Published online: 16.08.

2019

56
© 2019 IMIA and Georg Thieme Verlag KG

Health Information Management:

Implications of Artificial Intelligence
on Healthcare Data and Information
Management
Mary H. Stanfill1, David T. Marc2
1
United Audit Systems, Inc., Cincinnati, OH, USA
2
The College of St. Scholastica, Department of Health Informatics and Information Management,
Duluth, MN, USA

Summary 1 Introduction tries regarding AI developments [3]. There

are financial pressures on the healthcare
Objective: This paper explores the implications of artificial intel-
Health information technology has greatly industry globally, with increasing demands
ligence (AI) on the management of healthcare data and informa-
impacted the health information manage- due to growing and aging population. The
tion and how AI technologies will affect the responsibilities and
ment (HIM) profession. HIM professionals industry needs labor-saving technology and
work of health information management (HIM) professionals.
are part of the allied health team and they techniques to better understand the health
Methods: A literature review was conducted of both peer-reviewed
support efforts to ensure the availability, of the population while managing the health
literature and published opinions on current and future use of AI
accuracy, integrity, and security of health- of a greater number of people and saving
technology to collect, store, and use healthcare data. The authors
care data. The digitizing of healthcare data money [4]. AI, whether or not it eliminates
also sought insights from key HIM leaders via semi-structured
has greatly impacted the responsibilities the need for a person to fill a job, can make
interviews conducted both on the phone and by email.
and work of HIM professionals requiring the workforce more efficient [5-9]. Accen-
Results: The following HIM practices are impacted by AI
many to take on more technical roles ture estimates that “key clinical health
technologies: 1) Automated medical coding and capturing
related to the collection, storage, and use AI applications” can create $150 billion
AI-based information; 2) Healthcare data management and data
of healthcare data. in annual savings for the U.S. healthcare
governance; 3) Patient privacy and confidentiality; and 4) HIM
The digitizing of healthcare data, as well economy by 2026 [10]. Even if a fraction
workforce training and education.
as advancements in computer processing of that figure is realized, that is a powerful
Discussion: HIM professionals must focus on improving the qual-
and data storage, has also enabled the devel- incentive for adopting AI solutions.
ity of coded data that is being used to develop AI applications.
opment of advanced algorithms in the form Beyond economic concerns, an additional
HIM professional’s ability to identify data patterns will be an
of Artificial Intelligence (AI). As of 2011, driver of AI technology is the sheer volume
important skill as automation advances, though additional skills
the U.S. Agency for Healthcare Research of healthcare data. Healthcare is experi-
in data analysis tools and techniques are needed. In addition,
and Quality (AHRQ) had compiled over encing an information boom. “The rapid
HIM professionals should consider how current patient privacy
17,000 algorithms and computer programs expansion of scientific knowledge and pace
practices apply to AI application, development, and use.
for healthcare evaluation, treatment, and of technological development have resulted
Conclusions: AI technology will continue to evolve as will the
administration [1]. In a recent white paper in an overwhelming sea of data that is diffi-
role of HIM professionals who are in a unique position to take on
on AI in Radiology, the Canadian Associ- cult to decipher and apply.” [11]. Physicians
emerging roles with their depth of knowledge on the sources and
ation of Radiologists stated “In the next 5 are drowning in data that requires ever more
origins of healthcare data. The challenge for HIM professionals
years, Canadian radiologists will see more sophisticated interpretation, yet are still
is to identify leading practices for the management of healthcare
competent AI applications incorporated expected to perform efficiently. The prom-
data and information in an AI-enabled world.
into PACS workflows, especially for labo- ise that AI “augments decision making by
Keywords rious tasks prone to human error such as clinicians by uncovering clinically relevant
Artificial Intelligence; Health Information Management; detection of lung nodules on x-rays or bone information hidden in a massive amount of
automation; medical coding; health workforce metastases on CT.” [2]. data” [5] is extremely enticing, particularly
Multiple factors are driving the devel- now, when there are clinician shortages
Yearb Med Inform 2019:56-64 opment of AI in healthcare. In the United worldwide. The needs-based shortage of
http://dx.doi.org/10.1055/s-0039-1677913 States (U.S.), legislative pressures are healthcare workers globally is estimated at
mounting to keep pace with other coun- approximately 17.4 million [12]. According

