=Paper=
{{Paper
|id=None
|storemode=property
|title=To what entities does an ICD-9-CM code refer? A realist approach
|pdfUrl=https://ceur-ws.org/Vol-754/hogan_krmed2010.pdf
|volume=Vol-754
}}
==To what entities does an ICD-9-CM code refer? A realist approach==
https://ceur-ws.org/Vol-754/hogan_krmed2010.pdf
To what entities does an ICD-9-CM
code refer? A realist approach.
William R. Hogan*
Division of Biomedical Informatics,
University of Arkansas for Medical Sciences, Little Rock, AR
Abstract. I take a view of ICD-9-CM codes as diagnostic statements,
where these statements are about entities that exist in reality. I then
represent these entities according to a realist view of disease, disorder,
and diagnosis as defined by the Ontology for General Medical Science
and using Referent Tracking templates. I illustrate the approach using
ICD-9-CM codes that refer to systemic arterial hypertension. The
approach is general and could help enable interoperability of data
encoded with realist ontologies and data encoded with terminologies and
administrative classifications, such as data from electronic health records
and insurance claims.
1. Introduction
As several researchers have noted, the International Classification of Diseases, 9th
Revision, Clinical Modification (ICD-9-CM) is not truly a classification of disease, but
rather a classification of patients or statements about patients [1, 2]. For example,
Bodenreider et al. note that the term Tuberculosis of adrenal glands, tubercle bacilli
not found (in sputum) by microscopy, but found by bacterial culture is a sentence
schema that contains information about how the disease was diagnosed, which has no
bearing on the disease itself or the type of which it is an instance [1].
The ICD-9-CM Official Guidelines for Coding and Reporting document lends credence
to this view. First, it uniformly refers to ICD-9-CM codes as diagnosis codes, not
disease codes [3]. Second, it also describes combination codes, noting that they refer to
multiple diagnoses (and by extension, denote multiple diseases) [3]. Many ICD-9-CM
codes therefore classify multiple disease particulars in a single patient. They do not
assert that a single disease instantiates multiple types. For example, all codes
beginning with 404 combine heart disease and chronic kidney disease resulting from
hypertension. Thus these diagnosis codes, when applied to a particular patient, refer to
at least three disease particulars that are instances of hypertension, heart disease, and
chronic kidney disease.
Besides combination codes, the existence of epistemological criteria in ICD-9-CM also
supports the diagnosis perspective on ICD-9-CM. For example, all codes starting with
010 through 018 are differentiated by the manner in which tuberculosis was diagnosed
(e.g., tubercle bacilli found (in sputum) by microscopy). A realist-based ontology
should not have an exact match for these codes because they contain an
epistemological criterion. Thus, conversion of these codes to a realist representation is
more complex than a simple assertion of equivalence among codes.
If we view each ICD-9-CM code as a diagnostic statement, then what are these
statements about? As the name of the classification suggests, they are about diseases,
at least. We take the realist view that distinguishes carefully between a diagnosis, a
disorder, and a disease. In this view, diagnostic statements refer to diseases, disorders,
and possibly other entities that exist in reality (which in the case of ICD-9-CM, include
symptoms, findings, and epistemological methods).
The purpose of this work, therefore, is to generate a framework for creating
representations of the entities to which ICD-9-CM codes as diagnostic statements—or
perhaps more generally clinical statements—refer. That is, when a clinician assigns an
ICD-9-CM code to a particular patient, this framework enables a software application
to generate automatically representations of the disease, disorder, and other entities that
are thereby implied.
Another motivation for this approach is to facilitate interoperability of data annotated
using the Referent Tracking (RT) paradigm [4]. This paradigm accommodates data
annotated using ontologies constructed according to the realist approach, as well as
data annotated using “concept”-based terminologies and even administrative
classifications such as ICD-9-CM. If both types of such data are extant within a single
RT system (RTS), however, the data may not be fully interoperable. For example, a
query for type 2 diabetes mellitus using DOID:9352 from the Disease Ontology (DO)
will not return patients whose type 2 diabetes mellitus is recorded using only ICD-9-
CM and/or SNOMED-CT.
