q10/random_k2/random_1.md

Chapter2

ElectronicHealthRecords:ASurvey

Rajiur Rahman
Department of Computer Science
Wayne State University
Detroit, MI
rajiurrahman@wayne.edu

Chandan K. Reddy
Department of Computer Science
Wayne State University
Detroit, MI
reddy@cs.wayne.edu

2.1
2.2
2.3

2.4

2.5

2.6
2.7
2.8
2.9

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History of EHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Components of EHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Administrative System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.1
Laboratory System Components & Vital Signs . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2
Radiology System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.3
Pharmacy System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.4
Computerized Physician Order Entry (CPOE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.5
Clinical Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.6
Coding Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
International Classification of Diseases (ICD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1
ICD-9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1.1
ICD-10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1.2
ICD-11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1.3
Current Procedural Terminology (CPT)
2.4.2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT) . .
2.4.3
Logical Observation Identifiers Names and Codes (LOINC) . . . . . . . . . . . . . . . .
2.4.4
RxNorm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.5
International Classification of Functioning, Disability, and Health (ICF) . . . .
2.4.6
Diagnosis-Related Groups (DRG) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.7
Unified Medical Language System (UMLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.8
Digital Imaging and Communications in Medicine (DICOM) . . . . . . . . . . . . . .
2.4.9
Benefits of EHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Enhanced Revenue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.1
Averted Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.2
Additional Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.3
Barriers to Adopting EHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Challenges of Using EHR Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Phenotyping Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22
22
24
24
24
25
26
26
27
28
28
29
30
31
32
32
33
34
35
37
37
38
38
38
39
40
42
45
47
51
52

21

22

Healthcare Data Analytics

2.1 Introduction

An Electronic Health Record (EHR) is a digital version of a patient’s medical history. It is a
longitudinal record of patient health information generated by one or several encounters in any
healthcare providing setting. The term is often used interchangeably with EMR (Electronic Med-
ical Record) and CPR (Computer-based Patient Record). It encompasses a full range of data rel-
evant to a patient’s care such as demographics, problems, medications, physician’s observations,
vital signs, medical history, immunizations, laboratory data, radiology reports, personal statistics,
progress notes, and billing data. The EHR system automates the data management process of com-
plex clinical environments and has the potential to streamline the clinician’s workflow. It can gener-
ate a complete record of a patient’s clinical encounter, and support other care-related activities such
as evidence-based decision support, quality management, and outcomes reporting. An EHR sys-
tem integrates data for different purposes. It enables the administrator to utilize the data for billing
purposes, the physician to analyze patient diagnostics information and treatment effectiveness, the
nurse to report adverse conditions, and the researcher to discover new knowledge.

EHR has several advantages over paper-based systems. Storage and retrieval of data is obviously
more efficient using EHRs. It helps to improve quality and convenience of patient care, increase
patient participation in the healthcare process, improve accuracy of diagnoses and health outcomes,
and improve care coordination. It also reduces cost by eliminating the need for paper and other
storage media. It provides the opportunity for research in different disciplines. In 2011, 54% of
physicians had adopted an EHR system, and about three-quarters of adopters reported that using an
EHR system resulted in enhanced patient care [1].

Usually, EHR is maintained within an institution, such as a hospital, clinic, or physician’s office.
An institution will contain the longitudinal records of a particular patient that have been collected
at their end. The institution will not contain the records of all the care provided to the patient at
other venues. Information regarding the general population may be kept in a nationwide or regional
health information system. Depending on the goal, service, venue, and role of the user, EHR can
have different data formats, presentations, and level of detail.

The remainder of this chapter is organized as follows. Section 2.2 discusses a brief history
of EHR development and Section 2.3 provides the components of EHRs. Section 2.4 presents a
comprehensive review of existing coding systems in EHR. The benefits of using EHRs are explained
in more detail in Section 2.5, while the barriers for the widespread adoption of EHRs are discussed
in Section 2.6. Section 2.7 briefly explains some of the challenges of using EHR data. The prominent
phenotyping algorithms are described in Section 2.8 and our discussion is concluded in Section 2.9.

2.2 History of EHR

The first known medical record can be traced back to the fifth century B.C. when Hippocrates

prescribed two goals for medical records [2]:

• A medical record should accurately reflect the course of disease.

• A medical record should indicate the probable cause of disease.

Although these two goals are still appropriate, EHR has a lot more to offer. Modern EHR can
provide additional functionalities that could not be performed using paper-based systems.

Electronic Health Records: A Survey

23

Modern-day EHR first began to appear in the 1960s. Early EHRs were developed due to physi-
cians’ concerns about the increasing complexity and size of medical data. Data retrieval was much
faster using digital format. In 1967, Latter Day Saints Hospitals in Utah started using Health Eval-
uation through Logical Programming (HELP) software. HELP is notable for its pioneering logical
decision support features. In 1969, Harvard Medical School developed its own software Computer
Stored Ambulatory Record (COASTER) and Duke University began to develop The Medical Record
(TMR).

In 1970, Lockheed unveiled the Technicon Medical Information Management System/ Techni-
con Data System (TDS). It was implemented at El Camion Hospital in California. It came with a
groundbreaking Computer Provided Order Entry (CPOE) system. In 1979, Judith Faulkner, a com-
puter programmer established Human Services Computing Inc., which developed the Chronicles
data repository. The company later became Epic Systems. It was initially based on a single longi-
tudinal patient record and designed to handle enterprise-wide data from inpatient, ambulatory, and
payer environments.

In 1985, The Department of Veterans Affairs launched the automated data processing system,
Decentralized Hospital Computer Program (DHCP), which includes extensive clinical and admin-
istrative capabilities within its medical facilities. It received the Smithsonian Award for best use
of Information Technology in Medicine in 1995. The current variant of DHCP is VistA (Veterans
Health Information Systems and Technology Architecture). By providing care to over 8 million vet-
erans operating in 163 hospitals, 800 clinics, and 135 nursing homes, VistA manages one of the
largest medical system in the United States [4]. In 1983, Epic Systems launched a patient schedul-
ing software program called Cadence. This application helped clients to improve resource utiliza-
tion and manage patient access. In 1988, Science Application International Corporation (SAIC)
secured a $1.02 billion dollar contract from the U.S. Government to develop a composite healthcare
system. In 1992, Epic Systems introduced the first Windows-based EHR software named Epic-
Care. Allscripts released the first software with an electronic prescribing solution for physicians in
1998.

From 2000 and beyond, EHR software has been increasingly trying to incorporate other func-
tionalities to become an interactive companion for physicians and professionals. In January 2004,
President George W. Bush launched an initiative for the widespread adaptation of EHRs within the
next 10 years. He said in his State of the Union Address, “By computerizing health records, we can
avoid dangerous medical mistakes, reduce costs, and improve care” [5]. In January 2009, in a speech
at George Mason University, President Barack Obama said “[EHRs] will cut waste, eliminate red
tape, and reduce the need to repeat expensive medical tests. It just won’t save billions of dollars
and thousands of jobs – it will save lives by reducing the deadly but preventable medical errors that
pervade our health care system” [6]. The data from a National Ambulatory Medical Care Survey
(NAMCS) and Physicians Workflow mail survey shows that in the year 2011, 54% of the physicians
had adopted an EHR system. About three-quarters of the adopters reported that their system meets
the federal “meaningful use” criteria. Almost half (47%) of the physicians said they were some-
what satisfied, and 38% reported being very satisfied with their system. About three-quarters of the
adopters reported that EHR has resulted in enhanced patient care. Nearly one-half of physicians
without an EHR system at the time of the survey said they had plans for purchasing one within the
next year [1].

24

Healthcare Data Analytics

2.3 Components of EHR

The main purpose of EHR is to support clinical care and billing. This also includes other func-
tionalities, such as improving the quality and convenience of patient care, improving the accuracy
of diagnoses and health outcomes, improving care coordination and patient participation, improving
cost savings, and finally, improving the general health of the population. Most modern EHR systems
are designed to integrate data from different components such as administrative, nursing, pharmacy,
laboratory, radiology, and physician’ entries, etc. Electronic records may be generated from any de-
partment. Hospitals and clinics may have a number of different ancillary system providers; in that
case, these systems are not necessarily integrated to the main EHR system. It is possible that these
systems are stand-alone, and different standards of vocabularies have been used. If appropriate inter-
faces are provided, data from these systems can be incorporated in a consolidated fashion; otherwise
a clinician has to open and log into a series of applications to get the complete patient record. The
number of components present may also vary depending on the service provided. Figure 2.1 shows
different components of an EHR system.

2.3.1 Administrative System Components

Administrative data such as patient registration, admission, discharge, and transfer data are key
components of the EHR. It also includes name, demographics, employer history, chief compli-
ant, patient disposition, etc., along with the patient billing information. Social history data such
as marital status, home environment, daily routine, dietary patterns, sleep patterns, exercise pat-
terns, tobacco use, alcohol use, drug use and family history data such as personal health history,
hereditary diseases, father, mother and sibling(s) health status, age, and cause of death can also be
a part of it. Apart from the fields like “comments” or “description,” these data generally contain
<name-value> pairs. This information is used to identify and assess a patient, and for all other
administrative purposes. During the registration process, a patient is generally assigned a unique
identification key comprising of a numeric or alphanumeric sequence. This key helps to link all the
components across different platforms. For example, lab test data can create an electronic record;
and another record is created from radiology results. Both records will have the same identifier key
to represent a single patient. Records of a previous encounter are also pulled up using this key. It is
often referred to as the medical record number or master patient index (MPI). Administrative data
allows the aggregation of a person’s health information for clinical analysis and research.

2.3.2 Laboratory System Components & Vital Signs

Generally, laboratory systems are stand-alone systems that are interfaced to the central EHR sys-
tem. It is a structured data that can be expressed using standard terminology and stored in the form of
a name-value pair. Lab data plays an extremely important part in the clinical care process, providing
professionals the information needed for prevention, diagnosis, treatment, and health management.
About 60% to 70% of medical decisions are based on laboratory test results [7]. Electronic lab data
has several benefits including improved presentation and reduction of error due to manual data en-
try. A physician can easily compare the results from previous tests. If the options are provided, he
can also analyze automatically whether data results fall within normal range or not.

The most common coding system used to represent the laboratory test data is Logical Obser-
vation Identifiers Names and Codes (LOINC). Many hospitals use their local dictionaries as well
to encode variables. A 2009–2010 Vanderbilt University Medical Center data standardization study
found that for simple concepts such as “weight” and “height,” there were more than five internal rep-
resentations. In different places there are different field names for the same feature and the values

Electronic Health Records: A Survey

25

Nursing

Pharmacy

Healthcare Service
Provider

Lab

Radiology

Clinical

FIGURE 2.1: Various components of EHR.

are stored with different units (e.g., kilograms, grams, and pounds for weight; centimeters, meters,
inches, and feet for height).

Vital signs are the indicators of a patient’s general physical condition. It includes pulse, respi-
ratory rate, blood pressure, body temperature, body mass index (BMI), etc. A typical EHR system
must provide the option to accommodate these kinds of variables.

2.3.3 Radiology System Components

In hospital radiology departments, radiology information systems (RIS) are used for managing
medical imagery and associated data. RIS is the core database to store, manipulate, and distribute pa-
tient radiological data. It uses Current Procedural Terminology (CPT) or International Classification
of Diseases (ICD) coding systems to identify procedures and resources. Generally, an RIS consists
of patient tracking, scheduling, result reporting, and image tracking capabilities. RIS is usually used
along with a picture archiving communications system (PACS), which is a medical technology for

26

Healthcare Data Analytics

providing economical storage and convenient access to the digital images. An RIS can generate an
entire patient’s imagery history and statistical reports for patients or procedures. Although many
hospitals are using RIS, it may or may not be integrated with the central EHR system.

2.3.4 Pharmacy System Components

In hospitals and clinics, the pharmacy department’s responsibility is to maintain the inventory,
prescription management, billing, and dispensing medications. The pharmacy component in EHR
will hold the complete medication history of a patient such as drug name, dosage, route, quantity,
frequency, start and stop date, prescribed by, allergic reaction to medications, source of medication,
etc. Pharmacists serve an important public health role by administering immunizations and must
have the capabilities to document these services and share this information with other healthcare
providers and public health organizations. They assure safe and effective medication and support-
ing patient-centered care. Pharmacies are highly automated in large hospitals. Again, it may be
independent of central EHRs. The Food and Drug Administration (FDA) requires all the drugs to
be registered and reported using a National Drug Code (NDC). Coding systems used are NDC,
SNOMED, and RxNorm.

2.3.5 Computerized Physician Order Entry (CPOE)

Computerized Physician Order Entry (CPOE) is a very important part of EHRs. It is a system
that allows a medical practitioner to enter medical orders and instructions for the treatment of a
patient. For example, a doctor can electronically order services to laboratory, pharmacy, and radi-
ology services through CPOE. Then it gets propagated over a network to the person responsible
for carrying out these orders. As a digital system, CPOE has the potential to reduce medication-
related errors. It is possible to add intelligent rules for checking allergies, contradictions, and other
alerts. The primary advantages of CPOE are the following: overcomes the issue of illegibility, fewer
errors associated with ordering drugs with similar names, more easily integrated with decision sup-
port systems, easily linked to drug-drug interaction warning, more likely to identify the prescribing
physician, able to link the adverse drug event (ADE) reporting systems, able to avoid medication
errors like trailing zeros, create data that is available for analysis, point out treatment and drug of
choice, reduce under- and overprescribing, and finally, the prescriptions can reach the pharmacy
quicker. While ordering, a professional can view the medical history, current status report from a
different module, and evidence-based clinical guidelines. Thus, CPOE can help in patient-centered
clinical decision support.