IMIA Yearbook of Medical Informatics 2019

 57
Health Information Management: Implications of Artificial Intelligence on Healthcare Data and Information Management

to the Canadian Association of Radiologists,

“…there is evidence that AI can improve
2 Changing Health and suggest applicable diagnosis and proce-
dure codes. Typically, a person reviews the
the performance of clinicians and that both Information Management suggested codes to determine the final code
clinicians and AI working together are better
than either alone” [2]. Indeed, AI technology Practices selection. This computer-assisted approach
to the medical coding process is becoming
is necessary to achieve the goal of “preci- A core responsibility of the HIM profession more common and has been credited with
sion medicine”. Precision medicine is an is ensuring the right information is provided measurable gains in coder productivity [19,
emerging medical model where medical to the right people to enable quality patient 20]. However, productivity impacts vary
decisions and treatments are tailored to the care [15]. Increased adoption of AI-enabled widely, depending on the specific deployment.
patient. “Precision medicine presupposes the applications and more sophisticated use of Some studies reported a drop in productivity
availability of massive computing power and AI systems by healthcare providers at the when medical coders were forced to validate,
algorithms that can learn by themselves at point of care have significant implications and frequently eliminate, a large number of
an unprecedented rate” [5]. for HIM practices. These include practical suggested codes. Still, a Cleveland Clinic
Predictions on when healthcare will expe- implications both for common HIM pro- study found that CAC increased their coder
rience widespread deployment of disruptive cesses, such as medical coding, as well as productivity by over 20% without reducing
AI applications vary widely. Though AI is more generally the core HIM responsibility quality when suggested codes were reviewed
developing rapidly, and there are current and to manage health data and information. This and edited by a medical coder [19]. The ref-
imminent uses of AI in healthcare, it is still section explores the impact of AI systems on erenced Cleveland Clinic study found that
largely immature. According to witnesses HIM practices for the following use cases: CAC alone, without the intervention of a
who testified before the U.S. Subcommittee • Automated medical coding; credentialed coder, however had a lower recall
on Information Technology of the House • AI-based diagnosis specificity; and precision rate.
Committee on Oversight and Government • AI-based early detection information. Adoption of CAC requires reengineer-
Reform at a series of hearings on AI held in ing the medical coding workflow to fully
2018, “narrow” AI, i.e. systems focused on Each use case includes examples of the integrate the CAC tool in the process and
specific tasks, is commonly used today but anticipated use of AI, discusses the associ- gain optimal efficiency [21]. Early adopters
more general systems that can work across ated impact to current HIM processes and of CAC in the U.S. reported that CAC had
multiple tasks are underdeveloped [13]. practices, and explores new opportunities “…improved medical coding workflows,
However, given the pace of development, and challenges to adapt HIM practices. increased medical coding accuracy, and
the timeline for AI in healthcare is years, balanced medical coding resources to focus
not decades [14]. on more volume and complex cases” [22].
Presuming AI will eventually be wide- Not all hospitals however have experienced
spread and affordable, there are implications 2.1 Automated Medical Coding these benefits [23]. Some implementations
for the management of healthcare data and A systematic literature review of published have failed entirely. Effective implementa-
information in an AI-enabled world which studies evaluating the performance of auto- tion of a CAC application requires interfaces
can greatly impact the HIM profession. mated coding and classification systems to work properly so the application can read
The purpose of this paper is to describe the indicated that automated coding systems have all documents relevant for coding. In addi-
results of a literature review and the findings been in use since at least the mid 1990’s [16]. tion clinical documents must comply with
from interviews with key HIM leaders. The Computer-assisted coding (CAC) is the term a consistent format dictated by the CAC
paper explores the relationship of the HIM that refers to the automated generation of vendor [24]. And where CAC has been most
profession and AI, focusing on the following medical codes reported on healthcare claims effective, a new role has emerged to fine tune
key aspects: 1) Changes in HIM practices that are derived from clinical documentation. the rules and train the system to adapt as the
for specific HIM use cases, including auto- CAC applications have been available since code sets and reporting requirements change.
mated medical coding and management the early 2000’s [17] with adoption rates As the technology advances, and machine
of AI-based information; 2) Changes in increasing markedly in recent years. Accord- learning techniques improve the capabilities
management of healthcare data and the ing to a report available through Research and of CAC tools, the medical coding workflow
need for evolving data practices and data Markets, the global market for CAC software will further evolve. A WinterGreen market
governance; 3) Legal, ethical, and regulatory is projected to reach $4.75 billion by 2022 at a shares research report released in 2017 stated
data challenges; and 4) Changes in the HIM compound annual growth rate of 11.5% [18]. that as much as 88% of medical coding in
workforce, including foreshadowing new North America is seeing the largest growth physician offices for billing purposes could
roles and skills that are required. The conclu- followed by Europe, Asia-Pacific, and the occur automatically without human review
sion presents steps the HIM profession can rest of the world. [25]. This report requires independent val-
take now to help advance the development CAC applications use natural language idation and more research is needed on the
of reliable AI applications and to respond to processing (NLP) to read and interpret clin- accuracy of these systems to rely on them,
their use in healthcare. ical documentation in patient health records but advancements in CAC are poised to

