Anda di halaman 1dari 5

A Findings Model for an Ambulatory Pediatric Record:

Essential Data, Relational Modeling, and Vocabulary Considerations

Richard N. Shiffman, M.D., M.C.I.S.
Center for Medical Informatics and the Department of Pediatrics
Yale School of Medicine, New Haven, Connecticut

Effective, computer-based representation of Clinical Environment

clinical observations requires balancing the advantages The Department of Pediatrics at the Yale School
of structured, coded descriptions against those offree of Medicine has established a center for the care of
text narrative. An essential data set of relevant signs Children with Special Health Care Needs (CSHCN).
and symptoms was defined by a multidisciplinary It provides diagnostic and therapeutic services to
group based on management goals published in a pediatric patients who have a broad range of disabling
national guideline to meet the needs of clinicians in conditions, including spina bifida, cerebral palsy,
the Spina Bifida Clinic at Yale-New Haven Hospital. congenital and acquired heart disease, cleft lip and/or
These core data elements are stored in a structured palate, rheumatoid arthritis, and cystic fibrosis. A
format. Additional material is stored as free text. replicable and extensible information architecture was
A relational schema was devised that permits needed to manage the clinical, research, and
storage of both coded findings and narrative. administrative data generated in caring for children
Symptoms and signs are represented as subtypes of a with these chronic diseases. A prototype clinical
supertype patient finding entity; they inherit common information system has been developed using
attributes and specialize others. Microsoft Access Version 2.0 (Microsoft Corp.,
The IVORY vocabulary was supplemented and Redmond, Wash.) with a plan to migrate to a client-
modified to provide terms that describe relevant server architecture.
clinical observations. For this application, fields were Initially, the system is being implemented in the
added that enable predictive data entry of findings Spina Bifida Clinic at Yale-New Haven Hospital.
based on patient age and gender. Comprehensive care of children who have spina
bifida-a congenital disorder that results from
INTRODUCTION abnormal intrauterine development of the central
nervous system and spine-requires coordination of
A major challenge that faces the designers of a multiple disciplines, including neurosurgery,
computer-based medical record is to effectively orthopedics, urology, pediatrics, nursing, social
represent the clinical data that are captured during a services, physical and occupational therapy, orthotics,
health care encounter. Free-text narrative, the and genetic counseling. The clinical information
conventional representation in the paper-based record, requirements in this clinic are therefore quite
is flexible, expressive and familiar to clinicians [1]. extensive and should be representative of the needs of
With current technology, however, computers are able other ambulatory care environments.
to make only limited use of information stored in a
free-text format [2]. On the other hand, coded data- CORE DATA SET
precisely defined and based on controlled
vocabularies-can be used to organize the medical Several initiatives in the United States and
record for efficient search, retrieval, and display, to Europe have focused on the specification of basic
aggregate data for research and administrative information to support clinical services. For the
functions, and to trigger decision support. However, ambulatory setting, the Uniform Ambulatory Care
coded data may be insufficiently expressive and Data Set defines a common core of data items and
structured data entry may be unwieldy. provides standard definitions [3]. Likewise, the
This paper describes an approach that was applied Advanced Information in Medicine Program of the
to balance these representations in a computer-based European Economic Community has defined
ambulatory record. A core data set was prospectively minimum data sets to characterize clinical
identified that is maintained in coded format to meet information requirements and constraints and to
the needs of clinicians, followed by those of enhance uniformity [4]. Both are predicated on an
researchers and administrators. Additional information assumption that there is a core of data common to the
is maintained in free-text format. We describe the needs of multiple users that should have a priority for
process for choosing the core data sets, the standardization. However, neither of these multi-
development of a relational architecture for storage of purpose data sets is sufficiently detailed to adequately
these data, and the selection of a controlled vocabulary describe the findings observed during a clinical
for representation of the core data. encounter.