If used properly, CPOE decreases delay in order completion, reduces errors related to hand-
writing or transcriptions, allows order entry at point-of-care or off-site, provides error checking for
duplicate or incorrect doses or tests, and simplifies inventory and positing of charges. Studies have
shown that CPOE can contribute to shortened length of stay and reduction of cost [8]. There are
some risks involved in adopting CPOE as well. It may slow down interpersonal communication in
an emergency situation. If each group of professionals (e.g., physicians and nurses) works alone in
their workstations, it may create ambiguity about the instructions. These factors led an increase in
mortality rate by 2.8%–6.5% in the Children’s Hospital of Pittsburgh’s Pediatric ICU when a CPOE
system was introduced [8]. Frequent alerts and warnings may also interrupt workflow. The adapta-
tion rate of CPOE is slow. It may be partly due to physicians’ doubt about the value of CPOE and
clinical decision support.

Electronic Health Records: A Survey

27

2.3.6 Clinical Documentation

A clinical document contains the information related to the care and services provided to the
patient. It increases the value of EHR by allowing electronic capture of clinical reports, patient
assessments, and progress reports. A clinical document may include [9]

• Physician, nurse, and other clinician notes

• Relevant dates and times associated with the document

• The performers of the care described

• Flow sheets (vital signs, input and output, and problems lists)

• Perioperative notes

• Discharge summaries

• Transcription document management

• Medical records abstracts

• Advance directives or living wills

• Durable powers or attorney for healthcare decisions

• Consents (procedural)

• Medical record/chart tracking

• Release of information (including authorizations)

• Staff credentialing/staff qualification and appointments documentations

• Chart deficiency tracking

• Utilization management

• The intended recipient of the information and the time the document was written

• The sources of information contained within the document

Clinical documents are important because documentation is critical for patient care, serves as a
legal document, quality reviews, and validates the patient care provided. Well-documented medical
records reduce the re-work of claims processing, compliance with CMS (Centers for Medicare and
Medicaid Services), Tricare and other payer’s regulations and guidelines, and finally impacts coding,
billing, and reimbursement. A clinical document is intended for better communication with the
providers. It helps physicians to demonstrate accountability and may ensure quality care provided
to the patient. A clinical document needs to be patient centered, accurate, complete, concise, and
timely to serve these purposes.

The clinical document architecture (CDA) [10] is an XML-based electronic standard developed
by the Health Level 7 International (HL7) to define the structure. It can be both read by human eyes
and processed by automatic software.

28

Healthcare Data Analytics

2.4 Coding Systems

Standards play an important role in enhancing the interoperability of health information systems
and the purposeful use of EHR systems. Collecting and storing information following standard cod-
ing systems provide better and accurate analysis of the data, seamless exchange of information,
improved workflow, and reduced ambiguity. A complete healthcare system is complex and requires
various EHR products. Different vendors have implemented standards in their own way. This prac-
tice has resulted in a significant variation in the coding practices and implemented methods for
which systems cannot interoperate. To create an interoperable EHR, standardization is critical in
the following four major areas:

• Applications interaction with the users

• System communication with each other

• Information processing and management

• Consumer device integration with other systems and application

Interoperability between the different EHR systems is a crucial requirement in the “meaningful use
of certified EHR technology” to receive incentives. That is why conforming to a standard coding
system is very important. In a practical EHR, we need standards for

• Clinical vocabularies

• Healthcare message exchanges

• EHR ontologies

There are three organizations mainly responsible for developing the related standards: Health Level
Seven (HL7), Comit´e Europeen de Normalisation-Technical Committee (CEN-TC), and the Amer-
ican Society of Testing and Materials (ASTM). HL7 develops healthcare-related standards that are
widely used in North America. CEN-TC is a prominent standard developing organization working
in 19 member states in Europe. Both HL7 and CEN-TC collaborate with ASTM. Along with the
standards developed by these organizations, EHR systems must comply with the Health Insurance
Portability and Accountability (HIPAA) Act [11] to conserve the security and privacy of patient
information.

2.4.1 International Classification of Diseases (ICD)

ICD stands for International Classification of Diseases, which is the United Nations-sponsored
World Health Organization’s (WHO) official coding standard for diseases, diagnoses, health man-
agement, and clinical purposes [12]. It first appeared as the International List of Causes of Death
in 1893, adopted by the International Statistical Institute. Since then it has been revised according
to advancements in medical science and healthcare. Since the creation of WHO in 1948, WHO has
maintained ICD. WHO published ICD-6 in 1949, and it was the first coding system in which mor-
bidity was incorporated [13]. It also included mental disorders for the first time. The U.S. Public
Health Services issued International Classification of Diseases, Adapted for Indexing of Hospitals
Records and Operation Classification (ICDA) in 1959. It was revised regularly and used to classify
diseases and mortality until WHO published the ninth revision of ICD.

The 1967 WHO Nomenclature Regulations specified that the member nations should use the
most recent ICD version for mortality and morbidity statistics. Along with the storage and retrieval

Electronic Health Records: A Survey

29

of epidemiological and clinical information, it allows for the compilation of morbidity statistics for
more than 100 WHO member nations. About 70% of the world’s health expenditure in reimburse-
ment and resource allocation is also done using ICD codes [14]. It is used to classify diseases and
related problems, and provides a system of codes for a wide variety of diseases, signs, symptoms,
abnormal findings, complaints, social circumstances, and external causes of injury or disease. It
is the global foundation for providing common language in disease and health-related information
and statistics exchange. ICD is comprehensive and organizes information into standard groups that
allows for the following [15]:

• Easy storage, retrieval, and analysis of health information for evidence-based decision-

making.

• Sharing and comparing health information between hospitals, regions, and countries.

• Data comparison in the same location across different time periods.

2.4.1.1 ICD-9

ICD ninth revision is the most popular coding system published by WHO in 1978. It was de-
signed to promote comparability of classification, collection, processing, and presentation of mortal-
ity statistics. Its clinical modification, ICD-9-CM, was published by the U.S. Public Health Services
in the following year to meet the statistical needs. The modified version had expanded the number
of diagnostic codes and developed a procedure coding system. It has more than 13,000 codes and
uses more digits representing the codes compared to ICD-9. It is the system that is used to encode
all the diagnoses for healthcare services in the United States. It is maintained by the National Center
for Health Statistics (NCHS) and the Center for Medicare and Medicaid Services (CMS). Both the
departments are part of the federal department of Health and Human Services. The ICD-9-CM code
set is organized in three volumes and consists of tabular lists and alphabetical indices.

• Volume 1: Disease and Injuries Tabular List

• Volume 2: Disease and Injuries Alphabetical Index

• Volume 3: Procedures Tabular List and Alphabetic Index

ICD-9-CM is updated every year to keep up-to-date with medical trends and diseases. NCHS has
the responsibility to update Volumes 1 and 2, and CMS maintains Volume 3. Concerned parties
from both the public and private sectors can propose changes to it. The major updates take effect
on October 1 every year and minor updates occur on April 1. It is a statistical tool that converts the
diagnoses and procedures into number codes. Its primary applications are

• Reporting and research

• Monitoring the quality of patient care

• Communication and transactions

• Reimbursement

• Administrative uses

30

Healthcare Data Analytics

2.4.1.2 ICD-10

The tenth version was endorsed by WHO in 1990 during the 43rd World Health Assembly.
The first full version of ICD-10 was released in 1994. The first step of implementing ICD-10 was
taken by NCHS awarding a contract to the Center for Health Policy Studies (CHPS) to evaluate
ICD-10 for morbidity purposes within the United States. A prototype of clinically modified ICD-10
was developed after a thorough evaluation of ICD-10 by a technical advisory panel. After strong
recommendations, NCHS proceeded with implementing a revised version of ICD-10-CM. During
1995–1996, further work on the enhancement of ICD-10-CM was done incorporating experiences
from ICD-9-CM and through collaborating with many speciality groups like American Association
of Dermatology, American Academy of Neurology, American Association of Oral and Maxillo-
facial Surgeons, American Academy of Orthopedic Surgeons, American Academy of Pediatrics,
American College of Obstetricians and Gynecologists, American Urology Institution, and National
Association of Children hospitals and other related institutions. In 1999, ICD-10 was implemented
in the United States for mortality reporting. Death statistics and data regarding leading causes of
death for the years 1999 and 2000 were published using ICD-10 [16]. In October 2002, ICD-10
was published in 42 languages. In June/July 2003, the American Health Information Management
Association (AHIMA) and American Hospital Association (AHA) jointly conducted a pilot study
to test ICD-10-CM. In their study, they have compared ICD-9-CM and ICD-10-CM and the initial
results indicated ICD-10-CM is an improvement over ICD-9-CM; and ICD-10-CM is more applica-
ble in non-hospital environments compared to ICD-9-CM. Canada, Australia, Germany, and others
countries have their own revision of ICD-10 by adding country specific codes. The revisions are
ICD-10-CA, ICD-10-AM, ICD-10-GM, and so on. The standard for procedure codes ICD-10-PCS
was also developed during the same time frame to replace the Volume 3 of ICD-9-CM. The first
revision of it was released in 1998.

ICD-9-CM is around thirty years old. Many of its categories are full, and there have been
changes in technology. Some of them are also not descriptive enough. A newer coding system is
needed, which would enhance reimbursement, better facilitate evaluation of medical processes and
outcomes, and be flexible enough to incorporate emerging diagnoses and procedures. For exam-
ple, in a scenario where a patient had a fractured left wrist and, after a month a fractured right
wrist, ICD-9-CM cannot identify left versus right; additional information is required. However,
ICD-10-CM can report distinguishing left from right. It can also characterize initial and subsequent
encounters. Further, it can describe routine healing, delayed healing, nonunion, or malunion.

The major differences between ICD-10 and ICD-9-CM are [17]

• ICD-10 has 21 categories of diseases; while ICD-9-CM has only 19 categories.

• ICD-10 codes are alphanumeric; while ICD-9-CM codes are only numeric.

• ICD-9-CM diagnoses codes are 3–5 digits in length, while ICD-10-CM codes are 3–7 char-

acters in length.

• Total diagnoses codes in ICD-9-CM is over 14,000; while ICD-10-CM has 68,000.

• ICD-10-PCS procedure codes are 7 characters in length; while ICD-9-CM procedure codes

are 3–4 numbers in length.

• ICD-10-PCS total number of codes is approximately 87,000. The number of procedure codes

in ICD-9-CM is approximately 4,400.

The Center for Medicare and Medicaid Services (CMS) guidelines mandated a conversion from
ICD-9-CM to ICD-10-CM by October 1, 2014 in the United States. Adopting a new coding system
will have the following benefits:

Electronic Health Records: A Survey

31

• Improve patient care. The increased detail in the coding system will improve the measurement
of quality, safety, and efficacy of care, which will ultimately lead to improved patient care.

• Determine the severity of illness and prove medical necessity. ICD-10 codes are more granular
and provide option to input the level of sickness along with complexity of disease of a patient
in a code-based system.

• Improve research. The better and more accurate organization of code will be able to more
precisely classify diseases and injuries, and correlate them with the cause, treatment, and out-
come. The collected data will be less ambiguous and such a better-defined structure of the
information will make data analysis easier. Information processing will be easier with newer
coding system and it will open new opportunities for developing an intelligent prediction sys-
tem. It will also allow the United States. to conduct comparative research with other countries
that are already using ICD-10.

• Lend insight to the setting of health policy. With improved data analytics made possible

through ICD-10, policy makers will be able to make informed policy decisions.

• Facilitate improved public health reporting and tracking. The comprehensive coding structure

will allow concerned agencies to track public health risks and trends in greater detail.

• Improve clinical, financial, and administrative performance and resource allocation. The qual-
ity of data can reveal essential insights. It will allow the administrators to track time and work-
force spent for procedures. This will help administrators to allocate resources more efficiently
and achieve positive financial and managerial outcomes.

• Increase the accuracy of payment and reduce the risk that claims will be rejected for incorrect
coding. Reduced number of claim denials is expected due to higher specificity of ICD-10. It
will also create a better electronic record of evidence to receive proper payment from govern-
ment payers, insurers, hospitals, health systems, and others.

• Make room for new procedures and techniques. The adaptation ability of ICD-9-CM is lim-
ited, where all the codes are already utilized and has no more room for new codes. The ex-
panded coding of ICD-10 will be able to accommodate new procedures.

• It will have other facilities like reduced hassle of audits, help preventing and detecting health-

care fraud and abuse.