IMIA Yearbook of Medical Informatics 2019

58
Stanfill et al.

further augment the medical coding process. skin lesions has also performed as well as tiple points to consider including whether
Medical coding is a significant responsibility board-certified dermatologists [26, 27]. It reporting of diagnosis specificity using
of many HIM professionals currently and has been suggested that what might take an diagnostic test results should vary depending
this role will continue to evolve. experienced radiologist 30 years of radiolo- on the AI application itself. Some method is
There are significant opportunities for gy-pathology correlation to master may only needed to demonstrate that the AI application
medical coding professionals as CAC take an AI system hours or days to analyze meets the same degree of accuracy as phy-
advances to increase coding efficiency. The and learn in the future [28]. sicians. For example, reporting guidelines
fully automated coding workflow requires Code reporting guidelines for using might depend on whether the AI applica-
reengineering and a focus on data quality, diagnostic test results to add specificity to tion is approved or credentialed in some
which medical coders, with their intimate a diagnosis code vary by country. As AI manner. Reporting specificity based on AI
knowledge of the code sets and reporting systems become more adept and are proven results might also depend on whether the AI
requirements, are uniquely qualified to reliable in visual diagnosis, the need for application is employing supervised verses
address. In addition to assigning or validat- physicians to read images may become less unsupervised ML techniques. Unsupervised
ing codes on complex cases, medical coders necessary, perhaps done only by exception. ML is well known for feature extraction,
could also focus on validating aberrant This change in responsibilities could result whereas supervised ML, which goes through
coded data patterns across large groups of in either a decrease in code specificity or a training process to determine the best out-
cases. For example, a medical coder has the less consistency of international diagnosis puts, is more suitable for predictive modeling
knowledge to question the use of a code for code data, depending on a country’s code and is generally considered to provide more
an acute phase of a condition repeatedly reporting guidelines and how the guidelines clinically relevant results [6]. Thus, the type
for a patient, when the more likely data are adjusted to account for AI. For example, of AI and how the AI application is used in
pattern would be the acute code followed by currently in the U.S., “code assignment is the clinical workflow (e.g. whether AI-gener-
codes for the chronic phase or sequela. This based on the documentation by patient’s ated interpretations are validated or certified
code-specific pattern recognition is key in provider (i.e., the physician or other qualified as equally accurate compared to physicians)
validating accurate reporting for risk-scor- healthcare practitioner legally accountable could potentially be factors in determining
ing payment methodologies for example. for establishing the patient’s diagnosis)” future reporting requirements for diagnosis
Clearly, HIM professionals’ ability to iden- [29]. U.S. guidelines specifically state that code specificity.
tify data patterns to enhance business intel- clinically significant “laboratory, x-ray,
ligence or improve compliance with code pathologic, and other diagnostic results”
reporting requirements will be an important can be used for coding only if the test has
skill as automation advances. been “interpreted by a physician” [29]. In 2.3 Early Detection Information
the U.K., the NHS National Clinical Cod- AI systems are expected to assist healthcare
ing Standards, while less explicit than U.S. providers with early detection of likely or
guidelines, also imply that a physician has impending conditions, allowing for faster
2.2 Diagnosis Specificity to interpret diagnostic test results [30]. In intervention. ML algorithms are proving
AI systems are expected to assist healthcare contrast, the Canadian Coding Standards are effective in making inferences about specific
providers with diagnosis accuracy and speci- much more amendable to AI development. health risks and predicting health events. For
ficity. Medical specialties that utilize images Canadian medical coders are directed to use example, neural network algorithms have
for diagnosis (e.g. radiology, pathology, diagnostic results “when they clearly add proven effective in detecting strokes. Input
dermatology, ophthalmology) are partic- specificity in identifying the appropriate diag- variables analyzed by the algorithm include
ularly amendable to AI-aided diagnoses. nosis code for conditions documented in the stroke-related symptoms such as paresthesia
AI machine learning (ML) is very good at physician/primary care provider notes” [31]. of the arm or leg, acute confusion, vision
detecting anomalies in images, for example In Canada, there is no specific requirement alteration, problems with mobility, etc.
it has been proven effective in detecting lung that the test itself has to be interpreted by a This input data is analyzed to determine the
nodules on a radiologic image [2, 6, 9] and physician. Based on this varying guidance, probability of stroke [6]. There are other
congenital cataract as well as diabetic reti- in the instance where a physician has docu- examples of healthcare data being used to
nopathy on ocular image data [6, 26]. The mented a diagnosis, additional specificity of detect and predict future events including
sensitivity and specificity of deep learning that diagnosis in images interpreted by an AI hospital readmissions, sepsis, and surgical
algorithms, in detecting diabetic retinop- system alone (without a physician over-read) complications [32-34].
athy through retinal fundus photographs, would be lost in diagnosis data in the U.S. and Coding guidelines and standards for
for example, are both over 90%, which is possibly the U.K., whereas specificity would reporting suspected or impending conditions
“competitive against experienced physicians not necessarily be lost in Canada. also vary from one country to the next. In the
in the accuracy for classifying both normal Medical coding and reporting guidelines U.S., coders are directed to report a condition
and disease cases”[6]. An algorithm that can and standards will need to be adjusted to that remains “suspected and/or impending”
identify skin cancer by analyzing images of account for AI applications. There are mul- at the time of discharge as if it existed or

IMIA Yearbook of Medical Informatics 2019

 59
Health Information Management: Implications of Artificial Intelligence on Healthcare Data and Information Management

was established for a hospital inpatient SNOMED CT. Prefixes, such as Canada’s and exchange data. ‘In IoT, Things’ refers to
admission, but not to code it on an outpa- Q prefix, could be defined and appended to a device which is connected to the Internet
tient encounter [29]. For outpatient cases ICD codes. Perhaps ICD-11 extension codes and transfers the device information to other
the condition is coded to the highest degree could be defined to characterize the degree devices. “The future’s data will not be col-
of certainty [29]. Similarly, NHS National of certainty of a condition (e.g. unconfirmed, lected solely within the health care setting.
Coding Standards instructions are to code impending) or identify the source for the The proliferation of mobile sensors will
the diagnosis being “treated or investigated” diagnosis (e.g. clinician, AI system, patient). allow physicians of the future to monitor,
and an example is given of a “probable Again, there are multiple factors to consider. interpret, and respond to additional streams
myocardial infarction” reported with the Use of a status, prefix or extension to a code of biomedical data collected remotely and
code for an acute unspecified myocardial would require some mechanism to ensure it automatically” [7]. Such applications have
infarction [30]. According to the Canadian remains linked with the base code. Otherwise been in development for several years. More
Coding Standards however, impending or data validity would be a major concern. This than five years ago, a blood pressure cuff
threatened conditions are coded only when is the case for example when an “impend- that connects to a smartphone, and transmits
indexed as such in the Canadian version of ing” stroke is identified as an actual stroke data to a care provider was already available
the 10th revision of the International Statis- because the “impending” qualifier was lost. [36]. Devices are also available that measure
tical Classification of Diseases and Related Implications for insurance coverage or glucose levels, provide electrocardiogram
Health Problems (ICD) ICD-10-CA. In payment policy have also to be considered. readings, or even collect measures of peo-
addition, unconfirmed diagnoses in Canada As the industry continues to refine what is ple’s cognition and emotional health [37].
are reported with a specific “Q prefix” to deemed clinically relevant data/information, As wearable sensors improve, they will
denote the uncertainty associated with the medical coding standards and guidelines increasingly allow specific health parameters
code [31]. This variability and the inabil- will need to align with such data standards. to be tracked constantly and discreetly. They
ity in some countries to qualify reported may replace commonly worn items such as a
diagnoses as unconfirmed or uncertain is watch, may be worn under regular clothing,
concerning. Consider for example, if an or even built into “smart” clothing [38].
AI system triggers an alert for suspected 3 Changing Data These types of devices would conceivably
sepsis on a patient and the healthcare team
takes immediate action, thus intervening
Management Practices transmit data back to a healthcare provider,
potentially directly into an EHR, which
and preventing severe sepsis, the coding Increased adoption of AI-enabled applica- presents numerous challenges. It will be
and reporting of this circumstance may be tions, and more sophisticated use of these AI critical to track the source of this data as
missed, or inconsistently reported at best. applications by healthcare providers at the the accuracy, value, and clinical significance
Coding guidelines and standards will need point of care, holds practical implications for may be uncertain. In addition, today’s data
to be revised to capture this sequence of managing the data. HIM professionals have practices are entirely oriented toward an
events and support AI developments in early an opportunity to help develop, implement, episode of care. In AI-enabled healthcare,
detection of likely or impending conditions. and manage the policies and procedures the underlying organizing schema for health
This has broad implications and will require related to governing healthcare data, as well data needs to shift from dates of service
an interdisciplinary team to address the issue as to support the development, deployment, to the patient. It may require completely
fully, including standards developers and and assessment of AI models to ensure that different data architecture to collect, store,
members of the healthcare team as well as the technology can be trusted to improve care process, validate, interpret, and potentially
HIM professionals. and support greater efficiency. retrieve non-episodic ongoing streams of
One solution is to capture qualifiers to New and more varied data types are gen- patient-specific data.
diagnoses. If the functionality was built erated by AI-enabled applications affecting Manogaran and colleagues [39] proposed
into Electronic Health Records (EHRs), data practices and data governance. Today, a framework to support the collection,
the Health Level 7 (HL7) Fast Healthcare healthcare data is almost entirely encoun- transfer, and storage of data from multiple
Interoperability Resources (FHIR) standard ter-based. Healthcare data is collected during data streams. They emphasized that the
framework could potentially be leveraged an encounter with specific interaction with security of data must occur at numerous
to qualify diagnoses [35]. For example, the a care provider. However, healthcare data stages including during the collection of
FHIR code system verification status defines also includes streams of data collected data from devices, the transfer of data
codes as provisional, differential, confirmed, remotely and automatically from multiple between devices, the storage of data, and
and refuted [35]. A status could potentially data sources. As the Internet of Things (IoT) during the application and use of the data.
be added to reflect AI as the source for a expands further into healthcare, it is neces- Additionally, how the data is received from
condition or diagnosis. Alternatively, diag- sary to develop infrastructures to support the various streams and integrated into a single
nosis qualifiers could also be addressed by proliferation and use of these data streams. system poses a challenge. Data streams
the clinical terminology or classification IoT is a connection of physical objects with may include structured, semi-structured, or
system itself, which is demonstrated in network connectivity that are used to collect unstructured data and for integration to occur