0195-4210/95/$5.00 C) 1995 AMIA, Inc. 411

The American Society of Testing and Materials neurosurgery, nursing, and pediatrics identified 17
Standard ASTM E1384-91 identifies the essential symptoms and 22 signs that are directly relevant to
content and logical structure of an electronic primary the above-mentioned clinic goals. It was agreed that
health record [5]. The data elements included are these data elements would be coded in a structured
extremely broad-based and appear comprehensive, but manner. Additional findings data that are collected
components that define clinical observations during during clinical encounters can be captured as free text
encounters are minimal. and are available in clinical reports, but they cannot
Moidu et al. suggest that an essential data set be aggregated or used for record organization.
should include "just as many data elements as required
to provide the essential foundation for decision- RELATIONAL DESIGN
making.. .to identify the need for care, [and] to
monitor and evaluate the impact of the care provided" The Canon group has focused collaboratively on
[6]. This definition was applied to prospectively the representation of medical concepts related to
define data relevant to clinical management and radiography [11] and their publications have
outcomes assessment for the CSHCN program. concentrated on that domain, e.g. [12, 13]. Other
efforts to model clinical observations have dealt with
Use of Guidelines to Define Core Data the realms of laboratory medicine [14], and endoscopy
Well-crafted guidelines represent a valuable [15]. However, these publications are only indirectly
knowledge resource. They provide up-to-date applicable to pragmatic issues of database design for
information that has been sanctioned by the symptom and sign data.
sponsoring organization. Such information can be Recently, Dolin has presented a high-level
used to break the knowledge acquisition bottleneck for conceptual model of symptoms that uses a nested,
the development of knowledge-based systems and can polyhierarchical representation [16]. His use of a
serve as a basis for decision support [7]. In this work, single table to store symptoms, modifiers, and
knowledge from a published practice guideline was relationships is simpler than a conventional, multi-
reused to define an essential data set for an ambulatory table design but manipulation of the data is more
patient record [8]. complex.
The Spina Bifida Association of America has
published guidelines for the management of patients Structured Coding of Findings Data
with spina bifida, which are intended to maintain The approach described by Fleming and von
optimal health status, prevent secondary disabilities, Halle was applied to the logical design of a findings
maximize the potential to participate in society, and model for the Spina Bifida Clinic database [17]. This
foster independence [9]. The guidelines provide age- methodology systematically analyzes and graphically
and discipline-specific recommendations to health care models information requirements and translates the
providers. model into a stable relational implementation. The
For example, the Neurosurgical guidelines from procedure is data-driven, i.e., it is based on an
infancy through adolescence focus on 3 major goals: understanding of how the information is used without
* maintenance of normal intracranial pressure consideration of specific processing patterns. Clear,
* recognition of presence of the Chiari malformation graphical diagrams are produced that facilitate
* recognition of evidence of cord-tethering symptoms communication among designers, developers, and
Clinical management pertinent to these goals requires users. The end-result is a consistent, sharable, and
regular consideration and documentation of a flexible database.
significant number of historical findings and In this report, attention is focused on modeling
examination observations. of symptom and sign data. A skeletal user view is
Comparatively little attention has been paid to created to define and model the major relevant data
data management for the most fundamental objects-initially without supportive detail. The
observations made by clinicians-information elicited entities of interest in this model include PATIENTs,
in the clinical history and physical examination. It is PROVIDERs, ENCOUNTERs, and the FINDINGs
estimated that these findings lead to a diagnosis in themselves. SYMPTOMs and SIGNs are subtypes of
90% of cases, without requiring the expense and the supertype entity FINDING (as are LAB
discomfort associated with supplemental laboratory RESULTs and IMAGING RESULTs). Symptoms
tests and imaging procedures [10]. Findings data are represent phenomena experienced subjectively by the
also essential to characterize clinical outcomes, an patient and reported by the patient or a surrogate
area of increasing concern in the current managed care historian. Signs are objectively observed by
environment. clinicians. Each of these subtype instances represents
For the neurosurgical component of the Spina the same "object" in the real world as the supertype;
Bifida Clinic database, representatives from


