPC), comprising over 100,000 patients. 300 social history sections were randomly selected and annotated.
- Chronic Pain Cohort Dataset: Clinical notes from local adult patients with noncancer chronic pain at the Mayo Clinic and Olmsted Medical Center (OMC). 507 clinical notes were retrieved.
- "All of Us" Research Program SDoH Survey: A national survey containing 13 domains and 81 items designed to evaluate participants' perceived feelings influenced by social surroundings.
Key Insights and Ideas
- The Power of Standardization: The core insight is that a standardized, comprehensive ontology is crucial for unlocking the full potential of SDoH data. Without it, understanding complex health determinants and their impact remains fragmented.
- Bridging the Gap: SDoHO effectively bridges the gap between unstructured clinical text/survey data and structured, computable knowledge about SDoH.
- Flexibility and Extensibility: The ontology's design emphasizes flexibility, allowing it to accommodate diverse SDoH factors and be extended for future applications.
- Inference Capabilities: By formally representing relationships, SDoHO enables inferential reasoning, allowing for the identification of implicit information (e.g., inferring MSM status from gender and sexual orientation).
- Addressing Data Heterogeneity: The ontology provides a unified framework to handle the inherent heterogeneity in SDoH data from various sources.
Key Results
- Ontology Size: The current version of SDoHO comprises 706 classes, 105 object properties, and 20 data properties, with 1,542 logical axioms.
- Hierarchical Structure: The ontology features a nine-category top-level hierarchy, with a maximum class hierarchy depth of six.
- Semantic Evaluation: Achieved a final rational agreement score of 0.967 and an inter-evaluator agreement of 0.923 after three rounds of expert review, indicating high semantic quality.
- Coverage Evaluation:
- Psychiatric Notes: 100% coverage at the domain level, 100% at the main concept level, and 100% at the value/measurement level.
- Chronic Pain Notes: 100% coverage at the domain level, 100% at the main concept level, and 66.67% at the value/measurement level.
- AoU Survey: 100% coverage at the domain level, with 44.44% coverage at the value/measurement level.
Limitations and Future Work
- Limitations:
- Relation Expression: In its current stage, relationships are based on existing sources and may not fully express proper hierarchies, leading to incomplete representation.
- Calculation Annotation: Current calculations are annotated in natural language, and query functions are not fully developed.
- Instance Population: The ontology currently lacks instances and value sets, which will be enriched in future work.
- Future Work:
- Relationship Extension: Leveraging literature to expand and refine relationship hierarchies.
- Automated Querying: Utilizing Semantic Web Rule Language (SWRL) to automate calculation query functions.
- Instance and Value Set Enrichment: Evaluating existing ontologies and guidelines to populate the ontology with instances and value sets.
- Alignment with Other Ontologies: Planning to align with Basic Formal Ontology (BFO) and other ontologies to enhance interoperability.
- NLP Integration: Combining SDoHO with NLP to volumize instances from unstructured text and transform it into structured data, aiding clinical decision support.
Summary updated