IMIA Yearbook of Medical Informatics 2019

60
Stanfill et al.

there is a need for standardization. Initiatives requiring millions of observations to reach quality controls, and data quality manage-
such as International Standard for Metadata acceptable performance levels” [14]. Thus ment, must keep pace with evolving data
Registries (ISO/IEC 11179) aim to support researchers and developers need access to types and data uses. For example, data qual-
what is referred to as ‘semantic interopera- large sets of health data from thousands of ity management in healthcare organizations
bility’ between data that may be expressed patients. The reliability of an AI application today focuses on assuring data is fit for use
differently across devices and technologies is dependent upon the quality of the data for the organization’s business operations,
[40]. Semantic interoperability is intended that was used to develop and train it. “At its decision-making and planning. More focus
to support the unambiguous exchange of core, AI is reliant upon data. If the data itself is needed on detecting, assessing, and fixing
data. One method for standardization is is incomplete, biased, or skewed in some data defects in a systematic way. Data gov-
to create globally unique cross-reference other fashion, the AI system is at risk of ernance has never been a higher priority in
identifiers for data elements that are seman- being inaccurate” [13]. However, it’s widely healthcare as it “empowers users to trust the
tically equivalent using eXtensible Markup recognized in the U.S. that data in EHRs predictions of analytics models in their deci-
Language (XML) standards, even though and claims databases need “careful curation sion-making because there is certainty that
the data elements may have different names and processing before they are usable” [14]. the data and algorithms can be trusted” [44].
[40]. The Open Data Element Framework Healthcare data are highly heterogeneous, As advances in AI enable precision
(O-DEF) was developed by The Open Group ambiguous, noisy, and incomplete [26]. medicine, HIM professionals will need to
and can support the categorization, naming, Data curation (i.e., managing data to make develop practices to enable precision HIM.
and indexing of data using a controlled it more useful) requires significant financial Treating all healthcare data and information
vocabulary that associates data elements investment and without investing resources the same will no longer be practical or
with structured unique identifiers so that to support data curation the healthcare efficient in an era of big data. More robust
equivalencies and similarities between data industry risks producing ML models based data analytics and processes need to be
can be easily determined [41]. These identi- on factually inaccurate data [8]. The adop- established to identify data patterns and
fiers can be the basis of an indexing schema tion of data governance principles can trends and address data outliers. “Precision
where a data element from one device can be help organizations ensure that the people, medicine attempts to ensure that the right
integrated with a data element from another processes, and systems involved in AI ini- treatment is delivered to the right patient at
device because they both share the same tiatives are held accountable for ethical use the right time by taking into account several
equivalent content evidenced by the same and deployment, the process is transparent, aspects of patient’s data, including variability
structured unique identifier. O-DEF works the result has integrity, the information is in molecular traits, environment, EHRs and
well for collaborating enterprises, but may protected, the approach is compliant with lifestyle” [26]. Precision HIM attempts to
not serve the purpose of integrating data organizational and legal practices, the tech- ensure the right data and information is
from disparate systems and organizations. nology is available, the method of AI devel- delivered to the right person at the right time
Alternatively, other frameworks such as opment is retained, and when appropriate by taking into account the data source and
those from the World Wide Web Consor- the healthcare data is disposed of properly the people, processes, and technology that
tium (W3C) that focus on data integration [43]. These principles can help support the interface with that data to ensure it is used
of web-based data like RDF (Resource use of AI models that minimize the risk to and reused appropriately.
Description Framework), OWL (Ontology patients, providers, developers, and health-
Web Language), and SKOS (Simple Knowl- care organizations.
edge Organization System) may be more Evolving data governance principles
useful [42]. Data integration challenges will are necessary and must be a priority for all 4 Legal, Ethical, and
require an interdisciplinary team to address
the issue. HIM professionals can seek to
healthcare organizations. Developing clear,
consistent, and standardized policies and Regulatory Data Challenges
examine how existing information models procedures for creating and managing cur- The use of healthcare data to develop AI
can be leveraged within an organization to rent and emerging sources of data is a key applications has introduced substantial legal,
support a data governance framework that enabler to development of AI applications. ethical, and regulatory challenges. Patient
accommodates multiple data streams. The Data sources can include EHR data, lab data, privacy is a key concern affecting how AI
utilization of existing vocabularies may imaging data, claims data, various types of is developed and tested. Development of AI
serve to accelerate the collection and use of master data (e.g., enterprise master patient applications may require updates to privacy
data from non-episodic sources. index), patient-generated data, and metadata and confidentiality laws and regulations,
An additional challenge is the need for as well as a real-time streaming data from which vary widely. In the U.K., protection
quality healthcare data. ML techniques medical devices. Several issues need to be of health information centers on obtaining
require substantial amounts of data to ensure managed, such as data sparsity, redundancy, explicit consent from the patient in order
algorithms work accurately and are applied and missing values [26]. Data governance, to share information with any third party
appropriately to their targeted goals. “ML including data modeling, data standards and that is not in a direct care relationship with
algorithms are highly data hungry, often definitions, data mapping, data auditing, data the patient. Researchers must apply to the