Figure 1. Logical model of findings for the Spina Bifida Clinic database.
each subtype instance has all the properties of the Non-key attributes for the FINDING entity are
supertype plus some additional properties; and, for common to all occurrences of the supertype. General,
each instance of the subtype, there exists exactly one non-key attributes that are pertinent to all findings
instance of FINDING, although the converse need not include:
be true. * Finding ID: a pointer to the clinical vocabulary term
Relationships link these major entities. list (see below)
PATIENTs and PROVIDERs are linked one-to-many * Comment: this attribute permits free-text
to ENCOUNTERs. ENCOUNTERs and FINDINGs description to capture the richness of the clinical
are related many-to-many. Therefore, an associative encounter. This text field can be used to modify
entity (PATIENT_FINDING) is defined that relates specific findings, e.g., Rector's example of angina
one-to-many with both ENCOUNTER and a evoked by walking past the frozen food counter, or
FINDING list. SYMPTOMs, SIGNs, LAB it can store narrative not associated with any coded
RESULTs, and IMAGING RESULTs represent data item.
mutually exclusive categories of * Context: abdominal pain may occur both as a
PATIENT_FINDING, each associated 1:1 with the SYMPTOM reported by the patient and as a SIGN
supertype. elicited by the examiner in response to deep or
Each instance of these entities is defined by a superficial palpation. Likewise, patients may report
key, a unique numeric identifier. The primary key of wheezing or clinicians may detect it on auscultation
each subtype entity is the same as that of the of the chest. The context field identifies the specific
supertype FINDING entity. source of the finding.
The next step in the creation of the logical model * Absent: this Boolean attribute allows expression of
is to add non-key attributes to each entity (Figure 1). negation
Fleming and von Halle recommend that non-key * Certainty: this attribute expresses various levels of
attributes be placed as high as possible in the logical certainty, e.g., probable, possible, P=.75
model; following normalization principles, each must * Comparison with: this attribute stores the identifier
be fully determined by the entire primary key. of another finding with which the current finding is
Non-key attributes for PATIENT and compared
PROVIDER entities are implementation specific and * Comparison operator: the relationship of the current
beyond the scope of this paper. ENCOUNTERs are finding to the comparison finding, e.g., smaller
typically described by a type of service and a site of than, milder than, >=, etc.
service, as well as provider ID, patient ID, and Date- * EnteredBy, TimeStamp: The identifier of the user
time (an alternate, composite key). and the time the finding is recorded.

