Page 18 - AIH-1-3

Artificial Intelligence in Health Optimizing EHRs to support AI

based on proprietary system architectures requiring the use cases. For example, Zhang et al. found that multi-stage
16
adoption of transformation strategies. The bigger the health data flow chains in the UK do not fulfill recommended
1
ecosystem, the greater the difficulty to change foundational best practices for safe data access and that its existing
architectural system design. Ingram has documented these infrastructure produces aggregation of duplicate data
historical developments and evolutionary discoveries in assets. Multi-stage data flow chains limit the diversity of
great detail. Ingram also explains the scientific foundations data required to add value to end users.
2
of new discoveries over time. “There are gaping holes in data platform infrastructure
The literature included in this review has collectively that supports deployment of data-driven tools (such as
exposed that most current health ecosystem governance digitally/AI-enabled trials, or AI deployment).” 16
infrastructures, including legislation, regulations, and Information, Communication, and Technology (ICT)
policies, essentially determine not only any nation’s health and Information Systems (IS) research tend to be undertaken
ecosystem infrastructure but also the governance of its data, as action research (through trial and error, providing local
information, knowledge, and wisdom assets. Such high- solutions to problems identified). The action research approach
level infrastructures determine their strategic directions enables constant evaluation of implementation providing
including the mandating of standards compliance. Poorly a process of checking for and affirming understanding that
3
informed policy decisions have resulted in numerous is specific, non-evaluative, manageable, and focused on the
costly failures and continue to impede efficient progress. 4-8 target of interest (assessments feedback loops). Standards
A lack of foundational technical knowledge in emerging development activities consist of collaborative problem
developments and the continuing use of legacy systems has solving, making use of international experts, and user
resulted in a large digital health ecosystem-wide architectural feedback regarding issues encountered when testing new
patchwork. Natural language processing is compromised standards.
9
due to the prevalence of duplicate information in EMR The ISO TC 215 is responsible for developing
systems; secondary data use or advanced data analytics is standards specifically to suit the health industry along
10
compromised by incomplete data from EHRs. 11-13 We are with a few other Standards Development Organizations
also witnessing a continuing proliferation of applications (SDOs), including Health Level Seven (HL7), SNOMED
17
(Apps). Most apps are standalone and unable to share data International ), Digital Imaging and Communications
18
with EHR/EMR systems making most EHRs incomplete in Medicine (DICOM), and Clinical Data Interchange
and unable to provide a comprehensive overview of a Standards Consortium (CDISC). 19-22 A number of these
person’s health status at any point in time and across SDOs are working collaboratively through the Joint
different tiers of the health-care system. Incomplete EHRs Initiative Council (JIC) established in 2007. However,
23
represent a significant patient safety issue. it needs to be remembered that few governments have
Software developers tend to focus on meeting mandated compliance with any specific set of standards
procurement requirements, with a focus on how they can although this is changing.
best meet market needs and be competitive. Consequently,
we continue to have chaos, fragmentation, and data silos 2.1. Interoperability
as very few of these decisions are being coordinated to The interoperability issue is primarily being addressed
best suit the digital health ecosystem. Its impact is that by ICT professionals making use of various versions
collectively we are generating large amounts of real- of HL7 messaging standards for data exchange. Their
world data that cannot be used effectively. Yet, health data implementation and use require extensive data mappings
represent a valuable asset that needs to be well governed between proprietary data models and these standards.
and managed. All health-related concepts need to be represented by
The impact of the continuing fragmentation within data in a re-interpretable form to represent information
any health ecosystems not only contributes to physician in a formalized manner suitable for communication,
14
burnout but also limits AI developments aiming to interpretation, or processing by people or by automation.
support clinical practice. This limitation is primarily due Data mapping frequently results in a loss of information.
to health ecosystems’ inability to manage all relevant data Health data can be represented by any one of many
flows required to compile a comprehensive and complete terminologies. Anecdotally, we learned that many data
15
health record required to support any person’s continuity mapping activities are undertaken by administrative staff
of care. Incomplete health records, in turn, prevent the not necessarily suitably qualified to accurately interpret the
aggregation of quality data sets required to make “big meaning of terms or codes to accurately retain meaning
health data sets” available for research and multiple other when mapping data from one data set to another. Some

Volume 1 Issue 3 (2024) 12 doi: 10.36922/aih.3056