IMIA Yearbook of Medical Informatics 2019

 61
Health Information Management: Implications of Artificial Intelligence on Healthcare Data and Information Management

Health Research Authority’s Confidenti- rithms they cannot understand? Will physi- to go about the business of “providing the
ality Advisory Group (CAG) for approval cians use a system they cannot understand? clinical truth in their coding and resisting
to access confidential patient information Can the developer be held responsible? The the perverse incentives” [50]. Therefore,
without patients’ consent [45]. In the U.S., problem is complicated since the reasoning with the access and use of health informa-
government regulation is less strict. Privacy in an AI application is difficult, often too tion for the development and deployment
and confidentiality of protected health complex to understand [48]. AI applications of AI models, HIM professionals should be
information are addressed in the Health evolve and change constantly in unforesee- involved to ensure policies and procedures
Insurance Portability and Accountability Act able ways as they are “learning” from data are being developed, amended accordingly,
(HIPAA). HIPAA provides data privacy and [9]. Though mechanisms to ensure AI appli- and followed to account for the influence of
security provisions for safeguarding medical cations are safe and effective are still being AI technology. Although HIM professionals
information and allows for sharing protected formulated, prevailing approaches include an are just beginning to work with AI technol-
health information without patient consent expectation that algorithms can be inspected. ogy, there has already been notable impacts
specifically for the purposes of “treatment, “Each algorithm should be able to explain on the HIM workforce.
payment and operations”[46]. How the U.K. its output” [13]. To advance deployment and
or U.S. approaches will be interpreted on acceptance of AI applications, developers
cases related to data sharing for AI devel- will need to be able to produce the algorithm
opment is largely undetermined. The U.K.
consent requirement, and the definition of
for inspection, support why the algorithm
works, and ensure the application can meet
5 Response of the Health
a “direct care relationship,” was challenged expected outcomes in testing or certification Information Management
in 2017 in a published case study. The case
study alleged that a technology company,
procedures. Product master data, which
includes data about the components that
Workforce
Google DeepMind, did not have a direct make up the product, may include informa- Healthcare technology has greatly impacted
patient care relationship with every patient tion on the algorithm deployed. In the future, the way care is approached and delivered.
included in the data shared and thus “held individual patient health information may The digitizing of healthcare data has sup-
data on millions of Royal Free patients and include the algorithm that was applied to the ported efforts to automate processes that
former patients since November 2015, with patient’s data in order to validate or authenti- were previously done manually. These pro-
neither consent, nor research approval” cate healthcare decisions. In addition, there cesses have inevitably impacted the health-
[47]. This case study underscores the need may be a need to audit AI events for reporting care workforce, including the HIM profes-
to examine current privacy laws and regu- purposes. HIM professionals can establish sion. There is a greater need for employees
lations to determine how they may apply to the necessary data governance principles that that have technical skills to better collect,
AI applications. The U.S. Subcommittee on must be adopted for AI applications to be manage, and use healthcare data. Sandefer
Information Technology recommends that implemented successfully within healthcare and colleagues [51] evaluated data from a
federal agencies conduct such a review and, organizations. workforce survey that yielded responses
where necessary, update existing regulations Another aspect that deserves attention is from 6,475 healthcare professionals that
to account for the addition of AI [13]. HIM the need to balance the financial incentive were largely from HIM. The survey asked
professionals are involved with developing to make processes more efficient with the respondents to rate the percentage of their
and implementing organizational policies ethical and legal uses of health information. time they spent on current tasks and how
regarding privacy and security of health For example, the financial motivators to much they anticipate they will spend on these
information, training staff, and ensuring adopt CAC for the sole purpose of coding tasks 10 years in the future. The findings of
compliance. Therefore, with the access and a higher level of care must be tempered by the study suggested that many HIM profes-
use of health information for the develop- ethical considerations. HIM professionals sionals spent significant time on diagnostic
ment and deployment of AI models, HIM involved in the clinical coding process can and procedural coding and records process-
professionals should explore current privacy greatly impact the amount of funding pro- ing, but they expected these tasks to decline
practices considering how they might apply vided to a healthcare organization. Hoyle the most in the future while leadership,
to AI applications and how they might be [49] and Shepherd [50] argued that HIM teaching, and informatics tasks are expected
amended to account for AI technology. professionals are positioned as advocates for to increase. Historically, the HIM profession
In addition to data privacy and protection, the ethical use of technology and data. HIM has focused on medical records and coding.
another looming legal issue is liability and professionals must urge healthcare organiza- However, the profession has evolved into
accountability for the use of AI applications. tions to consider the ethical frameworks and more diverse roles and continues to change
Questions on who is ultimately liable for practice guides not just deemed appropriate with technological advances. Today, many
patient care decisions based on, or aided by, for health information professionals, but HIM professionals find themselves in diverse
an AI application are yet to be answered. also for CAC and AI technologies. These roles related to healthcare leadership, teach-
Should healthcare providers be held fully activities will provide support for the HIM ing, technology, compliance, quality, and
responsible for decisions suggested by algo- professionals in healthcare organization informatics [51, 52].