2.4.1.3 ICD-11

The World Health Organization is currently working on the eleventh revision of ICD. The final
publication of ICD-11 is expected by 2017 [18]. The beta draft [19] was made public online for
initial comments and feedback in May 2012. This development of ICD-11 revisions is taking place
in a web-based platform called iCAT, where all the concerned parties collaborate. For interested
groups or people, there are options to give structured input and field testing of revised editions. It
will be available in multiple languages and free to download for personal use. In ICD-11, disease
entries will have definitions and descriptions of the entry and category in human readable forms.
The current version ICD-10 has only the title headings. There are 2,400 codes in ICD-11 that are
different in the ICD-10 code set, where 1,100 codes are related to external causes and injury [20].

Although the beta version does not support any social network platforms, the support of web-
sites such as Wikipedia, Facebook, Social Reader, LinkedIn, etc. is in the plan. The structure of
definitions and other contents related to diseases and procedures will be defined more accurately. It
will be more compatible with EHRs and other technologies.

32

Healthcare Data Analytics

2.4.2 Current Procedural Terminology (CPT)

Current Procedural Terminology (CPT) is a set of medical codes developed, maintained, and
copyrighted by the American Medical Association (AMA). CPT codes are a list of descriptive terms,
guidelines, and identifying codes of medical, surgical, and diagnostic services designed to provide
uniform communication language among physicians, coders, patients, accreditation organizations,
and payers for administrative, financial, and analytic purposes.

It was first created by the AMA in 1966. The first edition contained mainly surgical codes. A
significant development took place for the second edition, which was published in 1970. The sec-
ond edition contained 5 digits instead of 4 digits, and it included lab procedures. In 1983, the Health
Claim Financial Administration (HCFA), which is now known as the Center for Medicine and Med-
icaid Services (CMS), merged its own Common Procedure Coding System (HCPCS) with CPT and
mandated CPT would be used for all Medicare billing. Every year the new version is released in
October. The Healthcare Common Procedures Coding System (HCPCS, often pronounced as “hick
picks”) is another set of codes developed by AMA based on CPT. Although the CPT coding system
is similar to ICD-9 and ICD-10, it describes the treatment and diagnostic services provided while
ICD codes describe the condition or the disease being treated. CPT is used only in inpatient settings.

2.4.3 Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT)

Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT) is a comprehensive,
computer-processible, multilingual clinical and healthcare terminology, originally created by the
College of American Pathologists (CAP). SNOMED was started as Systematic Nomenclature of
Pathology (SNOP) in 1965 [21]. It was enhanced further and SNOMED was created in 1974. It
had two major revisions in 1979 and 1993. In 1999, SNOMED-CT was created by the merger of
SNOMED Reference Terminology (SNOMED-RT) developed by the CAP and Clinical Terms Ver-
sion 3 (CTV3) developed by the National Health Services of the United Kingdom. This merged
version was first released in 2002. SNOMED-RT had a vast coverage of medical specialities with
over 12,000 concepts. It was designed for the retrieval and aggregation of healthcare information
produced by multiple organizations or professionals. The strong suit of CTV3 was its coverage
of terminologies for general practice. With more than 200,000 concepts, it was used to store pri-
mary care encounter information and patient-based records [22]. Currently SNOMED has more
than 311,000 concepts with logic-based definitions organized into a hierarchy. In July 2003, the
National Library of Medicine (NLM) on behalf of the U.S. Department of Health and Human Ser-
vices signed a contract with CAP to make SNOMED-CT available for users. Since April 2007, it
has been owned, maintained, and distributed by a newly formed Denmark-based nonprofit organiza-
tion named International Health Terminology Standards Development Organization (IHTSDO) [9].
CAP collaborates with IHTSDO and continues to provide support for SNOMED-CT operations.
More than 50 countries use SNOMED-CT.

SNOMED-CT is a valuable part of EHR. Its main purpose is to encode medical and healthcare-
related concepts and support recording of data. It provides a consistent way to store, index, retrieve,
and aggregate clinical data across different sites. It also helps to organize data in a more meaningful
way and reduce the variability of the data collection and management process. Its extensive coverage
includes clinical findings, symptoms, diagnoses, procedures, body structures, organisms and other
etiologies, substances, pharmaceuticals, devices, and specimens [23].

SNOMED-CT has a logical and semantic relationship between concepts. It has a multiaxial
hierarchy, which allows different level of details of information. Its extensible design permits the
integration of national, local, and vendor specific requirements. It primarily consists of four compo-
nents.

• Concept Codes: numerical codes to identify terms

Electronic Health Records: A Survey

33

• Descriptions: textual descriptions of the concept codes

• Relationships: represents relationships between the concept codes

• Reference Sets: used for grouping concept codes or descriptions. Supports cross mapping to

other classification standards.

SNOMED-CT can be mapped to other well-known terminologies like ICD-9-CM, ICD-10, and
LOINC. Renowned standards like ANSI, DICOM, HL7, and ISO are supported by it. In a joint
project with WHO, it is providing insights for the upcoming ICD-11.

SNOMED-CT has some fundamental differences from ICD. It is mainly a terminology system
while ICD is a classification system. SNOMED-CT is designed to encode and represent data for
clinical purposes [24]. Information coded with ICD is used for statistical analysis, epidemiology,
reimbursement, and resource allocation. SNOMED-CT facilitates the information input into the
EHR and provides standardization for primary data purposes while ICD codes enable retrieval for
secondary data purposes.

2.4.4 Logical Observation Identifiers Names and Codes (LOINC)

Logical Observation Identifiers Names and Codes (LOINC) is a universal code system for iden-
tifying laboratory observations and clinical test results. In response to the demand for electronic
clinical data, it was created in 1994 by Regenstrief Institute Inc., an Indianapolis-based nonprofit
research organization affiliated with Indiana University. It was originally called Laboratory Ob-
servations, Identifiers, Names, and Codes and the development was sponsored by NLM and other
government and private agencies. Original sources of information include the following [25]:

• Silver book for International Union of Pure and Applied Chemistry

• International Federation of Clinical Chemistry

• Textbooks of Pathology

• EuCliD (European Clinical Database)

• Expertise and work of the LOINC members

LOINC coding system helps to improve the communication of information. In January 2009, Re-
genstrief Institute released a Windows operating system-based mapping software called Regenstrief
LOINC Mapping Assistant (RELMA) where codes can be searched and local codes can be mapped
to a LOINC database. The current version of LOINC is LOINC 2.46 released in December 2013.
With more than 600 new users per month, it has 27,000 users from 158 different countries. LOINC
vocabulary continues to grow till today.

Each LOINC record represents a single test result. A record consists of six fields [26].

• Component: what is measured and evaluated (e.g., glucose, hemoglobin)

• Kind of property: characteristics of the component that is measured (e.g., mass, length, con-

centration, volume, time stamp, etc.)

• Time: observation period of the measurement

• System: the specimen or the substance, in context of which the measurement was done (e.g.,

blood, urine)

• Scale: the measurement scale (e.g., quantitative, nominal, ordinal, or narrative)

34

Healthcare Data Analytics

• Method (optional): the procedure performed for measurement

Certain parameters and descriptors related to the test are explicitly excluded in LOINC from obser-
vation name. They are made as fields of test/observation report message [25]. These fields are

• The instrument used for testing

• Fine details of the sample or the site of collection

• The priority of the test

• Who verified the result

• Size of the sample

• Place of testing

LONIC’s overall organization is divided into four categories: laboratory, clinical, attachments, and
surveys. The laboratory component is further divided into subcategories such as chemistry, hema-
tology, serology, microbiology (includes parasitology and virology), and toxicology. The clinical
attributes are vital signs, hemodynamics, intake/output, EKG, obstetric ultrasound, cardiac echo,
urologic imaging, gastroendoscopic procedures, pulmonary ventilator management, and other clin-
ical observations [25]. It also contains information about nursing diagnoses and nursing interven-
tions.

2.4.5 RxNorm

RxNorm is a drug vocabulary maintained and distributed by the National Library of Medicine
[27]. It assigns standard names to the clinical drugs and drug delivery devices available in the United
States. It is used as a basis for the capture and presentation of drug-related information in EHRs. In
2001, NLM started to develop RxNorm for modeling clinical drugs in the Unified Medical Language
System (UMLS) in consultation with the HL7 vocabulary technical committee and the Veterans
Administration [28]. It was developed to standardize the medication terminology that would reduce
the missed synonymy in clinical drugs [29]. Additional goals were to facilitate electronic capture
of related data, improve interoperability by supporting information exchange across platforms and
systems, develop clinical decision support, and provide opportunity for research.

RxNorm follows a standard for naming drugs. The normalized name of a drug include the fol-

lowing components [28]:

• IN: Ingredient of the drug.

• DF: Dose form of the drug.

• SCDC: Semantic clinical drug component. It represents the ingredients and strength.

• SCDF: Semantic clinical drug form. It represents the ingredient and dose form.

• SCD: Semantic clinical drug. It represents the ingredient, strength, and dose form.

• BN: Brand name. This is the formal name for a group of drugs containing a specific active

ingredient.

• SDBC: Semantic branded drug component. It represents the branded ingredient and strength.

• SBDF: Semantic branded drug form. It represents the branded ingredient and dose form.

• SDB: Semantic branded drug. It represents the branded ingredient, strength, and dose form.

Electronic Health Records: A Survey

35

RxNorm organizes drugs by concept. A concept is a set of names with similar meaning at a specific
level of abstraction. It can distinguish similar drugs from different providers using concepts. The
concepts and relationships between each other form a semantic network.

2.4.6 International Classification of Functioning, Disability, and Health (ICF)

The International Classification of Functioning, Disability, and Health, commonly known as
ICF, is a classification of health-related components of function and disability. ICF concentrates on
the functionality and body structure of people with a given health condition or disability rather than
diagnosis or diseases. It does not account for the cause of disability. It is a unified and standard
framework first developed by the World Health Organization (WHO) in 1980 [30]; initially it was
known as International Classification of Impairments, Disabilities, and Handicaps (ICIDH). After
years of coordinated revision, in May 2001, the 191 member states of WHO agreed to adopt ICF
as the standard coding method of functioning and disability. In June 2008, the American Physical
Therapy Association (APTA) joined WHO for endorsing ICF. ICF is the only method of its kind. It
has been developed and tested for applicability in more than 40 countries.

Body functions and disability can be viewed as interactions between health condition and per-
sonal and environmental factors. ICF has mainly two parts: Functioning and disability, and Con-
textual factors. It can be categorized into further subparts. The components of ICF are listed below
[31]:

• Functioning and disability

– Body functions

* Mental functions
* Sensory functions and pain
* Voice and speech functions
* Functions of the cardiovascular, hematological, immunological, and respiratory

systems

* Genitourinary and reproductive functions
* Neuromusculoskeletal and movement-related functions
* Functions of the skin and related structures

– Body structures

* Structure of the nervous system
* The eye, ear, and related structures
* Structures involved in voice and speech
* Structures related to cardiovascular, immunological, and respiratory systems
* Structures related to digestive, metabolic, and endocrine systems
* Structures related to genitourinary and reproductive systems
* Structures related to movement
* Skin and related structures

– Activities and participation

* Learning and applying knowledge
* General tasks and demands
* Communication
* Self-care
* Domestic life

36

Healthcare Data Analytics

* Interpersonal interactions and relationships
* Major life areas
* Community, social, and civic life

• Contextual factors

– Environmental factors

* Products of technology
* Natural environment and human-made changes to the environment
* Support and relationships
* Attitudes
* Service, systems, and policies

– Personal factors
* Gender
* Age
* Coping styles
* Social background
* Education
* Profession
* Past and current experience
* Overall behavior pattern
* Character and other factors

ICF complements WHO’s classification of disease scheme, ICD-10. ICD contains diagnosis and
health condition-related information, but not functional status. Together they constitute the WHO
Family of International Classifications (WHO-FIC) shown in Figure 2.2.

FIGURE 2.2: WHO Family of International Classifications taken from [32].

Electronic Health Records: A Survey

37

Diagnosis is used to define cause and prognosis of diseases, but by itself it does not predict
service needs, length of hospitalization, or level of care of functional outcomes. Nor can it accurately
provide support for disability. ICF allows incorporating all aspects of a person’s life. The current
ICF creates a more understandable and comprehensive profile of health forming of a person instead
of focusing on a health condition [33]. It is used as a clinical, statistical, research, social policy, and
educational tool. A common misconception about ICF is that it deals with only the disabled people.
However, ICF has some limitations regarding the ability to classify the functional characteristics of
developing children [34].

2.4.7 Diagnosis-Related Groups (DRG)

Diagnosis-Related Groups (DRG) are a patient classification scheme that group related patients
and relate these groups with the costs incurred by the hospital. DRGs divide diagnosis and illness
into 467 categories identified in ICD-9-CM [35]. The 467th group is “ungroupable.” The classifica-
tion is based on a patient’s principal diagnosis, ICD diagnoses, gender, age, sex, treatment proce-
dure, discharge status, and the presence of complications or comorbidities. The goals of developing
DRGs were to reduce healthcare cost, and improve quality of care and efficiency of the hospitals.
DRGs are by far the most important cost control and quality improvement tool developed [36].