In this paper, I outline a general approach to representing the disease, disorder, etc.
particulars in a given patient that are referred to by ICD-9-CM codes relating to
systemic arterial hypertension. These data are the most common form of diagnostic
data, occurring in electronic medical records (EMRs) and insurance claims data. I also
describe an implementation of the approach in the RT paradigm. The approach is
general and could be applied to ICD-10-CM and terminologies like SNOMED-CT.
SNOMED-CT has built-in constructs that might enable a more automated approach.
However, the vast majority of clinical diagnoses to be integrated with basic science
data are encoded with ICD-9-CM.
2. Ontological representation
2.1. Selecting a subset of ICD-9-CM
I searched the tabular list of the current version of ICD-9-CM on the word
‘hypertension’, selecting only codes that represent statements about systemic arterial
hypertension. Hypertension is the most common chronic condition that results in visits
to physician offices [5]. It causes substantial morbidity and mortality [6]. Furthermore,
hypertension is a significant subject of translational research [7, 8], and thus linking
clinical to genomics and proteomics data about patients with hypertension will
increasingly be important.
I excluded ICD-9-CM codes for pulmonary hypertension, hypertension complicating
pregnancy and childbirth (for which there are another 55 ICD-9-CM codes), and
hypertension of other body substances in liquid phase. For example, I excluded 365.04
Ocular hypertension, 416.0 Primary pulmonary hypertension, 459.3 Chronic venous
hypertension (idiopathic), and all codes beginning with 642 Hypertension complicating
pregnancy, childbirth, and the puerperium.
2.2. ICD-9-CM diagnoses of hypertension and related conditions
I found a total of 51 ICD-9-CM codes representing statements that implied either
directly (e.g., essential hypertension) or indirectly (e.g., hypertensive heart disease) that
the patient suffers from systemic arterial hypertension.1 This set excludes 55 ICD-9-
CM codes beginning with 642, which refer to hypertension complicating pregnancy
and childbirth.
A key issue was how to treat ICD-9-CM codes with the phrases ‘unspecified’ or ‘not
elsewhere classified’ in their title. Although one could handle ‘unspecified’ and ‘not
elsewhere classified’ in the RT paradigm, I did not. Typically, ICD-9-CM codes are
assigned within the context of a single chart, and thus the ‘unspecified’ reference refers
to statements made within only that chart. Indeed, the ICD-9-CM Guidelines
specifically state codes … titled “unspecified” are for use when the information in the
medical record is insufficient to assign a more specific code [3].
Given that the vision for the RT paradigm includes semantic interoperability across
organizations within a community, it would be inappropriate to include templates that
deny the presence of other templates within the larger community as a whole.
However, anyone wishing to include templates to handle ‘unspecified’ and ‘not
elsewhere classified’ could adapt this work.
2.3. Definitions of disease, disorder, and diagnosis
In this work, I use the realist definitions of disease, disorder, and diagnosis of
Scheuermann et al. [9] and the Ontology of General Medical Science (OGMS) [10].
Specifically, the definitions from version 0.5 of OGMS2 are as follows:
Disorder: A material entity which is clinically abnormal and part of an
extended organism. Disorders are the physical basis of disease.
Disease: A disposition (i) to undergo pathological processes that (ii) exists in
an organism because of one or more disorders in that organism.
Diagnosis: The representation of a conclusion of an interpretive process that
has as input a clinical picture of a given patient and as output an assertion …
to the effect that the patient has a disease of such and such a type.
The notion is that a disorder is a physical aberration in the body that confers a
disposition to undergo a particular pathological process. Note that absence of
realization of the disposition (e.g., in the presence of medications that maintain blood
1
This list of 51 codes is publicly available at:
http://spreadsheets.google.com/ccc?key=0AnlWz1vGIs2hdF9acjBjRWpheG9DWXdw
MTFhaGc4NGc&hl=en
2
The author is one of the developers of OGMS.
pressure in the normal range) does not imply absence of the disease (disposition).