IMIA Yearbook of Medical Informatics 2019

62
Stanfill et al.

In 2018, Sandefer [53] evaluated data from The evolving use of healthcare data for and guidelines with evolving data types and
a workforce survey of 274 senior-level profes- AI applications is already impacting the roles standards. In addition, as AI technologies
sionals within clinical (e.g., hospitals, clinics) and responsibilities of HIM professionals. present new and varied types of source data,
and non-clinical (e.g., software vendors, con- HIM professionals are findings themselves in HIM professionals have an opportunity to
sulting firms) organizations. The goal of the more leadership roles that govern healthcare influence the development of mechanisms
survey was to identify the needed job skills, data and technology, and more technical to collect and integrate emerging data types,
competencies, and education required by HIM roles that involve the access and use of including non-episodic ongoing streams of
professionals to meet future workforce needs. healthcare data for reporting and evalua- patient data and algorithms in product master
Seventy-two percent of clinical respondents tion purposes [51]. With some tasks being data for example. The adoption of data stan-
reported that at least half of coding functions automated, there will likely be continuing dards and vocabularies that support semantic
will be automated, and 50 % reported that opportunity for HIM professionals to take on interoperability is part of the solution to the
more than half of the coding functions will more tasks that focus on the data collection, data integration challenge [40] and one that
be automated in the near future. The paper validation, analysis, and overall the ethical HIM professionals should participate in
suggests that the application of natural lan- use of that data. HIM professionals who evaluating and testing.
guage processing combined with the quality currently find themselves working in medical HIM professionals should also partic-
of voice to text translation will support coding who embrace automated coding have ipate in developing the data governance
improvements in extracting meaning from an opportunity to transition into a role that framework within healthcare organizations
unstructured data, which will greatly revo- focuses on data validation to improve the to establish mechanisms to collect emerging
lutionize the healthcare industry. quality of healthcare data. However, to emerge data types from various sources, manage
Automation is also expected to impact into these roles, these professionals will need the policies and procedures related to the
the HIM workforce beyond just influencing technical training related to methods and tools access and use of data, and develop methods
how diagnostic and procedural coding is for data storage, acquisition, and analytics. to validate the reliability and impact of AI
approached. Data analytics has been more With advancements in technology, many technology. This includes considering how
prolific across the profession. More profes- professions are realizing the need for greater evolving data structures impact the use and
sionals are moving into roles to evaluate data competence in computational thinking skills reuse of data and the related policy impli-
related to financial, operational, and clinical to better translate data into abstract concepts cations (e.g., data reporting requirements,
performance [54]. HIM professionals are and understand data-based reasoning [57]. payment policy). It also includes for example
becoming more involved with developing Although exact details on how AI technol- ensuring data governance practices include
solutions for healthcare organizations to ogies will impact the future of HIM are not product master data (e.g., data about the
better manage and use data. For instance, yet known, current workforce studies suggest algorithms deployed) to support efforts to
HIM professionals are actively participating that HIM professionals are going to continue audit, inspect, or certify AI applications.
in the development of policies, procedures, to work in more technical roles and will These endeavors will require HIM profes-
and best practices to ensure data are being therefore support AI developments and use. sionals to have the technical knowledge to
used ethically and abiding by the required analyze and monitor AI tools and the nec-
laws when research or data reporting is essary technical skills related to collecting
being adopted [55]. However, in the future, and managing healthcare data in AI-enabled
HIM professionals are going to need to be
more involved in developing similar pol-
6 Conclusion healthcare. To acquire such technical skills,
HIM professionals may need to seek addi-
icies and procedures to accommodate AI AI has and will continue to impact the way tional education or training.
developments. To date, there is very little decisions are made in healthcare. For exam- There are significant data management
attention on the needs for data governance ple, decisions are influenced by ML algo- practices as well as laws and regulations
to support AI. Without having a workforce rithms that support the prediction of future surrounding the use of healthcare data that
to support AI data governance, there will events, or the use of clinical decision support have the potential to either impede or enable
likely be barriers to widespread adoption and systems that aid in the detection of anomalies development of AI applications. HIM profes-
use. For instance, past efforts to implement in diagnostic images. The decisions that HIM sionals can support future AI developments
ICU mortality risk scores have been met professionals make are also being impacted. today by increasing data validation efforts
with reluctance due to a lack of trust in the For instance, CAC has supplemented a med- and beginning to evaluate relevant policies
technology, despite the obvious benefits the ical coder’s role in selecting diagnostic and and processes. HIM professionals should
technology may serve [56]. By engaging procedural codes for healthcare claims. The analyze coded data patterns and establish
more stakeholders in the development of the promise that AI can support a more efficient processes to validate coded data across large
technology, including HIM professionals, a decision-making process with greater accu- groups of cases. HIM professionals must
culture of acceptance may be achieved by racy is certainly a promise worth exploring. focus on detecting, assessing, and fixing
adopting principles of data governance that HIM professionals should participate in data defects in a systematic way in order to
offer enterprise-wide technology support. efforts to align medical coding standards improve the quality of current healthcare data