It was first created at Yale University with the support from the Health Care Financing Admin-
istration, now known as the Center for Medicine and Medicaid Service (CMS). In 1980, it was first
implemented in a small number of hospitals in New Jersey [37]. It is used to define the reimburse-
ment amount of hospitals from Medicare. Medicare pays hospitals per patient and efficient hospitals
receive better incentives. DRGs help to decide the efficiency of the hospital.

2.4.8 Unified Medical Language System (UMLS)

The Unified Medical Language System (UMLS) is a collection of comprehensive biomedical
concepts and ontologies. It was developed by the U.S. National Library of Medicine (NLM) in 1986.
It provides the development of computer-based systems that can behave as through they understand
the biomedical and health concepts [38]. It is intended to be mainly used by medical informat-
ics professionals. NLM maintains and distributes UMLS knowledge sources (database) and related
software tools for developers to build enhanced electronic information system that can create pro-
cess, retrieve, integrate, and/or aggregate health and biomedical-related information. The knowledge
sources of UMLS are as follows [39]:

• Metathesaurus

– Source Vocabularies

– Concepts

• Relationships, Attributes

– Semantic Network

– Semantic Types (categories)

– Semantic Relationships

• Lexical Resources

– SPECIALIST Lexicon

– Lexical Tools

38

Healthcare Data Analytics

Metathesaurus is a very large, multipurpose, and multilingual vocabulary database. It contains health
and biomedical-related concepts of their various names and the relationships among them. It has
126 vocabularies in 17 languages [27]. It clusters similar terms into a concept. The semantic net-
work provides consistent categorization of concepts defined in Metathesaurus. The network contains
information regarding basic semantic types/categories that may be assigned to concepts and rela-
tionships between semantic types. In the semantic network, the semantic types are nodes and the
relationships are links between them. In the current version of semantic network, there are 135 se-
mantic types and 54 relationships [38]. The SPECIALIST Lexicon provides the lexical information
needed for the SPECIALIST natural language processing tool.

2.4.9 Digital Imaging and Communications in Medicine (DICOM)

The Digital Imaging and Communications in Medicine (DICOM) is a medical imaging standard.
It determines the data exchange protocol, digital image format, and file structure for biomedical
images and related information [40]. DICOM was developed by the American College of Radiology
(ACR) and National Electric Manufacturers Association (NEMA). The first version ACR/NEMA
300 was released in 1985. DICOM is generally used in the following application areas [40]

• Network image management

• Network image interpretation management

• Network print management

• Imaging procedure management

• Offline storage media management

DICOM allows the integration of scanners, servers, workstations, printers, and network hardware
into a Picture Archiving and Communication Systems (PACS). It has been extensively used by the
hospitals and other organizations. It provides a widely accepted foundation for medical imaging
standards. It promotes interoperability between radiology systems.

2.5 Benefits of EHR

EHRs are transformational tools. The scope of paper-based systems is severely limited. We
need EHRs to improve the quality of patient care and increase productivity and efficiency. In terms
of the overall management and costs, EHRs are a better choice. They also help in complying with
government regulations and other legal issues. The benefits of EHRs are described in this section.

2.5.1 Enhanced Revenue

An EHR system can capture the charges and bills for clinical services provided, laboratory tests,
and medications more accurately. Utilization of electronic systems decrease billing errors [41]. They
also provide a better documentation opportunity for these services that can be used to resolve fi-
nancial disputes. Better management of information yield more accurate evaluation and increase
reimbursements. According to experts, due to inaccurate coding systems, 3%–15% of a healthcare
provider’s total revenue is lost [42]. An EHR system can be programmed or configured to generate
alerts for both patients and doctors when a healthcare service is due. This can aid better manage-
ment of collecting revenue. It can be used to garner more revenues by incorporating services like

Electronic Health Records: A Survey

39

telemedicine, e-visits, virtual office visits, etc. It is true that all kinds of services are not possible
over the Internet or telephone network, but not all diseases will require extensive diagnosis and lab-
oratory testing. Diseases commonly treated through telemedicine include acne, allergies, cold and
flu, constipation, diabetes, fever, gout, headache, joint aches and pains, nausea and vomiting, pink
eye, rashes, sinus infection, sore throat, sunburn and urinary tract infections, anxiety and depression,
etc.

2.5.2 Averted Costs

After adopting electronic systems, some costs associated with the previous way of operating a
business are eliminated. The Center for Information Technology leadership suggested that the use
of EHRs will save a total of $44 billion each year [43]. Adopting EHR has the following averted
costs [44].

• Reduced paper and supply cost: To maintain paper-based health records an organization
will require a lot of paper, printing materials, and other supplies. Adopting EHR will reduce
these costs. After adopting EHRs, one organization estimated a reduction of 90% of paper
usage within a few months [45].

• Improved utilization of tests: In electronic systems, test results are better organized. A
healthcare staff no longer needs to carry the reports from one place to another. Identifying
redundancy or unnecessary tests is easier. This can reduce the loss of information and ensure
improved utilization of tests. A study by Wang et al. [41] reports better utilization of radiology
tests after adopting EHRs.

• Reduced transcription costs: An EHR can reduce transcription costs for manual administra-
tive processes [46, 47]. It utilizes structured flow sheets, clinical templates, and point-of-care
documentation. In a typical outpatient setting, physicians generate about 40 lines of tran-
scription per encounter. For a group of three practicing physicians, treating 12,000 patients
annually at the cost of $0.11 for each transcription line results in over $50,000 per year [46].
A study of fourteen solo or small-group primary care practices in twelve U.S. states reports
the median transcription cost saving to be $10,800, where a minimum saving was $8,500
and a maximum was $12,000 for the year 2004–2005 [47]. Other related research work also
describes saving $1,000–$3,000 per physician, per month [48].

• Improved productivity: EHR helps to improve workflows by utilizing resources more ef-
ficiently and reducing redundancies. As a result, the overall productivity of individuals in-
creases.

• Better availability of information and elimination of chart: In EHR, all the charts are in
digital format. It eliminates the need to pull, route, and re-file paper charts [46]. A significant
amount of effort is spent on creating, filing, searching, and transporting paper charts [49]. A
study estimated that the elimination of paper charts can save $5 per chart pull [41]. It is also
comparatively easier to manage digital charts.

• Improved clinician satisfaction: Electronic technology can save time by reducing the pa-
perwork burden, which can create additional time for patient encounters and delivery of care
[3]. A study reports the use of EHR has reduced the physician’s office visit time by 13%
and a nurse’s pre-exam interview time by 1 minute [50]. This can improve satisfaction for
professionals, which can indirectly enhance revenue.

40

Healthcare Data Analytics

2.5.3 Additional Benefits

EHR offers many additional benefits that are discussed in more detail below.

• Improved accuracy of diagnosis and care: EHR provides comprehensive and accurate pa-
tient information to physicians that can help to quickly and systematically identify the correct
problem to treat. EHRs do not just contain the patient information; they have the capability to
perform computation and make suggestions. They can also present comparative results of the
standard measurements. A U.S. national survey of doctors demonstrates the following [51]:

– 94% of the providers report EHR makes records readily available at the point of care.
– 88% report that EHR produces clinical benefits for their practice.
– 75% report that EHR allowed them to deliver better patient care.

The gathered information can guide a physician in the emergency department to take prudent
and safer actions. Such services are unimaginable with paper-based systems. Diagnostic er-
rors are difficult to detect and can be fatal to a patient. A new study suggests that EHR can
help to identify potential diagnostic errors in primary care by using certain types of queries
(triggers) [52].

• Improved quality and convenience of care: EHRs have the potential to improve the qual-
ity of care by embedding options such as Clinical Decision Support (CDS), clinical alerts,
reminders, etc. Research suggests that EHRs are linked to better infection control [53], im-
proved prescribing practices [12], and improved disease management [42] in hospitals. In
such applications, convenience is also an important measure. EHRs greatly reduce the need
for patients to fill out similar (or even sometimes the same) forms at each visit. Patients can
have their e-prescriptions ready even before they leave the facility and can be electronically
sent to a pharmacy. Physicians and staff can process claims insurance immediately. Following
are the results of a study on the effects of e-prescribing reports [54].

– 92% patients were happy with their doctor using e-prescribing.
– 90% reported rarely or only occasionally having prescriptions not ready after going to

the pharmacy.

– 76% reported e-prescribing made obtaining medications easier.
– 63% reported fewer medication errors.

• Improved patient safety: Just like improving the quality of care, clinical decision support
systems (CDSS) and computerized physician order entry (CPOE) have the potential to im-
prove patient safety. Medication errors are common medical mistakes and in the United States
it is responsible for the death of a person every day on average as well as injuring more than a
million annually [55]. Research shows that utilization of CPOE can reduce medication errors
[56, 57]. Medication errors can occur at any stage of the medication administration process
from a physician ordering the drug, followed by the dispensing of the drug by the pharma-
cist, and finally the actual administration of the drug by the nurse. CPOE is a technology
that allows physicians to act on a computerized system that introduces structure and control.
Along with patient information, EHR holds the medication records for a patient. Whenever
a new medication is prescribed, it can check for potential conflicts and allergies related to
the particular medication and alert the physician. The system also can provide the chemi-
cal entities present in the drug and cross-reference allergies, interactions, and other possible
problems related to the specific drug. Introducing technologies such as Barcode Medication
Administration can make the system even more accurate. The Institute of Medicine (IOM)
recommends CPOE and CDS as main information technology mechanisms for increasing
patient safety in the future [58].

Electronic Health Records: A Survey

41

• Improved patient education and participation: In an EHR system, certain features can
provide simplified patient education [42]. EHRs can be used by the provider as a tool to
illustrate procedures and explain a patient’s conditions. It can increase a patient’s participation
by offering follow-up information, self-care instructions, reminders for other follow-up care,
and links to necessary resources. Information technology affects every part of our life. In this
digital era, patients may feel more comfortable with an electronic system.

• Improved coordination of care: EHRs are considered essential elements of care coordi-
nation. The National Quality Forum defines care coordination as the following [59]: “Care
coordination is a function that helps ensure that the patient’s needs and preferences for health
services and information sharing across people, functions, and sites are met over time. Coordi-
nation maximizes the value of services delivered to patients by facilitating beneficial, efficient,
safe and high-quality patient experiences and improved healthcare outcomes.” For a patient
with multiple morbidities, a physician is responsible for providing primary care services and
coordinating the actions of multiple subspecialists [60]. According to a Gallup poll [61], it is
a common scenario for older patients to have multiple doctors: no physician 3%, one physi-
cian 16%, two physicians 26%, three physicians 23%, four physicians 15%, five physicians
6%, and six or more physicians 11%. EHRs allow all clinicians to document services provided
and access up-to-date information about their patient. It streamlines the transition process and
knowledge sharing between different care settings. This facilitates an improved level of com-
munication and coordination [62]. Research suggests that the clinicians having 6+ months use
of EHRs reported better accessing and completeness of information than clinicians without
EHRs. Clinicians having EHRs have also reported to be in agreement on treatment goals with
other involved clinicians [63].

• Improved legal and regulatory compliance: As organizations develop their systems, it is
important to understand and comply with many federal, state, accreditation, and other reg-
ulatory requirements. A health record is the most important legal and business record for a
healthcare organization. The use of an EHR system will provide more security and confiden-
tiality of a patient’s information and thus, comply with regulations like HIPAA, Consumer
Credit Act, etc. Moreover, the Center for Medicare and Medicaid Services (CMS) has fi-
nancial incentive programs for hospitals regarding the meaningful use of health information
technology. To receive the financial reimbursement, professionals have to meet a certain cri-
teria and can get up to $44,000 through Medicare EHR Incentive Program and up to $63,750
through the Medicaid EHR Incentive Program [64]. Adaptation of certified EHR can help
providers get reimbursed.

• Improved ability to conduct research and surveillance: In conjunction with the direct use
of EHR in primary patient care, there is an increasing recognition that secondary use of EHR
data can provide significant insights [65]. Using quantitative analysis of functional values, it
has the potential to identify abnormalities and predict phenotypes. Pakhomov et al. demon-
strated the use of text processing and NLP to identify heart failure patients [66]. EHR data can
be used to predict survival time of patients [67]. Data from different EHRs can be integrated
into a larger database and geo-location specific surveillance is also possible.

• Improved aggregation of data and interoperability: Standards play a crucial role in data
aggregation and interoperability between different systems. EHRs maintain standard proce-
dure and follow defined coding system while collecting data. This accommodates easier ag-
gregation of data and greater interoperability, which offer the following benefits [68].

– Manage increasingly complex clinical care

– Connect multiple locations of care delivery

42

Healthcare Data Analytics

– Support team-based care

– Deliver evidence-based care

– Reduce errors, duplications, and delay

– Support ubiquitous care

– Empower and involve citizens

– Enable the move to the Personal Health Paradigm

– Underpin population health and research

– Protect patient privacy

We need high-quality aggregated data from multiple sources in order to make evidence-
based decisions. The level of achievable interoperability using EHRs is unthinkable from
paper-based systems. The American Medical Association recognizes that enhanced interop-
erability of EHRs will further help to attain the nation’s goal of a high-performing healthcare
system.