The SYMPTOM subentity inherits all of the location, frequency, severity, and aggravating factors.
above attributes from the FINDING entity and adds Findings terms are linked by associative tables to
specialized attributes. A primary difference between relevant modifiers. Many of the modifiers match
symptoms and signs is the fact that symptoms are attributes in the relational model described here. The
described by temporal modifiers. These attributes expressiveness of this vocabulary makes it a
include: reasonable choice for basic terminology.
* Onset: e.g., abrupt, indolent, one day, 2 weeks For this application, predictive data entry "forms"
* Past history: references previous episodes of a were designed to display only the symptoms and
SYMPTOM, e.g., 1 year ago signs relevant to a given patient's age group and
* Duration: the length of time a SYMPTOM gender. This necessitated adding attributes to the
continues vocabulary terms (Figure 2). With human growth and
* Frequency: e.g., rare, constant, seasonal development, the symptoms and signs that indicate
* Trend: unchanging, improving, worsening the outcomes of interest change. For instance,
Both SYMPTOMs and SIGNs are modified by: symptoms of increased intracranial pressure in infancy
* Intensity, e.g., moderate, disabling, copious, Grade include irritability, lethargy, and vomiting. Signs
II/VI, include a full anterior fontanel, accelerating head
THE SIGN entity adds specializations for: circumference growth rate and sunset sign. On the
* Measured values other hand, older children may complain of headaches
* Units and suffer from memory disturbance, blurred vision,
e.g., 120 (Measured value) millimeters of mercury and decreased school performance. Additionally, the
(Units). fontanel closes, head circumference does not change in
Some SYMPTOM and SIGN attributes are response to acute pressure changes, and papilledema is
multi-valued, i.e., a given FINDING may take on more easily ascertained. Similarly, gynecologic terms
several modifier values simultaneously. These are pertinent only when the patient is female.
attributes are classified as child entities:
* Location: a modifier may be necessary when the
finding does not connote location, e.g., a rash may ModHifers Associative Tables
be present on the right arm and left leg. SevertylD
Severky Nanie |
* Laterality; a topographic modifier, e.g., left, right _Severity
bilateral |Location[) is|FindnglD
* Quality: e.g., red, papular, matted, coarse |Location Name | LocalorlD !
* Radiation: e.g., chest pain can radiate to the left arm QuIkylD FindngiD
and the left jaw QudlyName QuakylD
* Elicited by: e.g., meals, 1 flight of steps, deep * U
palpation, Ortolani's maneuver, caloric stimulation * U Cl'linical Ybcabulary
* Relieved by: e.g., sitting up, OTC analgesics, * U

bowel movement. Figure 2. Relational structure of the modified

The logical model depicted in Figure 1 can be controlled vocabulary.
readily transformed into a relational implementation.
Tables are created to represent each entity. Columns Such restrictions are easily added as fields to the
in these tables represent each attribute. Indexing, IVORY findings vocabulary. The fact that the finding
redefinition of columns and tables, denormalization, identifier carries no contextual information (as it does
and other "tuning" procedures may then be performed in hierarchical classification systems such as ICD-9-
to meet performance requirements. CM and SNOMED) facilitates addition of new terms.
As noted by Lindberg and Humphreys, "There are
VOCABULARY no existing controlled vocabularies that can be
Encoding data related to symptoms and signs has recommended without reservation for even selected
been challenging because the concepts are often elements of the automated patient record" [21] The
IVORY vocabulary lacks many of the terms necessary
complex and a comprehensive controlled vocabulary for the Spina Bifida Clinic application. Of the 43
has been lacking. We modified the IVORY terms required to encode core data for the
vocabulary for use in this application [18]. This neurosurgical component of the database, the IVORY
vocabulary, originally derived from the Wisconsin vocabulary lacks terms for 22 and has marginal
Ambulatory Research Project, was modified by semantic matches for another 2.
Campbell and Musen for use in their IVORY Moorman et al. caution that the use of a limited
progress note tool [19]. set of modifiers prohibits domain completeness and
The IVORY vocabulary includes unique results in a limited descriptive level of detail [20]. A
identifiers for 966 findings terms and a large number facility is included in the database that allows the
of modifier terms grouped by categories such as