IMIA Yearbook of Medical Informatics 2019

 63
Health Information Management: Implications of Artificial Intelligence on Healthcare Data and Information Management

that is being used to develop AI applications. of Radiologists. White Paper on Artificial In- CAC Practice Guidance (2013 update). AHIMA
Other examples of steps HIM professionals telligence in Radiology. Can Assoc Radiol J Practice Brief 2013.
2018;69(2):120-35. 22. O’Dowd E. HIT Infrastructure: HIT Infrastructure;
can take now include ensuring the proper 3. Trump DJ. Executive order on maintaining 2018. [cited 2018]. Available from: https://hitin-
laws and regulations are being followed (e.g., American leadership in artificial intelligence. In: frastructure.com/news/computer-assisted-cod-
ensuring only authorized personnel and tech- President EOot, editor. Federal Register: White ing-systems-help-orgs-handle-increasing-data.
23. Butler M. Computer-Assisted Coding Reality
nology accesses clinical data), beginning to House; 2019. p. 3967-72.
Check. J AHIMA 2019;90(6):10-3.
explore current privacy practices in light of 4. Noorbakhsh-Sabet N, Zand R, Zhang Y, Abedi V.
Artificial Intelligence Transforms the Future of 24. Eramo LA. J AHIMA 2017. [cited 2019]. Available
how they may apply to AI applications, and Health Care. Am J Med 2019. from: https://journal.ahima.org/2017/06/01/has-
establishing collaborative relationships with cac-lived-up-to-its-promise-providers-vendors-
5. Ganapathy K, Abdul SS, Nursetyo AA. Artificial
weigh-in/.
data standards developers and informaticists intelligence in neurosciences: A clinician’s per- 25. Curtiss ET, Susan E. Computer Assisted Coding:
involved in developing AI applications. spective. Neurol India 2018;66(4):934-9. Market Shares, Strategies, and Forecasts, World-
Although there is an emphasis on creating 6. Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, et al. wide 2016 to 2022. WinterGreen Research; 2017.
Artificial intelligence in healthcare: past, present 26. Miotto R, Wang F, Wang S, Jiang X, Dudley
policies and procedures to accommodate and future. Stroke Vasc Neurol 2017;2(4):230-43. JT. Deep learning for healthcare: review, op-
AI technology, HIM professionals will also 7. Johnson KW, Torres Soto J, Glicksberg BS, portunities and challenges. Brief Bioinform
find that there are emerging opportunities Shameer K, Miotto R, Ali M, et al. Artificial 2018;19(6):1236-46.
for careers related to the greater adoption of Intelligence in Cardiology. J Am Coll Cardiol 27. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM,
AI. HIM professionals are well situated to 2018;71(23):2668-79. Blau HM, et al. Dermatologist-level classification
8. eHealth Initiative. The Machines are Here: The of skin cancer with deep neural networks. Nature
proactively manage and monitor data gover- Influence of AI on Healthcare. EHIDC; 2018. 2017;542(7639):115-8.
nance, data sets, and data models related to 9. Pesapane F, Volonte C, Codari M, Sardanelli F. 28. King BF, Jr. Artificial Intelligence and Radiology:
the implementation and use of AI. AI tech- Artificial intelligence as a medical device in radiol- What Will the Future Hold? J Am Coll Radiol
nologies are not intended to replace health- ogy: ethical and regulatory issues in Europe and the 2018;15(3 Pt B):501-3.
United States. Insights Imaging 2018;9(5):745-53. 29. CMS. ICD-10-CM Official Guidelines for Coding
care workers, but individuals who are able and Reporting FY 2019. CDC; 2018.
10. Collier M, Fu R, Yin L, Christiansen P. Artificial
to adapt to new workflows and processes Intelligence (AI): Healthcare’s New Nervous 30. NHS Digital. National Clinical Coding Standards
may replace those who cannot. There are System; 2017. ICD-10 5th Edition. NHS; 2018.
31. Canadian Institute for Health Information. Canadi-
wonderful opportunities for career moves 11. Miriovsky BJ, Shulman LN, Abernethy AP. Impor-
an Coding Standards for Version 2018 ICD-10-CA
and advancements for those who continue to tance of health information technology, electronic
health records, and continuously aggregating data and CCI. CIHI; 2018.
increase their knowledge of data analytical to comparative effectiveness research and learning 32. Saqib M, Sha Y, Wang MD. Early Prediction
methods and tools. health care. J Clin Oncol 2012;30(34):4243-8. of Sepsis in EMR Records Using Traditional
ML Techniques and Deep Learning LSTM
The future of AI holds the promise of 12. Mesko B, Hetenyi G, Gyorffy Z. Will artificial
Networks. Conf Proc IEEE Eng Med Biol Soc
a more effective and efficient healthcare intelligence solve the human resource crisis in
healthcare? BMC Health Serv Res 2018;18(1):545. 2018;2018:4038-41.
system built on a strong foundation of reli- 13. Hurd H, Kelly R. Rise of the Machines: Artificial 33. Golas SB, Shibahara T, Agboola S, Otaki H, Sato J,
able and accurate data. HIM professionals Intelligence and its Growing Impact on U.S. Nakae T, et al. A machine learning model to predict
the risk of 30-day readmissions in patients with
manage and support the entire continuum Policy. https://oversight.house.gov/wp-content/
heart failure: a retrospective analysis of electronic
of healthcare data from the collection of the uploads/2018/09/AI-White-Paper-.pdf: Subcom-
mittee on Information Technology Committee on medical records data. BMC Med Inform Decis
data to the use and disposition of that data. Oversight and Government Reform; 2018.
Mak 2018;18(1):44.
AI technology will continue to evolve and so 34. Bertsimas D, Dunn J, Velmahos GC, Kaafarani
14. Obermeyer Z, Emanuel EJ. Predicting the Future
HMA. Surgical Risk Is Not Linear: Derivation
will the role that HIM professionals would - Big Data, Machine Learning, and Clinical Med-
and Validation of a Novel, User-friendly, and
play to support this technology. The chal- icine. N Engl J Med 2016;375(13):1216-9.
Machine-learning-based Predictive OpTimal Trees
15. Dimick C. Health information management 2025.
lenge for HIM professionals is to identify J AHIMA 2012;83(8):24-31.
in Emergency Surgery Risk (POTTER) Calculator.
leading practices to achieve precision HIM Ann Surg 2018;268(4):574-83.
16. Stanfill MH, Williams M, Fenton SH, Jender, RA, 35. HL7. FHIR Release 3 HL7.org2017 [cited 2018
and develop practice standards for the man- Hersh WR. A systematic literature review of auto- December 10]. Available from: https://hl7.org/
agement of healthcare data and information mated clinical coding and classification systems. J implement/standards/fhir/codesystem-condi-
Am Med Inform Assoc 2010;17(6):646-51. tion-ver-status.html.
in an AI-enabled world. 17. Workgroup AC-ACe-H. Delving into computer-as- 36. Cohn J. The Health Tech Report: Robot Will See
sisted coding. J AHIMA 2004;75(10):48A-H. You Now. The Atlantic 2013(March):59-67.
18. Research and Markets Report. Computer Assisted 37. Metz R. The smartphone app that can tell you’re
Coding Market by Software (NLP, Structured depressed before you know it yourself: MIT Tech-
References input, Integrated Systems), Service, Application
(Automated Encoding, Clinical Coding Audit,
nology Review; 2018 [cited 2018 December 10].
Available from: https://www.technologyreview.
1. AHRQ. Medal.org (Medical Algorithms): Agency Management Reporting) – Global Forecast to com/s/612266/the-smartphone-app-that-can-tell-
for Healthcare Research and Quality, Innovation 2022. Research and Markets; 2017. youre-depressed-before-you-know-it-yourself/.
Exchange; 2011 [updated November 9, 2011; 19. Dougherty M, Seabold S, White SE Study reveals 38. Sawh M. The best smart clothing: From biometric
cited 2018 December 10]. Available from: https:// hard facts on CAC. J AHIMA 2013;84(7):54-6. shirts to contactless payment jackets: Smart fabrics
innovations.ahrq.gov/qualitytools/medalorg-med- 20. Crawford M. The truth about computer-assisted and garments on sale now or hitting the shops
ical-algorithms. coding. J AHIMA 2013;84(7):24-7. soon Wareable.com2018 [cited 2019 January
2. Tang A, Tam R, Cadrin-Chênevert A, Guest W, 21. Comfort A, D’Amato C, Isom C, Rihanek T, Thom- 14]. Available from: https://www.wareable.com/
Chong J, Barfett J, et al; Canadian Association as-Flowers D. Automated Coding Workflow and smart-clothing/best-smart-clothing.