• Improved business relationships: A healthcare provider organization equipped a with su-
perior EHR system can be in a better bargaining position with insurers and payers compared
with less equipped ones. The next generation of business professionals will expect and de-
mand a state-of-the-art information healthcare technology system.

• Improved reliability: Data is more reliable in a digital format. Due to the reduction of storage

costs, having multiple copies of data is possible.

2.6 Barriers to Adopting EHR

Despite of having great potential of EHRs in medical practice, the adoption rate is quite slow
and faces a range of various obstacles. Many other developed countries are doing far better than
the United States. Four nations (United Kingdom, the Netherlands, Australia, and New Zealand)
have almost universal use (each
90%) of EHRs among the general practitioners. In contrast, the
United States and Canada have only around 10–30% of the ambulatory care physicians using EHRs
[69]. Health informatics has been a high priority in other developed nations, while until recently,
the degree of involvement and investment by the U.S. government in EHRs has not been significant.
Major barriers to adopting EHRs are discussed below.

∼

• Financial barriers: Although there are studies that demonstrate financial savings after adopt-
ing EHRs, the reality is that the EHR systems are expensive. Several surveys report that the
monetary aspect is one of the major barriers of adopting EHRs [70, 71, 72, 73, 74, 75, 76].
There are mainly two types of financial costs, start-up and ongoing. A 2005 study suggests
that the average initial cost of setting up an EHR is $44,000 (ranging from a minimum of
$14,000 to a maximum of $63,000) and ongoing costs average about $8,500 per provider per
year [47]. Major start-up costs include purchasing hardware and software. In addition, a sig-
nificant amount of money is also required for system administration, control, maintenance,
and support. Long-term costs include monitoring, modifying, and upgrading the system as
well as storage and maintenance of health records. Besides, after the substantial amount of
investment, physicians are worried that it could take up to several years for the return on the
investment.

Electronic Health Records: A Survey

43

An EHR is not the only electronic system that exists in any healthcare provider like practice
management. There might be other old systems that also need integration into the new system.
It is important that an EHR system is integrated into other systems, and this integration can
sometimes be very expensive. Surveys show that due to the high financial investment required,
EHR adaptation was far higher in large physician practices and hospitals [77].

• Physician’s resistance: To adopt EHRs, physicians have to be shown that new technology
can return financial profits, saves time, and is good for their patients’ well-being. Although
research-based evidence is available, it is difficult to provide concrete proof of those benefits.
As given in a report by Kemper et al. [76], 58% of physicians are without any doubt that EHR
can improve patient care or clinical outcomes. Finally, adopting EHRs in a medical practice
will significantly change the work processes that physicians have developed for years.

Besides, physicians and staffs might have insufficient technical knowledge to deal with EHRs,
which leads them to think EHR systems are overly complex. Many physicians complain about
poor follow-up services regarding technical issues and a general lack of training and support
from EHR system vendors [72]. A study reports that two-thirds of physicians expressed in-
adequate technical support as a barrier to adopting EHRs [75]. Some physicians are also
concerned about the limitation of EHR capabilities. Under certain circumstances or as time
passes, the system may no longer be useful [71, 74]. Besides, all physicians do not perform
the same operations. EHR systems have to be customizable to best serve each purpose. Sur-
veys suggest that one of the reasons for not adopting EHRs is that the physicians cannot find
a system that meets their special requirements [71, 72, 73, 75, 78, 76]. However, an increased
effort and support from vendors may play a role in motivating physicians towards adopting
EHRs.

• Loss of productivity: Adoption of an EHR system is a time-consuming process. It requires
a notable amount of time to select, purchase, and implement the system into clinical practice.
During this period physicians have to work at a reduced capacity. Also, a significant amount
of time has to be spent on learning the system. The improvement will depend on the quality
of training, aptitude, etc. The fluent workflow will be disrupted during the transition period,
and there will be a temporary loss of productivity [79].

• Usability issues: EHR software needs to be user-friendly. The contents of the software must
be well-organized so that a user can perform a necessary operation with a minimal number
of mouse clicks or keyboard actions. The interface of software workflow has to be intuitive
enough. In terms of usability, a comprehensive EHR system may be more complex than ex-
pected. It has to support all the functionalities in a provider’s setting. There might be a number
of modules and submodules, so the user might get lost and not find what he is looking for. This
has the potential to hamper clinical productivity as well as to increase user fatigue, error rate,
and user dissatisfaction. Usability and intuitiveness in the system do not necessarily correlate
to the amount of money spent. The Healthcare Information and Management Systems Soci-
ety (HIMSS) has an EHR usability task force. A 2009 survey by the task force reported 1,237
usability problems, and the severity of 80% of them was rated “High” or “Medium” [80].
Apart from the workflow usability issue, other related issues are configuration, integration,
presentation, data integrity, and performance. The task force defined the following principles
to follow for effective usability [81]: simplicity, naturalness, consistency, minimizing cogni-
tive load, efficient interactions, forgiveness and feedback, effective use of language, effective
information presentation, and preservation of context.

• Lack of standards: Lack of uniform and consistent standards hinders the EHR adoption.
Standards play an integral role in enabling interoperability. CMS reimbursement for mean-
ingful use requires EHR systems to demonstrate the ability to exchange information. Many

44

Healthcare Data Analytics

of the currently used systems have utility only for certain specific circumstances. Different
vendors have developed systems in different programming languages and database systems.
They do not have any defined best practice or design patterns. This makes the data exchange
difficult or impossible between the systems [73, 74, 78]. This lack of standardization limits
the proliferation of EHRs [82]. While large hospital systems have moved to EHRs, many
others are skeptical about the available systems. They fear that the EHR software they buy
now might not work with standards adopted by the healthcare industry or mandated by the
government later on.

• Privacy and security concerns: Health records contain personal, diagnostics, procedures,
and other healthcare related sensitive information. Due to the immense importance of this
information, an EHR system may be subjected to attack. Some of the medical diagnoses are
considered socially stigmatized, like sexually transmitted disease. Some information relates
to direct life threats, like allergies. Employers as well as insurance companies may be inter-
ested to know more about a patient to make unethical decisions whether to cover a patient
and/or his specific diagnosis. It can also influence some of the hiring decisions. EHRs contain
information like social security numbers, credit card numbers, telephone numbers, home ad-
dresses, etc., which makes EHRs attractive target for attackers and hackers. A patient might
even be motivated to alter his or her medical records to get worker’s compensation or to obtain
access to narcotics. Therefore, it is important that the privacy and security of EHRs are well
maintained. The most used certification for privacy and security is given by the Certification
Commission for Healthcare Information Technology (CCHIT). The CCHIT website claims
that by mid-2009, 75% of EHR products in the marketplace were certified [83]. In addition
to that, the Health Information Technology for Economic and Clinical Health (HITECH) Act
introduced a new certification process sponsored by the Office of the National Coordina-
tion for Health Information Technology (ONC) in 2009. In January 2010, the ONC released
the interim final rule that provides an initial set of standards, implementation specifications,
and certification criteria of EHR technology. Its requirement includes database encryption,
encryption of transmitted data, authentication, data integrity, audit logs, automatic log off,
emergency access, access control, and account of HIPPA release of information [84]. Physi-
cians doubt the level of security of patients’ information and records. According to Simon
et al. [74], physicians are more concerned about this issue than patients. The inappropri-
ate disclosure of information might lead to legal consequences. Testing the security of EHR
products, a group of researchers showed that they were able to exploit a range of common
code-level and design-level vulnerabilities of a proprietary and an open source EHR [85].
These common vulnerabilities could not be detected by 2011 security certification test scripts
used by CCHIT. EHRs pose new challenges and threats to the privacy and security of patient
data. This is a considerable barrier to EHRs proliferation. However, this risk can be mitigated
by proper technology, and maintaining certified standards with the software and hardware
components.

• Legal aspects: Electronic records of medical information should be treated as private and
confidential. Various legal and ethical questions obstruct adoption and use of EHRs. The le-
gal system that relies on the paper-era regulations does not offer proper guidance regarding
the transition to EHRs. EHRs may increase the physicians’ legal responsibility and account-
ability [86]. With computer-based sophisticated auditing, it is easy to track what individuals
have done. The documentation is comprehensive and detailed in EHRs. It can both defend and
expose physicians regarding malpractice. According to a Health Affairs article, malpractice
costs around $55 billion in the United States, which is 2.4% of total healthcare spending [87].
A 2010 research reveals that it was unable to determine whether the use of EHR increases or
decreases malpractice liability overall [86]. HIPAA’s privacy standards also present reason-
able barriers to EHR adaptation.

Electronic Health Records: A Survey

45

2.7 Challenges of Using EHR Data

The primary purpose of EHR data is to support healthcare-related functionalities. As a vast
amount of data is being collected every day, the secondary use of EHR data is gaining increased
attention in research community to discover new knowledge. The main areas of use are clinical
and transitional research, public health, and quality measurement and improvement. Using the EHR
data, we can conduct both patient-oriented and public health research. EHR data can be used for
the early detection of epidemics and spread of diseases, environmental hazards, promotes healthy
behaviors, and policy development. The integration of genetic data with EHRs can open even wider
horizons. But the data does not automatically provide us the knowledge. The quality and accuracy
of the data is an issue to be taken care of. Beyley et al. [88] presents an excellent survey of the
challenges posed by the data quality.

• Incompleteness: Data incompleteness or missingness is a widespread problem while using
EHR data for secondary purpose [88, 89, 90]. Missing data can limit the outcomes to be
studied, the number of explanatory factors to be considered, and even the size of population
included [88]. Incompleteness can occur due to a lack of collection or lack of documentation
[91]. Hersh [92] reports the following reasons for inaccurate reporting by professionals.

– Unaware of legal requirements

– Lack of knowledge of which diseases are reportable
– Do not understand how to report

– Assumption that someone else will report

– Intentional failure for privacy reasons

A pancreatic malignancies study using ICD-9-CM code at the Columbia University Medical
Center found that 48% of the patients had corresponding diagnoses or disease documenta-
tion missing in their pathology reports [93]. Authors also report a significant amount of key
variables missing (see Table 2.1).

Patients’ irregularity of communicating with the health system can also produce incomplete-
ness. Based on the application in hand, type of data and proportion of data that is missing,
certain strategies can be followed to reduce the missingness of data [91].

TABLE 2.1: Percentage of Incompleteness of Variables in a Pancreatic Malignancies Study

Variables
Necrosis
Number of Mitoses
Lymph Node Metastasis
Perineural/Lymphovascula Invasion
Differentiation
Size
Chronic Pancreatitis
Smoking—Alcohol
History of Other Cancer
Family History of Cancer
Tumor Markers

Source: Taken from Botsis et al. [93].

Endocrine
20%
21%
28%
15%
38%
6%
14%
27%–29%
35%
39%
46%

46

Healthcare Data Analytics

• Erroneous Data: EHR data can be erroneous as well. Data is collected from different service
areas, conditions, and geographic locations. Data is collected by busy practitioners and staff.
Therefore, the data can be erroneous due to human errors. Faulty equipment can also produce
erroneous data. Validation techniques should be used to both identify and correct erroneous
data. Both internal and external validation measures can be applied. Internal validation is a
way to check the believability of the data, e.g., unrealistic blood pressure, BMI values, etc.
Dates can be used to check whether the result generated before a test has taken place. External
validation includes comparing the data with other patients or historical values.

• Uninterpretable Data: The captured EHR data might be uninterpretable to a certain extent.
It is closely related with data incompleteness. It may occur when some part of the data is cap-
tured but the rest is missing. For example, if a specific quantitative or qualitative measurement
unit is not provided with the result value, it will be difficult to interpret.

• Inconsistency: Data inconsistency can heavily affect the analysis or result. Data collection
technologies, coding rules, and standards may change over time and across institutions, which
may contribute to inconsistency. For multi-institutional studies this issue might be common,
especially because different healthcare centers use different vendors for providing apparatus,
softwares, and other technologies [88]. A study in Massachusetts of 3.7 million patients found
that 31% of patients have visited two or more hospitals in the course of five years [94].

• Unstructured Text: In spite of having many defined structures for collecting the data, a large
portion of the EHR data contain unstructured text. These data are present in the form of
documentation and explanation. It is easy to understand them for humans, but in terms of
automatic computational methods, detecting the right information is difficult. Sophisticated
data extraction techniques like Natural Language Processing (NLP) are being used to identify
information from text notes [95].

• Selection Bias: In any hospital, the patient group will mostly be a random collection. It varies
depending on the nature of practice, care unit, and the geographical location of the institution.
It will not contain the diversity of demography. This is an important challenge to overcome.
Therefore, EHR data mining findings will not be generalizable. This problem must be ad-
dressed while working with the secondary use of data.

• Interoperability: Lack of EHR interoperability is a major impediment towards improved
healthcare, innovation, and lowering costs. There are various reasons behind it. EHR software
from commercial vendors are proprietary and closed systems. Most software were not built
to support communication with a third party and developing new interfaces for that purpose
might be a costly undertaking. Absence of standard also contributes to the problem. Many
patients are not lenient towards sharing their information. Besides EHR systems must comply
with the HIPAA Act [11] to ensure the security and privacy of the data.

