Page 24 - AIH-1-3

Artificial Intelligence in Health Optimizing EHRs to support AI

data structures, types, and value sets and a universal EHR et al. explored if these three standards could work together.
71
architecture designed to process the high-level components They found all three to be useful for the purposes for which
of an EHR categorized as folder, composition, section, they were designed but that they have limitations when
and entry. The entry category is further categorized into used for different purposes. Selecting the most suitable set
observations, evaluations, instructions, and actions. Its of standards able to best meet a defined purpose is critical.
specifications include a platform model. Its open clinical Information models in FHIR are called resources. All
data repository provides access to nearly 1000 archetypes clinical and other findings can only be represented using a
to date. Collectively, these consist of the largest number single observation resource, which then requires adapting
of data points representing various levels of granularity and bringing together many of them together as a profile to
in the world. These archetype models are evidence- be able to represent even a simple concept like blood pressure.
30
based, developed, and reviewed by well over 1000 However, this HL7 FHIR design has been a deliberate
multidisciplinary experts from 114 countries. The number compromise for the simplicity of technical implementation
of available archetypes is growing exponentially relative to by developers at the expense of its expressivity. As a result,
openEHR-compliant implementations. The development FHIR is being widely adopted by many vendors and health
of universal openEHR archetypes is undertaken by the systems. There is now a common trend to use both FHIR and
potential data users who have a sound understanding of openEHR together, representing demographic, administrative
context that must be represented. and simple clinical data exchange using FHIR and rich clinical
A further modeling layer is the openEHR templates data using openEHR. Archetypes enable access to far more
which gather one or more archetypes and define use-case- detailed clinical data points as they represent the maximal
specific constraints (e.g., discharge summary, medication data elements for a given concept. The FHIR
order, and clinical reports). Templates are used to drive Resources, on the other hand, include minimal data
information systems. Archetypes and templates, along with elements that have been adopted by current health
annotated clinical terminology and ontology concepts, information systems. Therefore, FHIR allows for rapid data
define domain-specific information models and enable exchange between legacy systems.
semantic interoperability in healthcare through this multi-
level modeling approach. Their use for the management of 4.5. What’s possible with next-generation EHR
clinical data especially optimizes the use of EHR data for systems and data?
AI purposes. Current trends on building digital twins exclusively using
88
The Archetype Query Language allows formulation EHR data and AI without using atomic-level omics data
of portable queries using domain concepts unlike field and mechanistic knowledge and constraints of human
or table names in a traditional relational database. physiology and anatomy will fall short of driving next-
83
Several examples have been reported elsewhere. At the generation clinical decision support tools and research.
84
knowledge level, openEHR also defines a formal clinical Such limitations are due to not being able to train AI
guideline specification (GDL) to drive decision support – with reliable data. At present, available data are inevitably
all in a single standards stack. There is also ongoing work flawed by incorrect facts and associations as well as bias
85
to model and capture health-care processes and clinical due to known shortcomings of most EHR systems in use
workflows. and available data sets.
The openEHR Clinical Knowledge Manager (CKM) is Atalag has defined an ontology mapping-based
an online clinical models repository (archetypes, templates, framework leveraging a multitude of existing and mature
and clinical terminology subsets) and an advanced web- standards to bring together all these data sources (EHR,
based distributed knowledge curation tool. CKM supports physiome, and omics) in a way that preserves the clinical,
86
an editorial process resembling peer-review process of a biological, physiological, and anatomical context and
scientific journal where editors with the help of domain semantics that can drive these next-generation methods
89
experts can conduct online reviews using the CKM’s web and tools, as shown in Figure 2. This model represents
interface and then publish models. Figure 1 shows the an ontology mapping-based framework to show how
relationships between ontologies, models, and systems. compliance with various types of data exchange standards
enable an openEHR-compliant clinical data repository to be
4.4. HL7 FHIR populated and enable precision medicine supported by AI.
The HL7 FHIR standard represents an evolutionary result Current proprietary EMR/EHR systems and
27
of ongoing developments of the HL7 messaging standards. infrastructures are no longer considered “fit for purpose”
Its implementation is gaining momentum. Pedrera-Jiménez due to their many shortcomings. When machine learning

Volume 1 Issue 3 (2024) 18 doi: 10.36922/aih.3056