Patients with hypertension have a disposition to high blood pressure regardless of
treatment: cessation of treatment results in a return to high blood pressure because the
disposition never ceased to exist.
2.4. Representing Hypertension
In the case of hypertension (both the so-called essential and secondary varieties),
significant evidence exists that the underlying disorder is an abnormally elevated
quantity of salt in the blood, which itself is the consequence of a wide variety of
disorders that confer a disposition to enhanced reabsorption of salt by the kidneys [11].
The disease is then the resulting disposition of the blood to exert higher-than-normal
pressure on the walls of blood vessels.
In what follows I use italics for particulars and universals3 and bold for relations.4
Where necessary, I number particulars as dz_1, dz_2, … to distinguish among them.
Now, for a particular patient, to whom one of the 51 ICD-9-CM codes has been
assigned, I say the following:
hs instance_of Human
dz instance_of Hypertension
do instance_of Disorder
do part_of hs
dz inheres_in hs
dz disposition_of do
Now, I treat the portion of salt in the blood as the disorder:
do instance_of Scattered molecular aggregate
The portion of salt has at least one molecule of salt as its grain (else the system will not
know it is a salt aggregate):
mo instance_of NaCl
do has_grain mo
There is a portion of blood that is part of the patient, and the portion of salt is part of it:
pb instance_of Portion of blood
do part_of pb
pb part_of hs
2.5. Essential vs. Secondary vs. Complication of Treatment
Many of the 51 ICD-9-CM codes distinguish ‘essential’ vs. ‘secondary’ hypertension.
The term ‘essential hypertension’ refers to hypertension where the underlying disorder
that disposes to renal salt retention is unknown, whereas secondary hypertension refers
3
I set particulars in lower-case and universals with an upper-case first letter
4
All relations in this work are instance-level relations except instance_of and lacks,
which are instance–universal relations.
to cases whose underlying disorders/diseases are known. As an example of the latter,
reduced blood flow to the kidney due to stricture or stenosis of a renal artery causes the
kidney to retain salt to increase blood pressure (to in turn increase blood flow to itself).
These distinctions have an epistemological basis, and except for ICD-9-CM codes that
explicitly state the disorder that predisposes to renal retention of salt, we add nothing
further. Since the final common pathway of hypertension regardless of these
distinctions is elevated salt content of the blood, which in turn results from a different
disposition (disease) which in turn has a different physical basis (disorder), these
distinctions do not affect our expansions.
There is also an ICD-9-CM code for secondary hypertension resulting from
iatrogenesis (i.e., caused by treatment). Here, a disorder was caused by an instance of
treatment, giving rise to a disposition to renal salt resorption, whose realization caused
the disorder underlying hypertension:
tr instance_of Treatment
tr results_in do_2
do_2 part_of hs
dz_2 disposition_of do_2
dz_2 realized_by sr
sr instance_of Exaggerated renal salt resorption
sr results_in do
2.6. Benign vs. Malignant Hypertension
The words ‘benign’ and ‘malignant’ in ICD-9-CM diagnoses refer to hypertension, not
to other diseases mentioned such as heart disease, chronic kidney disease, etc. Whether
hypertension is benign or malignant refers to the course of the disease. Malignant
hypertension … is a sudden and rapid development of extremely high blood pressure
[12].
Thus, for malignant hypertension, an acute disease course:
dc instance_of Acute disease course
And for benign hypertension, a chronic disease course:
dc instance_of Chronic disease course
And in both cases, associate the course with the disease:
dz realized_by dc
2.7. Hypertensive Heart and Kidney Disease
When ‘hypertensive heart disease’ appears in the title of an ICD-9-CM code, there also
exists a disease of the heart:
dz_2 instance_of Heart disease
do_2 instance_of Disorder
dz_2 inheres_in hs
dz_2 disposition_of do_2
We assert that realization of the hypertension (pathological process) resulted in the
disorder of the heart:
pp instance_of Pathological process
dz realized_by pp
pp results_in do_2
And we need to say that the disorder is part of the heart, and the heart part of the
patient:
ht instance_of Heart
do_2 part_of ht
ht part_of hs
We represent hypertensive chronic kidney disease in the same manner, but additionally
assert a chronic course, and that the disease is realized by the course.
