Development of an Ontology of Social Determinants of Health for Pediatrics

Background and Purposes: Social determinants of health (SDoH) are psychosocial and non-medical factors that impact human health. While consideration of SDoH are included in risk assessment and clinical care, data on SDoH are not easily accessible or extracted from patient electronic health records (EHR). SDoH data is often found in unstructured text data such as social work notes and assessments, psychological evaluations, and nurses text documentation. In an effort to extract and reliably code SDoH information from unstructured text data in pediatric patients’ EHR, we proposed the development and evaluation of Social Determinants of Health for Pediatrics (SDoH-Pediatrics) ontologies to systematically interpret and code SDoH factors from unstructured text data. Ontologies are formal representations of concepts within a psychosocial domain and the relationships between these concepts. Ontological structures standardize the terminology to produce structured data for subsequent analysis.

Methods: The development of SDoH-Pediatrics was a multifaceted process that drew upon various valuable resources. We integrated information from multiple sources, including Healthy People 2030 and five pre-transplant assessment tools to ensure its comprehensive coverage and relevance to pediatric patients. By leveraging these assessment tools, we ensured that SDoH-Pediatrics was informed and relevant to practical settings. Related SDoH ontologies were used, such as SOHO, to further enrich our ontology. Relevant codes from ICD-10-CM (i.e., Z55.0 to Z65.0) were included as they effectively capture various social determinants of health. Lastly, recognizing the critical family-centered nature of pediatric care, we integrated concepts and categories that account for the sociodemographic characteristics of caregivers under the "caregiver information" category and "family context" categories. To ensure that SDoH-Pediatrics is consistent, adaptable, and semantically sound, it was evaluated using the HermiT reasoner (a commonly used ontology evaluation tool) and through inter-rater reliability tests from human expert evaluation.

Results: The current version of SDoH-Pediatrics was developed via Protégé in OWL format. It has 298 classes with 654 axioms and was made available on the American National Center for Biomedical Ontology (NCBO) BioPortal. The ontology's foundation is principally constructed upon two first-level classes: Patient History and Social Determinants of Health, constituting 80.9% of its framework. Results from the HermiT reasoner showed that SDoH-Pediatrics is a consistent and coherent ontology and, therefore, did not need further debugging. A Cohen's kappa of 0.5321 was obtained from the results of the human expert evaluation, indicating 76.0% agreement and moderate agreement about the SDoH-Pediatrics between the two evaluators.

Conclusion and Implications:

The development of SDoH-Pediatrics will enable data exchange across healthcare information systems and aid pediatric healthcare researchers in more easily integrating SDoH factors into analysis from unstructured text. SDoH-Pediatrics ontologies are able to provide a foundation for clinical decision support systems with accurately coded psychosocial data, support advanced data mining for identifying SDoH factors from patients' medical records, and aid in development of personalized treatment plans focusing on psychosocial factors. Perhaps most importantly, ontology development like the SDoH-Pediatrics will support inclusion of SDoH data within in healthcare research and clinical care.