In a recent JAMIA (Journal of the American Medical Informatics Association) article, the
authors have specified 11 specific areas that present barriers to interoperability of C-CDA
documents by inspecting 91 C-CDA documents from 21 technologies [96]. In June 2014, the
office of the National Coordinator for Health Information Technology (ONC) unveiled a plan
for robust healthcare information sharing and aggregation and interoperability increase by
2024 [97]. Its three-year agenda includes “Send, Receive, Find, and Use Health Information to
Improve Health Care Quality.” Its six-year agenda states “Use Information to Improve Health
Care Quality and Lower Cost,” and finally, its 10-year agenda proposes to achieve a “Learning
Health System.” The mentioned building blocks for attaining the goals are the following:

– Core technical standards and functions
– Certification to support adoption and optimization of health IT products and services

Electronic Health Records: A Survey

47

– Privacy and security protections for health information

– Supportive business, clinical, cultural, and regulatory environments

– Rules of engagement and governance

2.8 Phenotyping Algorithms

Phenotyping algorithms are combinations of multiple types of data and their logical relations to
accurately identify cases (disease samples) and controls (non-disease samples) from EHR as illus-
trated in Figure 2.3 [98]. Based on the structure, EHR data can be broadly divided into two parts,
structured and unstructured data. Structured data exists in a name–value pair while unstructured
data contains narrative and semi-narrative texts regarding descriptions, explanation, comments, etc.
Structured data include billing data, lab values, vital signs, and medication information. Billing and
diagnosis-related data are collected using various coding systems like ICD, CPT, and SNOMED-
CT. These codes are important parts of the phenotyping process. ICD codes generally have high
specificity but low sensitivity [99]. Table 2.2 lists different characteristics of EHR data.

The primary purpose of EHR data is to support healthcare and administrative services. Infor-
mation is produced as a byproduct of routine clinical services. They are not a suitable format for
performing research tasks. They often require further processing to be used for phenotyping al-
gorithms. Within existing EHR systems, querying for a particular diagnosis or lab test across all
patients can be a not-trivial task. An EHR can quickly pull the information related to a patient’s
current medications, and easily find any test results. But combining different data with a temporal
relationship might require manual processing of data. From clinical operational settings, data are
often extracted and reformatted to make them more convenient and suitable for doing research, typ-
ically storing them in relational databases. Researchers have created a number of Enterprise Data
Warehouses (EDWs) for EHR data. Examples include Informatics for Integrating Biology and the
Bedside (i2b2) [100], the Utah Population Database [101], Vanderbilt’s Synthetic Derivative [102],
etc. Commercial EHR vendors are also developing research repositories. For example, EPIC users
can add the “Clarity” module to their system, which will convert the EHR data into SQL-based
database for research purposes.

To build a phenotype algorithm, first we need to select the phenotype of interest, followed by the
identification of key clinical elements that define the phenotype. It may contain billing codes, lab-
oratory and test results, radiology reports, medication history, and NLP-extracted information. The
gathered information may be combined with a machine learning method. For example, in [103], the
authors have applied Support Vector Machine (SVM) to a both naive and well-defined collection of
EHR features to identify rheumatoid arthritis cases. A medication record can be used to increase the
accuracy of case and control identification of phenotyping algorithms. Patients who are believed to
be controls must be having a different medication profile. They may not even have any medications
prescribed to them at all. Sufficient dosage of a particular medication serves the confirmation that a
person is having the disease of interest. For example, a patient treated with either oral or injectable
hypoglycemic agents will be having diabetes. These medications are highly sensitive and specific
for treating diabetes.

Studies have shown that CPT codes can accurately predict an occurrence of a given procedure
[104]. The standard terminology codes for lab tests are LOINC. On the other hand, clinical notes
are in free-text format. To be used for phenotyping algorithms, it has to undergo subsequent text
processing. Certain procedures and test results may also exist in a combination of structured and
unstructured form. For example, an electrocardiogram report typically contains structured interval

48

Healthcare Data Analytics

(1)
PMRP

(2)Has Cataract
Surgery?

No

Yes

(3)Has Cataract
DX?

Yes

(8)How many DX?

2+Dx

1 Dx

Case

Yes

(9)NLP or ICR
Cataract found?

Yes

(10)Age >= 50 at
event

No

Yes

Yes

No

Exclude

No

No

(4)Any exclusion
Dx found?

No

(5)NLP or ICR
Cataract Found?

No

Exam < 5 Yrs?

Yes

No

(7)Age >= 50 at
event

Yes

Control

FIGURE 2.3: Flowchart for cataracts phenotyping algorithm taken from [98].

durations, heart rates, and overall categorization, along with a narrative text of cardiologist’s inter-
pretation of the result [105].

Recently, researchers have been linking EHR data with biological databanks (biobanks). The
most popular biobanks are the collection of DNA samples. Hospitals and clinics can collect DNA

Electronic Health Records: A Survey

49

TABLE 2.2: Characteristics of Different EHR Data

Availability
Recall

ICD
High
Medium

CPT
High
Poor

Lab
High
Medium

Precision

Medium

High

High

Structured
Easy to work
with, high
precision

Mostly
High data
validity

Medication
Medium
Inpatient: High
Outpatient:
Variable
Inpatient: High
Outpatient:
Variable
Structured
High data validity

Clinical notes
Medium
Medium

Medium/High

Structured
More details
about the
doctors’
thoughts

Missing data

Data nor-
malization
and ranges

Prescribed not
necessarily taken

Difficult to
process

Format
Pros

Cons

Structured
Easy to work
with, good
approximation
of disease
status
Disease code
often used for
screening,
therefore
disease might
not be there

Source: Taken from Denny [106].

samples from a patient’s blood sample that is used in routine tests. The Personalized Medicine
Research Population (PMRP) project in Marshfield Clinic has a biobank of 20,000 individuals [107].
Similar DNA biobanks exist at eMERGE Network sites, Northwestern University, Geisinger Health
System, Mount Sinai School of Medicine, and at other places. The eMERGE network is funded
and organized by the National Human Genome Research Institute (NHGRI) and until today it has
created and validated twenty-one EHR-derived phenotyping algorithms (see Table 2.3). Its mission
is to develop, disseminate, and apply methods to combine DNA biorepositories and EHR systems
for large scale and high throughput genetic research [108]. But the phenotype information extracted
from EHRs may be challenging. Validation of phenotypes is important before integration of EHRs
into genetic studies. By validating EHR-derived phenotypes from eMERGE network, Newton et al.
report the following points [109]:

• Multisite validation improves phenotype algorithm accuracy

• Targets for validation should be carefully considered and defined

• Specifying time frames for review of variables eases validation time and improves accuracy

• Using repeated measures requires defining the relevant time period and specifying the most

meaningful value to be studied

• Patient movement in and out of the health plan (transience) can result in incomplete or frag-

mented data

• The review scope should be defined carefully

• Particular care is required in combining EMR and research data

• Medication data can be assessed using claims, medications dispensed, or medications pre-

scribed

• Algorithm development and validation will work best as an iterative process

• Validation by content experts or structured chart review can provide accurate results

50

Healthcare Data Analytics

TABLE 2.3: Phenotyping Algorithms Developed by eMERGE Network

Phenotype

Atrial Fibrillation —
Demonstration Project
Cardiac Conduction(QRS)

Cataracts

Clopidogrel Poor
Metabolizers

Crohn’s Disease —
Demonstration Project
Dementia
Diabetic Retionapathy

Drug Induced Liver Injury

Height

High-Density Lipoproteins
(HDL)

Hypothyroidism

Lipids

Multiple Sclerosis —
Demonstration Project
Peripheral Arterial Disease

Red Blood Cell Indices

Rheumatoid Arthritis —
Demonstration Project
Severe Early Childhood
Obesity
Type 2 Diabetes —
Demonstration Project

Type 2 Diabetes Mellitus

Warfarin dose/response

White Blood Cell Indices

Source: Taken from [110].

EHR data used to characterize
phenotype
CPT Codes, ICD 9 Codes, Natural
Language Processing
CPT Codes, ICD 9 Codes,
Laboratories, Medications, Natural
Language Processing
CPT Codes, ICD 9 Codes,
Medications, Natural Language
Processing
CPT Codes, ICD 9 Codes,
Laboratories, Medications, Natural
Language Processing
ICD 9 Codes, Medications, Natural
Language Processing
ICD 9 Codes, Medications
CPT Codes, ICD 9 Codes,
Laboratories, Medications, Natural
Language Processing
ICD 9 Codes, Laboratories,
Medications, Natural Language
Processing
ICD 9 Codes, Laboratories,
Medications
ICD 9 Codes, Laboratories,
Medications, Natural Language
Processing
CPT Codes, ICD 9 Codes,
Laboratories, Medications, Natural
Language Processing
ICD 9 Codes, Laboratories,
Medications
ICD 9 Codes, Medications, Natural
Language Processing
CPT Codes, ICD 9 Codes,
Laboratories, Medications, Natural
Language Processing
CPT Codes, ICD 9 Codes,
Laboratories, Medications, Natural
Language Processing
ICD 9 Codes, Medications, Natural
Language Processing
ICD 9 Codes, Medications, Natural
Language Processing, Vital Signs
ICD 9 Codes, Laboratories,
Medications, Natural Language
Processing
ICD 9 Codes, Laboratories,
Medications
Laboratories, Natural Language
Processing
CPT Codes, ICD 9 Codes,
Laboratories, Medications

Institution

Vanderbilt University

Vanderbilt University

Marshfield Clinic Research Foundation

Denny’s Group at Vanderbilt, VESPA
— Vanderbilt Electronic Systems for
Pharmacogenomic Assessment
Vanderbilt University

Group Health Cooperative
Marshfield Clinic Research Foundation

Columbia University

Northwestern University

Marshfield Clinic Research Foundation

Vanderbilt University, Group Health
Cooperative, Northwestern University

Northwestern University

Vanderbilt University

Mayo Clinic

Mayo Clinic

Vanderbilt University

Cincinnati Children’s Hospital Medical
Center
Vanderbilt University

Northwestern University

Vanderbilt University

Group Health Cooperative

Electronic Health Records: A Survey

51

Before the use of a phenotyping algorithm, data has to be normalized to standard representation.
Natural Language Processing (NLP) based tools have gained much popularity to extract structured
information from free text. Several studies have shown that coded data are not sufficient or accurate
to identify disease cohorts [111, 112]. Information from narrative text complements the structured
data. There are studies that report NLP-processed notes provide more valuable data sources. For
example, Penz et al. reports ICD-9 and CPT codes identified less than 11% cases in detecting ad-
verse events related to central venous catheters, while NLP methods achieved a specificity of 0.80
and sensitivity of 0.72 [113]. Widely used general-purpose NLP tools include MedLEE (Medical
Language Extraction and Encoding System) [114], cTAKES (clinical Text Analysis and Knowledge
Extraction System) [115], MetaMap [116], and KnowledgeMap [117]. All of them have been suc-
cessfully applied to phenotyping using EHR data. Task-specific NLP methods are available that aim
to extract specific concepts from clinical text.

The DNA sequence of a person can be huge in size (ranging from hundreds of gigabytes to
terabytes) in raw format that exceeds the capability for using the current EHR systems. Storing,
managing, and transferring a repository of such a large volume of data is difficult. Efficient data
compression techniques can be applied to solve this problem. Genome Wide Association Study
(GWAS) became the mainstay of genetic analysis over the last decade. In general, GWAS investi-
gates around 500,000 genetic variants (Single Nucleotide Polymorphisms) or more to see the asso-
ciation of variations with observable traits. It compares the SNPs of cases versus controls to find
meaningful knowledge. Besides traits, we can also identify SNPs that determine a particular drug
response. One individual might react adversely to a particular drug while others might not. The ge-
netic profile of an individual can be used for personalized medicine. One big advantage of genetic
data is that the SNPs are the same for that individual and do not change based on a given/suspected
disease. The same set of data can be used for different phenotype investigations as well. Researchers
are working to integrate genetic information for enhanced clinical decision support. For example,
researchers in Vanderbilt University are working on implementing Pharmacogenomic Resource for
Enhanced Decisions in Care and Treatment (PREDICT) [118]. St. Jude Children’s Research Hospi-
tal also has a multiplexed genotyping platform for providing decision support [119].

2.9 Conclusions

Electronic health records are the obvious and inevitable future of patient care in hospitals and
medical practices. This chapter discusses several aspects of the EHRs. EHR systems are gaining
nationwide popularity in the United States recently due to “Meaningful use legislation and reim-
bursement [120]. It is being widely installed in hospitals, academic medical centers,” and outpatient
clinics throughout the nation. Besides healthcare benefits like improved patient care, safety and
reduced costs, it creates great opportunity for clinical and translational research. Widespread adop-
tion of EHRs can foster the improvement of quality in healthcare services, safety and efficiency,
and most importantly, public health. Having great potential for benefits, successful deployment of
EHRs has several challenges to overcome. There are notable limitations of the use of EHR data
in research purposes. In the era of technology, the necessary laws lag far behind. While other de-
veloped countries have showed widespread adoption, in the United States, the overall adoption is
considerably low. Bigger Government initiatives and enhanced standardization today can lead to a
brighter healthcare tomorrow.

