Page 22 - AIH-1-3

Artificial Intelligence in Health Optimizing EHRs to support AI

collection of data points to determine if a pattern exists, or of health information from citizen demographic or
for algorithm development to make decisions or support identification data by adopting a privacy-by-design
decision-making or make predictions. Training any AI approach. Every citizen needs to have control over
model requires large amounts of representative data. The their data and how it is used.
number and types of accessible data points determine the (3) Facilitate the linkage of health-care data with omics
accuracy of the model or a possible set of rules that can data, that is with the inclusion of data representing
be identified. The delivery of health services is data centric the various “omes” of an organism, to enable making
where access to accurate and timely data is critical for sense of vast amounts of collected data to build
decision-making. AI approaches making use of these data next generations of clinical decision support and
require the use of advanced analytics and access of a large research methods and tools. At present, the use of
amount of source data. Data-driven approaches are relevant genetic sequencing and variation information is not
for the provision of automated reporting as automation part of routine clinical practice because health-care
58
relies on pre-determined rules or assumptions. There professionals do not have the knowledge or skills.
are significant limitations regarding access to source data Most importantly, there is a lack of automated tools
collected and stored in legacy systems. that can reliably associate phenotypic data from EHRs
with many types of omics data to provide personal
4. New and emerging technologies and precision care. Large-scale, well-annotated, and

This review’s findings have confirmed that the interoperable high-quality EHR data will have an immense impact
and scalable ecosystem-wide architectures can be adopted, on bringing omics and healthcare together.
the knowledge about the health ecosystem’s data supply (4) Facilitate the linkage of healthcare and data with
chain, and the relationships between information models, the human physiome 59,60 comprising personal and
terminologies, and ontologies with data exchange protocols. mechanistic computational multi-scale models. Such
Health ecosystem-wide data supply chains need to: models enable the provision of new types of insight
(1) Include data/information flow requirements to into not only our understanding of human physiology
support collaborative, person-centered life-long, and and pathology but also predictions of disease and
episodic continuity of care. Episodic events of multiple prognosis. Such insights are the result of using
service episodes can also exist. Such episodes represent ontology-based EHR data linkage that parameterize
a treatment plan for one specific health issue such as these models that are able to run surprisingly reliable
for cancer care or a pregnancy as recorded by multiple simulations at individual or population levels.
systems over a period of time. Data collections able Computational physiology and systems biology
to meet all information needs associated with any provide us with unprecedented precision to provide
treatment/care plan require identifiable data transfers value-based and appropriate care as well as drive
between any number of individual and organizational more effective drug and medical device development
health-care service providers as well as devices. and faster compliance through in silico medicine and
61
Specific data needs will differ based on the individual’s clinical trials.
health status, treatment/care plans (life-long and Data governance protocols, legislation, and regulations
episodic), and geographical location relative to service need to facilitate or enable these requirements to deliver
availability at any point in time. optimal benefits of data use, including any type of effective
(2) Facilitate the aggregation of de-identified data and reporting automation and AI adoption.
identifiable data to classify any number of grouping
protocols (populations) or individuals to suit specific 4.1. Next-generation EHR/EMR system
data use cases. Data relationships will vary by use characteristics
case and need to include data from systems other Next-generation EHR/EMR systems are designed to
than data collected and stored by EHR/EMRs, such reduce or eliminate these gaps and improve the generation
as clinical registries. Over time, such registries are of quality data within a connected digital health ecosystem.
expected to be generated from vendor/technology- New health platforms need to be engineered to integrate
neutral federated cloud-based health data repositories personal health information received from emerging
including CDRs. For some use cases, linkages may technologies in the fields of personal health and well-being,
also need to include relationships between weather including apps and wearables. EHR/EMR systems and
events or environmental status at a specific point in CDRs should become a valuable computable data source
time or by geographical location, such as vaccination for research and evaluation purposes as well as be enriched
rates. CDR design needs to prioritize the separation by data from external data sources while complying with

Volume 1 Issue 3 (2024) 16 doi: 10.36922/aih.3056