option to add terms for modifiers. While initially 7. Shiffman RN. Towards effective implementation
coded as "Other", such terms can be reviewed and of a pediatric asthma guideline: integration of
added to the vocabulary. decision support and clinical workflow support.
The recent release of version 3.1 of SNOMED In: Ozbolt J, ed. Proc 18th SCAMC.
International (College of American Pathologists, Washington: 1994: 797-801.
Northfield, Illinois) promises access to a rich, 8. Musen MA. Dimensions of knowledge sharing
standardized nomenclature of concepts for symptoms and reuse. Comput Biomed Research
and signs. It contains more than 800 newly added 1992;25:433-67.
signs and symptoms. Of the neurosurgical concepts 9. Rauen K. Guidelines for Spina Bifida Health Care
required for the Spina Bifida Clinic database, Services Throughout Life.Washington, DC:
SNOMED lacks terms for only two (decreased school Spina Bifida Assocaition of America, 1990.
performance and sunset sign). SNOMED codes can be 10. Lipkin M. The care of patients: perspectives and
added as attributes of the IVORY vocabulary. practice. New Haven, Conn.: Yale University
Press, 1987:82.
FUTURE DIRECTIONS 11. Cimino JJ. Controlled medical vocabulary
construction: methods from the Canon Group. J
Each of the components described above will be Am Med Informatics Assoc 1994; 1:296-7.
extended as this clinical database is fully 12. Friedman C, Cimino JJ, Johnson SB. A schema
implemented. Additional disciplines will define for representing medical language applied to
essential data sets and the adequacy of this approach clinical radiology. J Am Med Informatics Assoc
for selection of core data will be evaluated. The 1994; 1:233-48.
relational model of findings will be extended to allow 13. Bell DS, Pattison-Gordon E, Greenes RA.
findings to support meta-level problems, and Experiments in concept modeling for
assessment and plan entities will be added. Necessary radiographic image reports. J Am Med
terminology will be added to the vocabulary and Informatics Assoc 1994;1:249-62.
linkages to SNOMED will be enhanced. 14. Stahlhut RW, McCallie DP, Waterman DM, et
Acknowledgments al. A relational model for clinical objective
This work was supported in part by a grant from the results. Proc 14th SCAMC 1990:354-58.
Bureau of Community Health and Addiction Services, 15. Gouveia-Oliveira A, Salgado NC, Azevedo AP,
Connecticut Dept. of Health and Addiction Services. et al. A unified approach to the design of clinical
Special thanks to Larry Wright for his insightful reporting systems. Meth Information Medicine
comments. We are grateful to the Section on Medical 1994;33:479-87.
Informatics at Stanford University for making the 16. Dolin RH. Modeling the relational complexities
IVORY vocabulary available. of symptoms.Meth Inform Med 1994;33:448-53.
17. Fleming CC, von Halle B. Handbook of
References relational database design. Reading, Mass.:
1. Dick RS, Steen EB, ed. The computer-based Addison-Wesley, 1989
patient record: an essential technology for health 18. Campbell KE, Musen MA. Creation of a
care. Washington, D.C.: National Academy systematic domain for medical care: the need for a
Press, 1991 comprehensive patient-description vocabulary. In:
2. Rector AL, Nowlan WA, Kay S. Foundations for Lun KC, et al., ed. MEDINFO 92. Elsevier
an electronic medical record. Meth Inform Med Science Publishers B.V. (North-Holland), 1992:
1991 ;30: 179-86. 1437-42.
3. Report of the National Committee on Vital and 19. Campbell KE, Wieckert K, Fagan LM, et al., A
Health Statistics and the Interagency Task Force computer-based tool for generation of progress
on the Uniform Ambulatory Care Data Set. U.S. notes. In: Safran CS, ed. Proc 17th SCAMC
Dept. of Health and Human Services, 1989. Washington: 1993: 284-8.
4. Parkin D, Hutchinson A, Fisher PJ. Minimum 20. Moorman PW, van Ginneken AM, van der Lei
data sets for ambulatory care. In: O'Moore R, et J, et al. A model for structured data entry based
al., ed. MedInfo Europe. Glasgow, Scotland: on explicit descriptional knowledge. Meth Inform
Springer Verlag, 1990: 88-93. Med 1994;33:454-63.
5. ASTM. Standard guide for description for content 21. Lindberg DAB, Humphreys BL. The Unified
and structure of an automated primary record of Medical Language System (UMLS) and
care E1384-91. 1991. computer-based patient records. In: Ball MJ,
6. Moidu K, Singh AK, Bostrom K, et al. Towards Collen MF, ed. Aspects of the computer-based
an essential data set: applicability in the domain patient record. New York: Springer-Verlag, 1992:
of maternal health services. Meth Inform Med 172.
1992;3 1:182-92.