Acknowledgments

This work was supported in part by the National Science Foundation Grant IIS-1231742.

52

Healthcare Data Analytics

Bibliography

[1] Eric Jamoom et al. Physician Adoption of Electronic Health Record Systems: United States,
2011, volume 98. US Department of Health and Human Services, Centers for Disease Control
and Prevention, National Center for Health Statistics, 2012.

[2] Mark A. Musen and Jan Hendrik Bemmel. Handbook of Medical Informatics. MIEUR, 1999.

[3] Richard S. Dick and Elaine B. Steen. The Computer-Based Patient Record: An Essential

Technology for Health Care. National Academics Press, 1991.

[4] http://yosemite.epa.gov/opa/admpress.nsf/8822edaadaba0243852572a000656841/

66a4a31db7c1ae178525703d0067d18b!OpenDocument, September, 2003.

[5] George W. Bush. Transforming health care: The president’s health information technology

plan. Washington, DC, 2004.

[6] http://www.cnbc.com/id/28559492/Text_of_Obama_Speech_on_the_Econom.

[7] http://www.cdc.gov/osels/lspppo/healthcare_news.html.

[8] Hagop S. Mekhjian, Rajee R. Kumar, Lynn Kuehn, Thomas D. Bentley, Phyllis Teater, An-
drew Thomas, Beth Payne, and Asif Ahmad. Immediate benefits realized following imple-
mentation of physician order entry at an academic medical center. Journal of the American
Medical Informatics Association, 9(5):529–539, 2002.

[9] Electronic health records overview. National Institutes of Health National Center for Re-
search Resources, MITRE Center for Enterprise Modernization, McLean, Virginia, 2006.

[10] Robert H. Dolin, Liora Alschuler, Sandy Boyer, Calvin Beebe, Fred M. Behlen, Paul V. Biron,
and Amnon Shabo Shvo. Hl7 clinical document architecture, release 2. Journal of the Amer-
ican Medical Informatics Association, 13(1):30–39, 2006.

[11] Centers for Medicare & Medicaid Services et al. The health insurance portability and ac-

countability act of 1996 (HIPAA). Online at http://www. cms. hhs. gov/hipaa, 1996.

[12] http://www.who.int/classifications/icd/en/.

[13] Heinz Katschnig. Are psychiatrists an endangered species? Observations on internal and

external challenges to the profession. World Psychiatry, 9(1):21–28, 2010.

[14] http://www.who.int/classifications/icd/factsheet/en/index.html.

[15] http://www.ncvhs.hhs.gov/031105a1.htm.

[16] http://wonder.cdc.gov/wonder/help/mcd.html.

[17] Hude Quan, Bing Li, L. Duncan Saunders, Gerry A. Parsons, Carolyn I. Nilsson, Arif Alibhai,
and William A. Ghali. Assessing validity of ICD-9-CM and ICD-10 administrative data in
recording clinical conditions in a unique dually coded database. Health Services Research,
43(4):1424–1441, 2008.

[18] http://www.who.int/classifications/icd/revision/timeline/en/.

[19] World Health Organization et al. ICD-11 beta draft, 2013.

Electronic Health Records: A Survey

53

[20] http://www.who.int/classifications/icd/revision/

icd11betaknownconcerns.pdf.

[21] Ronald Cornet and Nicolette de Keizer. Forty years of SNOMED: A literature review. BMC

Medical Informatics and Decision Making, 8(Suppl 1):S2, 2008.

[22] Michael Q. Stearns, Colin Price, Kent A. Spackman, and Amy Y. Wang. SNOMED clinical
terms: Overview of the development process and project status. In Proceedings of the AMIA
Symposium, page 662. American Medical Informatics Association, 2001.

[23] http://www.ihtsdo.org/snomed-ct.

[24] Karen Kostick et al. SNOMED CT integral part of quality EHR documentation. Journal of
AHIMA/American Health Information Management Association, 83(10):72–75, 2012.

[25] Clem McDonald, S.M. Huff, J. Suico, and Kathy Mercer. Logical observation identifiers
names and codes (LOINC R) users’ guide. Indianapolis: Regenstrief Institute, 2004.

[26] Alex A.T. Bui and Ricky K. Taira (Eds.). Medical Imaging Informatics. Springer New York,

2010.

[27] Casey C. Bennett. Utilizing RxNorm to support practical computing applications: Captur-
ing medication history in live electronic health records. Journal of Biomedical Informatics,
45(4):634–641, 2012.

[28] Simon Liu, Wei Ma, Robin Moore, Vikraman Ganesan, and Stuart Nelson. RxNorm: Pre-
scription for electronic drug information exchange. IT Professional, 7(5):17–23, 2005.

[29] https://www.nlm.nih.gov/research/umls/rxnorm/history.html.

[30] National Center for Health Statistics (US), World Health Organization, et al. Classification

of Diseases, Functioning, and Disability. US Centers for Disease Control and Prevention.

[31] Alan M. Jette. Toward a common language for function, disability, and health. Journal of

American Physical Therapy Association, 86(5):726–734, 2006.

[32] World Health Organization and others. Towards a common language for functioning, disabil-

ity and health: ICF. World Health Organization, 2002.

[33] Helena Hemmingsson and Hans Jonsson. An occupational perspective on the concept of par-
ticipation in the international classification of functioning, disability and healthsome critical
remarks. The American Journal of Occupational Therapy, 59(5):569–576, 2005.

[34] Rune J. Simeonsson, Anita A. Scarborough, and Kathleen M. Hebbeler. ICF and ICD codes
provide a standard language of disability in young children. Journal of Clinical Epidemiol-
ogy, 59(4):365–373, 2006.

[35] http://www.ihtsdo.org/snomed-ct/snomed-ct0/.

[36] Norbert Goldfield. The evolution of diagnosis-related groups (DRGS): From its beginnings in
case-mix and resource use theory, to its implementation for payment and now for its current
utilization for quality within and outside the hospital. Quality Management in Healthcare,
19(1):3–16, 2010.

[37] W.C. Hsiao, H.M. Sapolsky, D.L. Dunn, and S.L. Weiner. Lessons of the New Jersey DRG

payment system. Health Affairs, 5(2):32–45, 1986.

54

Healthcare Data Analytics

[38] Catherine Selden and Betsy L. Humphreys. Unified Medical Language System (UMLS):
January 1986 Through December 1996: 280 Selected Citations, volume 96. US Dept. of
Health and Human Services, Public Health Service, 1997.

[39] Donald A.B. Lindberg and Betsy L. Humphreys. Concepts, issues, and standards. current
status of the nlm’s umls project: The UMLS knowledge sources: Tools for building better user
interfaces. In Proceedings of the Annual Symposium on Computer Application in Medical
Care, page 121. American Medical Informatics Association, 1990.

[40] W. Dean Bidgood, Steven C. Horii, Fred W. Prior, and Donald E. Van Syckle. Understanding
and using DICOM, the data interchange standard for biomedical imaging. Journal of the
American Medical Informatics Association, 4(3):199–212, 1997.

[41] Samuel J. Wang, Blackford Middleton, Lisa A. Prosser, Christiana G. Bardon, Cynthia D.
Spurr, Patricia J. Carchidi, Anne F. Kittler, Robert C. Goldszer, David G. Fairchild, Andrew J.
Sussman et al. A cost-benefit analysis of electronic medical records in primary care. The
American Journal of Medicine, 114(5):397–403, 2003.

[42] Tricia L. Erstad. Analyzing computer-based patient records: A review of literature. Journal

of Healthcare Information Management, 17(4):51–57, 2003.

[43] Douglas Johnston, Eric Pan, and J. Walker. The value of CPOE in ambulatory settings.

Journal of Healthcare Information Management, 18(1):5–8, 2004.

[44] Nir Menachemi and Robert G. Brooks. Reviewing the benefits and costs of electronic health
records and associated patient safety technologies. Journal of Medical Systems, 30(3):159–
168, 2006.

[45] T. Ewing and D. Cusick. Knowing what to measure. Healthcare Financial Management:

Journal of the Healthcare Financial Management Association, 58(6):60–63, 2004.

[46] Abha Agrawal. Return on investment analysis for a computer-based patient record in the
outpatient clinic setting. Journal of the Association for Academic Minority Physicians: The
Official Publication of the Association for Academic Minority Physicians, 13(3):61–65, 2002.

[47] Robert H. Miller, Christopher West, Tiffany Martin Brown, Ida Sim, and Chris Ganchoff.
The value of electronic health records in solo or small group practices. Health Affairs,
24(5):1127–1137, 2005.

[48] Jeff Mildon and Todd Cohen. Drivers in the electronic medical records market. Health

Management Technology, 22(5):14, 2001.

[49] Karl F. Schmitt and David A Wofford. Financial analysis projects clear returns from elec-
tronic medical records. Healthcare Financial Management: Journal of the Healthcare Fi-
nancial Management Association, 56(1):52–57, 2002.

[50] K. Renner. Electronic medical records in the outpatient setting (Part 1). Medical Group

Management Journal/MGMA, 43(3):52–54, 1995.

[51] E. Jamoom. National perceptions of EHR adoption: Barriers, impacts, and federal policies.

National Conference on Health Statistics, 2012.

[52] Hardeep Singh, Traber Davis Giardina, Samuel N. Forjuoh, Michael D. Reis, Steven Kos-
mach, Myrna M. Khan, and Eric J. Thomas. Electronic health record-based surveillance of
diagnostic errors in primary care. BMJ Quality & Safety, 21(2):93–100, 2012.

Electronic Health Records: A Survey

55

[53] J. Michael Fitzmaurice, Karen Adams, and John M. Eisenberg. Three decades of research on
computer applications in health care medical informatics support at the agency for healthcare
research and quality. Journal of the American Medical Informatics Association, 9(2):144–
160, 2002.

[54] R. Lamar Duffy, Shih Shen Angela Yiu, Ehab Molokhia, Robert Walker, and R. Allen
Perkins. Effects of electronic prescribing on the clinical practice of a family medicine resi-
dency. Family Medicine, 42(5):358–63, 2010.

[55] Linda T. Kohn, Janet M. Corrigan, Molla S. Donaldson et al. To Err is Human: Building a

Safer Health System, volume 627. National Academies Press, 2000.

[56] David W. Bates, Lucian L. Leape, David J. Cullen, Nan Laird, Laura A. Petersen, Jonathan M.
Teich, Elizabeth Burdick, Mairead Hickey, Sharon Kleefield, Brian Shea et al. Effect of com-
puterized physician order entry and a team intervention on prevention of serious medication
errors. JAMA, 280(15):1311–1316, 1998.

[57] Ross Koppel, Joshua P. Metlay, Abigail Cohen, Brian Abaluck, A. Russell Localio,
Stephen E. Kimmel, and Brian L. Strom. Role of computerized physician order entry systems
in facilitating medication errors. JAMA, 293(10):1197–1203, 2005.

[58] Institute of Medicine (US). Committee on Quality of Health Care in America. Crossing the
Quality Chasm: A New Health System for the 21st Century. National Academies Press, 2001.

[59] Elizabeth E Stewart, Paul A. Nutting, Benjamin F. Crabtree, Kurt C. Stange, William L.
Miller, and Carlos Roberto Ja´en. Implementing the patient-centered medical home: observa-
tion and description of the national demonstration project. The Annals of Family Medicine,
8(Suppl 1):S21–S32, 2010.

[60] Christopher J. Stille, Anthony Jerant, Douglas Bell, David Meltzer, and Joann G. Elmore.
Coordinating care across diseases, settings, and clinicians: A key role for the generalist in
practice. Annals of Internal Medicine, 142(8):700–708, 2005.

[61] Gallup serious chronic illness survey 2002.

http://poll.gallup.com/content/

default.aspx?ci=6325&pg=1.

[62] Lynda C. Burton, Gerard F. Anderson, and Irvin W. Kues. Using electronic health records to

help coordinate care. Milbank Quarterly, 82(3):457–481, 2004.

[63] Ilana Graetz, Mary Reed, Thomas Rundall, Jim Bellows, Richard Brand, and John Hsu. Care
In AMIA Annual Sym-
coordination and electronic health records: Connecting clinicians.
posium Proceedings, volume 2009, page 208. American Medical Informatics Association,
2009.

[64] Medicare and medicaid EHR incentive program basics.

http://www.cms.gov/
Regulations-and-Guidance/Legislation/EHRIncentivePrograms/Basics.html.

[65] Charles Safran, Meryl Bloomrosen, W. Edward Hammond, Steven Labkoff, Suzanne Markel-
Fox, Paul C. Tang, and Don E. Detmer. Toward a national framework for the secondary use
of health data: An American Medical Informatics Association white paper. Journal of the
American Medical Informatics Association, 14(1):1–9, 2007.

[66] Serguei Pakhomov, James Buntrock, and Patrick Duffy. High throughput modularized NLP
system for clinical text. In Proceedings of the ACL 2005 on Interactive Poster and Demon-
stration Sessions, pages 25–28. Association for Computational Linguistics, 2005.

