Linking Computational Physiology Models with Clinical Data
Linking Computational Physiology Models with Clinical Data can be instrumental for real-world model validation as well as enable personalised and predictive clinical decision support systems. Electronic health records (EHR) are sinks of biomedical knowledge and include manifestations of genomic and environmental aspects that impact on biological systems.
In the computational physiology domain recent attempts to enable semantic interoperability heavily rely on Semantic Web technologies and utilise ontology-based annotations but a wealth of useful information and knowledge sits in EHRs where Semantic Web technologies have very limited use.
openEHR provides open standards for structuring and exchange of healthcare data. Core openEHR specifications have also been adopted by ISO and CEN making it an international standard which underpins many national programs and vendor implementations worldwide. At the heart of the openEHR formalism are Archetypes which are constraint based models of health information based on stable technical building blocks (Reference Model). While the Reference Model provides syntactic interoperability, reusable Archetypes define granular clinical concepts like blood pressure or lab results and enable semantic interoperability in healthcare.
openEHR also provides the means to annotate whole or parts of the information model by a mechanism called terminology binding. This can then be used to semantically link to annotated CellML models through ontology mapping. This linkage is bidirectional; e.g. from computational models to clinical data and vice versa.
Step by Step Methodology: 1) Select a source for which to find linked clinical data (computational model, disease state, etc.); 2) Retrieve all semantic annotations from PMR for the selected model; 3) Using mappings between ontology and clinical terminology concepts display matching clinical terms; 4) Find Archetypes with matching terminology bindings leveraging all types of ontology relations (e.g. is-a, part-of, subsumption) and display as a list/graph; 5) Discover clinical data with matching terminology bindings (both schema and instance level); 6) Display and export corresponding clinical data; 7) Feed clinical data to model parameters (as per mappings), and perform simulations.
We illustrate here how openEHR can link the world of health ICT and computational physiology and bioinformatics communities. More work is needed around ontology mapping, annotation methods and tooling.