TLDR: The NIRS ontology is a new knowledge framework developed to standardize concepts and relationships in non-invasive respiratory support (NIRS) for acute care. It uses Web Ontology Language (OWL) semantics, Semantic Web Rule Language (SWRL) rules, and SPARQL queries to enable rule-based clinical reasoning, therapy recommendations, and data analysis. This addresses current inconsistencies in documentation and data representation, providing a structured foundation for improved clinical decision-making and data integration in respiratory care.
Non-Invasive Respiratory Support (NIRS) plays a crucial role in managing patients with acute respiratory failure, offering alternatives to conventional oxygen therapy or invasive mechanical ventilation. However, a significant challenge in healthcare has been the lack of standardized ways to describe and categorize NIRS modalities. This inconsistency often leads to varying documentation practices in electronic health records (EHRs), making it difficult to collect and analyze large-scale NIRS data, and ultimately limiting the ability to optimize patient care.
To address this, researchers have developed the NIRS ontology, a specialized knowledge framework designed to bring consistency and clarity to non-invasive respiratory support in acute care settings. An ontology, in this context, is a structured way of representing knowledge, explicitly defining concepts and their relationships in a machine-interpretable format. This allows for better information exchange and more informed clinical decision-making.
Developing the NIRS Ontology
The NIRS ontology was built using Web Ontology Language (OWL) semantics and a tool called Protégé. It organizes clinical concepts and their relationships, creating a unified system for understanding NIRS. To go beyond simple hierarchical structures and enable more complex, rule-based clinical reasoning, the developers incorporated Semantic Web Rule Language (SWRL) rules. They also used SPARQL queries to test the ontology’s ability to retrieve specific information and make targeted inferences, evaluating it against 17 hypothetical patient clinical scenarios and data from the Electronic Intensive Care Unit (eICU) Collaborative Research Database.
The resulting ontology is comprehensive, featuring 132 classes, 12 object properties, and 17 data properties, all interconnected by 882 axioms that define concept relationships. To ensure clinical concepts are standardized, 350 annotations were added, including definitions based on controlled vocabularies. The evaluation showed that SPARQL queries successfully validated all test cases, demonstrating the ontology’s capability to support rule-based reasoning and therapy recommendations. For example, it could infer that a patient treated with a high-flow nasal cannula (HFNC) for two hours due to acute respiratory failure might avoid endotracheal intubation.
How the Ontology is Structured
The NIRS ontology is designed with a clear hierarchy, starting from a top-level class called ‘Thing’ and branching into primary superclasses: Indication, Patient, Therapy, and Outcome. The ‘Therapy’ superclass further breaks down into device types, parameters, and timing, covering various interventions like HFNC systems and positive airway pressure variants. The ‘Patient’ superclass includes demographics, comorbidities, and other key patient characteristics, while the ‘Outcome’ superclass represents clinical endpoints such as avoiding intubation or requiring escalation of treatment.
Object properties within the ontology define relationships between different classes. For instance, ‘hasComorbidity’ links patients to their health conditions, and ‘hasDevice’ connects therapies to the equipment used. Data properties, on the other hand, capture specific numeric or time-related details, such as a patient’s age, body mass index (BMI), or ventilation settings like oxygen flow rate and FiO₂ (fraction of inspired oxygen). This detailed structure allows for precise logical reasoning.
Also Read:
- Boosting Interactive Product Configuration with Intelligent ASP Techniques
- HypKG: Integrating Patient Data with Medical Knowledge Graphs for Enhanced Healthcare Predictions
Clinical Reasoning and Interoperability
The integration of SWRL rules is key to the ontology’s ability to perform clinical inferences. These rules link clinical criteria to patient scenarios, allowing the system to answer specific competency questions. For example, one rule suggests that if a patient with COPD requires a high FiO₂ (≥ 0.60), BiPAP (bi-level positive airway pressure) is the recommended therapy. Another rule identifies patients at higher risk of intubation if their IPAP (inspiratory positive airway pressure) is greater than 22 and PEEP (positive end-expiratory pressure) is greater than 10.
To enhance its usefulness and allow for seamless integration with existing healthcare systems, the NIRS ontology is mapped to standardized biomedical terminologies. Using tools like BioPortal, relevant classes are linked to external vocabularies such as ICD-10-CM, MedDRA, and SNOMED-CT. This means that concepts like CPAP (continuous positive airway pressure) are mapped to their corresponding codes, strengthening interoperability and enabling the ontology to fit effectively into diverse EHR environments.
In conclusion, the NIRS ontology provides a unified, machine-readable framework for non-invasive respiratory support. By formally representing complex, context-dependent therapies and enabling rule-based reasoning, it offers a foundation for more consistent documentation practices, better integration into clinical data models, and advanced analysis of NIRS outcomes. For more detailed information, you can refer to the full research paper here.