IMIA Yearbook of Medical Informatics 2019

64
Stanfill et al.

39. Manogaran G, Varatharajan R, Lopez D, Kumar 47. Powles J, Hodson H. Google DeepMind and health- hospitals better protect the privacy of electronic
PM, Sundarasekar S, Thota C. A new architec- care in an age of algorithms. Health Technol (Berl) medical records? Perspectives from staff members
ture of Internet of Things and big data ecosystem 2017;7(4):351-67. of health information management departments.
for secured smart healthcare monitoring and 48. Price N. Artificial Intelligence in Health Care: Ap- Health Inf Manag 2017;46(2):87-95.
alerting system. Future Gener Comput Syst plications and Legal Issues. The Sci Tech Lawyer 56. Delahanty RJ, Kaufman D, Jones SS. Development
2018;82:375-87. 2017;14(1):10-3. and Evaluation of an Automated Machine Learning
40. Bandyopadhyay D, Sen J. Internet of things: Appli- 49. Hoyle P. Health Information is central to changes Algorithm for In-Hospital Mortality Risk Adjust-
cations and challenges in technology and standard- in healthcare: A clinician’s view. Health Inf Manag ment Among Critical Care Patients. Crit Care Med
ization. Wirel Pers Commun 2011;58(1):49-69. 2018;48(1):48-51. 2018;46(6):e481-e8.
41. The Open Group. Open Data Element Framework 50. Shepheard J. Ethical leadership and why health 57. Davies AFD, Gorbis M. Future Work Skills 2020.
(O-DEF) Version 1.0: The Open Group; 2016 information management professionals need to be Institute for the Future, University of Pheonix
[cited 2019 January 9]. Available from: https:// involved. Commentary on Health Information is Research Institute; 2011.
publications.opengroup.org/c163. central to changes in healthcare: a clinician’s view.
42. W3C. Semantic Web: W3C; 2017 [cited 2019 Health Inf Manag 2018;2018.
January 9]. Available from: https://www.w3.org/ 51. Sandefer R, Marc D, Mancilla D, Hamada D.
standards/semanticweb/. Survey Predicts Future HIM Workforce Shifts. J Correspondence to:
43. Empel S. The way forward. AHIMA develops AHIMA 2015;86(7):32-5. Mary H. Stanfill
information governance principles to lead health- 52. Riley M, Robinson K, Prasad N, Gleeson B, 1379 N. Gale Mountain Rd.
care toward better data management. J AHIMA Barker E, Wollersheim D, et al. Workforce sur- Pocatello, ID 83204
2014;85(10):30-2; quiz 4. vey of Australian graduate health information USA
44. eHealth Initiative. Artificial Intelligence in Health- managers: Employability, employment, and
Tel: +1 513 477 2544
care. EHIDC; 2018. knowledge and skills used in the workplace.
45. NHS Health Research Authority. Confidentiality Health Inf Manag 2019. Fax: +1 513 873 9704
Advisory Group: NHS; 2018 [cited 2018 Decem- 53. Sandefer R. New Market Research Supports HIM E-mail: [email protected]
ber 14]. Available from: https://www.hra.nhs.uk/ Reimagined’s Call for Professional Adaptation. J
approvals-amendments/what-approvals-do-i-need/ AHIMA 2018;89(5):20-3. David T. Marc
confidentiality-advisory-group/. 54. Marc DT, Robertson J, Gordon L, Green-Lawson 1200 Kenwood Ave
46. U.S. Department of Health & Human Services. ZD, Gibbs D, Dover K, et al. What the Data Duluth MN 55811
The HIPAA Privacy Rule: HHS; 2015 [cited 2018 Say About HIM Professional Trends. J AHIMA USA
December 14]. Available from: https://www.hhs. 2017;88(5):25-31. Tel: +1 218-625-4892
gov/hipaa/for-professionals/privacy/index.html. 55. Sher ML, Talley PC, Cheng TJ, Kuo KM. How can E-mail: [email protected]

IMIA Yearbook of Medical Informatics 2019