56

Healthcare Data Analytics

[67] Bhanukiran Vinzamuri and Chandan K. Reddy. Cox regression with correlation based regu-
larization for electronic health records Proceedings of the 2013 IEEE International Confer-
ence on Data Mining, pages 757–766.

[68] Dipak Kalra and BGME Blobel. Semantic interoperability of ehr systems. Studies in Health

Technology and Informatics, 127:231, 2007.

[69] Ashish K. Jha, David Doolan, Daniel Grandt, Tim Scott, and David W. Bates. The use of
health information technology in seven nations. International Journal of Medical Informat-
ics, 77(12):848–854, 2008.

[70] Elizabeth Davidson and Dan Heslinga. Bridging the it adoption gap for small physician
practices: An action research study on electronic health records. Information Systems Man-
agement, 24(1):15–28, 2006.

[71] Catherine M. DesRoches, Eric G. Campbell, Sowmya R. Rao, Karen Donelan, Timothy G.
Ferris, Ashish Jha, Rainu Kaushal, Douglas E. Levy, Sara Rosenbaum, Alexandra E. Shields
et al. Electronic health records in ambulatory carea national survey of physicians. New
England Journal of Medicine, 359(1):50–60, 2008.

[72] Ebrahim Randeree. Exploring physician adoption of EMRS: A multi-case analysis. Journal

of Medical Systems, 31(6):489–496, 2007.

[73] Robert H. Miller and Ida Sim. Physicians’ use of electronic medical records: Barriers and

solutions. Health Affairs, 23(2):116–126, 2004.

[74] Steven R. Simon, Rainu Kaushal, Paul D. Cleary, Chelsea A. Jenter, Lynn A. Volk, E John
Orav, Elisabeth Burdick, Eric G. Poon, and David W. Bates. Physicians and electronic health
records: A statewide survey. Archives of Internal Medicine, 167(5):507–512, 2007.

[75] Glenn A. Loomis, J. Scott Ries, Robert M. Saywell, and Nitesh R. Thakker. If electronic
medical records are so great, why aren’t family physicians using them? Journal of Family
Practice, 51(7):636–641, 2002.

[76] Alex R. Kemper, Rebecca L. Uren, and Sarah J. Clark. Adoption of electronic health records

in primary care pediatric practices. Pediatrics, 118(1):e20–e24, 2006.

[77] 2003

national

http://www.
commonwealthfund.org/Surveys/2003/2003-National-Survey-of-Physicians-
and-Quality-of-Care.aspx.

physicians

quality

survey

care.

and

of

of

[78] Arun Vishwanath and Susan D. Scamurra. Barriers to the adoption of electronic health
records: using concept mapping to develop a comprehensive empirical model. Health In-
formatics Journal, 13(2):119–134, 2007.

[79] Nir Menachemi and Taleah H. Collum. Benefits and drawbacks of electronic health record

systems. Risk Management and Healthcare Policy, 4:47–55, 2011.

[80] Clinicians and EHR usability. http://www.himss.org/resourcelibrary/TopicList.

aspx?MetaDataID=1721&navItemNumber=17121.

[81] Jeffery L. Belden, Rebecca Grayson, and Janey Barnes. Defining and testing EMR usability:
Principles and proposed methods of EMR usability evaluation and rating. Technical report,
Healthcare Information and Management Systems Society (HIMSS), 2009.

Electronic Health Records: A Survey

57

[82] Ignacio Valdes, David C. Kibbe, Greg Tolleson, Mark E. Kunik, and Laura A. Petersen.
Barriers to proliferation of electronic medical records. Informatics in Primary Care, 12(1):
3–9, 2004.

[83] About CCHIT. https://www.cchit.org/about.

[84] US Department of Health, Human Services, et al. Health information technology: Initial
set of standards, implementation specifications, and certification criteria for electronic health
record technology. Federal Register, 75(8):13, 2010.

[85] Ben Smith, Andrew Austin, Matt Brown, Jason T. King, Jerrod Lankford, Andrew Meneely,
and Laurie Williams. Challenges for protecting the privacy of health information: Required
In Proceedings of the Second
certification can leave common vulnerabilities undetected.
Annual Workshop on Security and Privacy in Medical and Home-care Systems, pages 1–12.
ACM, 2010.

[86] Sandeep S. Mangalmurti, Lindsey Murtagh, and Michelle M. Mello. Medical malprac-
tice liability in the age of electronic health records. New England Journal of Medicine,
363(21):2060–2067, 2010.

[87] Michelle M. Mello et al. National costs of the medical liability system. Health Affairs,

29(9):1569–1577, 2010.

[88] K. Bruce Bayley, Tom Belnap, Lucy Savitz, Andrew L. Masica, Nilay Shah, and Neil S.
Fleming. Challenges in using electronic health record data for CER: Experience of 4 learning
organizations and solutions applied. Medical Care, 51:S80–S86, 2013.

[89] Jionglin Wu, Jason Roy, and Walter F. Stewart. Prediction modeling using EHR data:
challenges, strategies, and a comparison of machine learning approaches. Medical Care,
48(6):S106–S113, 2010.

[90] Chris Paxton, Alexandru Niculescu-Mizil, and Suchi Saria. Developing predictive models
using electronic medical records: Challenges and pitfalls. AMIA Annual Symposium Pro-
ceedings Archive, pages 1109–1115, 2013.

[91] Brian J. Wells, Amy S. Nowacki, Kevin Chagin, and Michael W. Kattan. Strategies for
handling missing data in electronic health record derived data. Strategies, 1(3):Article 7,
2013.

[92] William Hersh. Secondary Use of Clinical Data from Electronic Health Records. Oregon

Health & Science University.

[93] Taxiarchis Botsis, Gunnar Hartvigsen, Fei Chen, and Chunhua Weng. Secondary use of EHR:
Data quality issues and informatics opportunities. AMIA Summits on Translational Science
Proceedings, 2010:1, 2010.

[94] Fabienne C. Bourgeois, Karen L. Olson, and Kenneth D. Mandl. Patients treated at multi-
ple acute health care facilities: Quantifying information fragmentation. Archives of Internal
Medicine, 170(22):1989–1995, 2010.

[95] Harvey J. Murff, Fern FitzHenry, Michael E. Matheny, Nancy Gentry, Kristen L. Kotter,
Kimberly Crimin, Robert S. Dittus, Amy K. Rosen, Peter L. Elkin, Steven H. Brown et al.
Automated identification of postoperative complications within an electronic medical record
using natural language processing. JAMA, 306(8):848–855, 2011.

58

Healthcare Data Analytics

[96] John D. D’Amore, Joshua C. Mandel, David A. Kreda, Ashley Swain, George A. Koromia,
Sumesh Sundareswaran, Liora Alschuler, Robert H. Dolin, Kenneth D. Mandl, Isaac S. Ko-
hane et al. Are meaningful use stage 2 certified EHRS ready for interoperability? Findings
from the smart C-CDA collaborative. Journal of the American Medical Informatics Associa-
tion, 21(6):1060–1068, 2014.

[97] http://www.healthit.gov/sites/default/files/ONC10yearInteroperability

ConceptPaper.pdf.

[98] http://www.phekb.org/phenotype/cataracts.

[99] P. L. Elkin, A. P. Ruggieri, S. H. Brown, B. A. Bauer, D. Wahner-Roedler, S. C. Litin, J Bein-
born, K. R. Bailey, and L. Bergstrom. A randomized controlled trial of the accuracy of clin-
ical record retrieval using SNOMED-RT as compared with ICD-9-CM. American Medical
Informatics Association, 2001.

[100] Shawn N. Murphy, Griffin Weber, Michael Mendis, Vivian Gainer, Henry C. Chueh, Su-
sanne Churchill, and Isaac Kohane. Serving the enterprise and beyond with informatics for
integrating biology and the bedside (i2b2). Journal of the American Medical Informatics
Association, 17(2):124–130, 2010.

[101] John F. Hurdle, Stephen C. Haroldsen, Andrew Hammer, Cindy Spigle, Alison M. Fraser,
Geraldine P. Mineau, and Samir J. Courdy. Identifying clinical/translational research cohorts:
Ascertainment via querying an integrated multi-source database. Journal of the American
Medical Informatics Association, 20(1):164–171, 2013.

[102] Dan M. Roden, Jill M. Pulley, Melissa A. Basford, Gordon R. Bernard, Ellen W. Clayton, Jef-
frey R. Balser, and Dan R. Masys. Development of a large-scale de-identified DNA biobank
to enable personalized medicine. Clinical Pharmacology & Therapeutics, 84(3):362–369,
2008.

[103] Robert J. Carroll, Anne E. Eyler, and Joshua C. Denny. Na¨ıve electronic health record pheno-
type identification for rheumatoid arthritis. In AMIA Annual Symposium Proceedings, volume
2011, page 189. American Medical Informatics Association, 2011.

[104] Joshua C. Denny, Josh F. Peterson, Neesha N. Choma, Hua Xu, Randolph A. Miller, Lisa Bas-
tarache, and Neeraja B. Peterson. Extracting timing and status descriptors for colonoscopy
testing from electronic medical records. Journal of the American Medical Informatics Asso-
ciation, 17(4):383–388, 2010.

[105] Joshua C. Denny, Randolph A. Miller Anderson Spickard III, Jonathan Schildcrout, Dawood
Darbar, S. Trent Rosenbloom, and Josh F. Peterson. Identifying UMLS concepts from ECG
impressions using knowledgemap. In AMIA Annual Symposium Proceedings, volume 2005,
page 196. American Medical Informatics Association, 2005.

[106] Joshua C. Denny. Mining electronic health records in the genomics era. PLoS Computational

Biology, 8(12):e1002823, 2012.

[107] Catherine A. McCarty, Anuradha Nair, Diane M. Austin, and Philip F. Giampietro. Informed
consent and subject motivation to participate in a large, population-based genomics study:
The Marshfield Clinic personalized medicine research project. Public Health Genomics,
10(1):2–9, 2006.

[108] http://emerge.mc.vanderbilt.edu/.

Electronic Health Records: A Survey

59

[109] Katherine M. Newton, Peggy L. Peissig, Abel Ngo Kho, Suzette J. Bielinski, Richard L.
Berg, Vidhu Choudhary, Melissa Basford, Christopher G. Chute, Iftikhar J. Kullo, Rongling
Li et al. Validation of electronic medical record-based phenotyping algorithms: Results and
lessons learned from the emerge network. Journal of the American Medical Informatics
Association, 20(e1):e147–e154, 2013.

[110] http://www.phekb.org/.

[111] Elena Birman-Deych, Amy D. Waterman, Yan Yan, David S. Nilasena, Martha J. Radford,
and Brian F. Gage. Accuracy of ICD-9-CM codes for identifying cardiovascular and stroke
risk factors. Medical Care, 43(5):480–485, 2005.

[112] Elizabeth F.O. Kern, Miriam Maney, Donald R. Miller, Chin-Lin Tseng, Anjali Tiwari, Man-
gala Rajan, David Aron, and Leonard Pogach. Failure of ICD-9-CM codes to identify patients
with comorbid chronic kidney disease in diabetes. Health Services Research, 41(2):564–580,
2006.

[113] Janet F.E. Penz, Adam B. Wilcox, and John F. Hurdle. Automated identification of adverse
events related to central venous catheters. Journal of Biomedical Informatics, 40(2):174–182,
2007.

[114] Carol Friedman, George Hripcsak, William DuMouchel, Stephen B. Johnson, and Paul D.
Clayton. Natural language processing in an operational clinical information system. Natural
Language Engineering, 1(01):83–108, 1995.

[115] Guergana K. Savova, James J. Masanz, Philip V. Ogren, Jiaping Zheng, Sunghwan Sohn,
Karin C. Kipper-Schuler, and Christopher G. Chute. Mayo clinical text analysis and knowl-
edge extraction system (CTAKES): architecture, component evaluation and applications.
Journal of the American Medical Informatics Association, 17(5):507–513, 2010.

[116] http://metamap.nlm.nih.gov/.

[117] Joshua C. Denny, Jeffrey D. Smithers, Randolph A. Miller, and Anderson Spickard. Under-
standing medical school curriculum content using knowledgemap. Journal of the American
Medical Informatics Association, 10(4):351–362, 2003.

[118] Jill M. Pulley, Joshua C. Denny, Josh F. Peterson, Gordon R. Bernard, Cindy L. Vnencak-
Jones, Andrea H. Ramirez, Jessica T. Delaney, Erica Bowton, Kyle Brothers, Kevin Johnson,
et al. Operational implementation of prospective genotyping for personalized medicine: The
design of the Vanderbilt predict project. Clinical Pharmacology & Therapeutics, 92(1):87–
95, 2012.

[119] J. Kevin Hicks, Kristine R. Crews, James M. Hoffman, Nancy M. Kornegay, Mark R. Wilkin-
son, Rachel Lorier, Alex Stoddard, Wenjian Yang, Colton Smith, Christian A. Fernandez,
et al. A clinician-driven automated system for integration of pharmacogenetic consults into
an electronic medical record. Clinical Pharmacology and Therapeutics, 92(5):563, 2012.

[120] David Blumenthal and Marilyn Tavenner. The meaningful use regulation for electronic health

records. New England Journal of Medicine, 363(6):501–504, 2010.