2.8. Presence vs. Absence of Heart Failure
The title of several ICD-9-CM codes refers to the presence or absence of heart failure.
‘Heart failure’ is an ambiguous term that may refer to either a pathological process or
the disposition to it. We use ‘heart failure’ in the latter sense, which is consistent with
medical usage: we treat heart failure, and when the heart is not malfunctioning, the
disposition to undergo malfunctioning remains.
dz_3 instance_of Heart failure
do_3 instance_of Disorder
dz_3 disposition_of do_3
dz_3 inheres_in pt
do_3 part_of ht
Conversely, when the title says ‘without heart failure’, the patient is not the bearer of
any instance of heart failure:
pt lacks Heart failure with respect to bearer_of
3. Implementation in RT templates
RT is a realist approach to managing data by asserting the existence of particulars and
their instantiations and relationships over time [4]. RT has several types of templates
that capture different information about particulars [13]. An A-template asserts the
existence of a particular:
Ai = < IUIp, IUIa, tap >
Specifically, it …captures the assignment of an IUIp to a particular at time tap by the …
author IUIa [13]. Each particular to which the new RT templates refer—including
diseases, disorders, pathological processes, disease courses, and so on—requires an A-
template. For example,
A1 =
A PtoU template states that a particular IUIp instantiates a universal u at time tr, where
ta is time of authorship, o is the ontology from which the universal u derives, and inst is
the instance_of relation from the Relation Ontology (RO):
Ui = < IUIa, ta, inst, o, IUIp, u, tr >
Ex: U1 = < IUIh, 5/14/10, inst, DO, IUIdz, Hypertension, 5/7/10 >
A PtoP template states that a set of particulars P stand in relation r from ontology o to
each other:
Ri = < IUIa, ta, r, o, P, tr >
Ex: R1 = < IUIh, ta, part_of, RO, , tr >
A PtoLackU template states that a particular IUIp lacks a relation r from ontology o to
any instance of a universal u:
U i = < IUIa, ta, r, o, IUIp, u, tr >
Ex: U1=< IUIh, ta, bearer_of, RO, IUIpt, Heart failure, tr >
A PtoCO template states that the author has annotated a particular with a ‘concept’
code co from a terminology cbs:
Coi = < IUIa, ta, cbs, IUIp, co, tr >
Ex: Co1=< IUIh, ta, ICD-9-CM, IUIpt, 401.1, tr >
I refer to the process of converting the PtoCO template to templates that represent the
entities implied by a diagnosis—as outlined above—as an expansion. I use the term
‘expansion template’ to refer to an RT template created to represent these entities and
their relationships one another. I propose a new subsystem of the RTS called the
‘expansion subsystem (ESS)’ that performs the expansion. The templates I described
above are sufficient to represent all the statements about particulars I have listed here.5
Each expansion template requires the IUI of an author (IUIa). One possibility is to use
the IUIa from the original PtoCO template that contained the ICD-9-CM code in
question. However, this approach would incorrectly assign any errors made in the
expansion of the code to that author. Other alternatives include having a single IUIa for
the ESS or having a different IUIa for each expansion of a single ICD-9-CM. I adopt a
single IUIa for the ESS.
Except for A-templates and Meta-templates (which we do not discuss further here),
each template has two timestamps: one to capture the time at which the assertion holds
in reality, or tr and one to capture the time at which the assertion was made, or ta. The
latter timestamp is straightforward: I simply use the time at which the ESS completed
the expansion. For the former timestamp, I use the tr of the PtoCO template, because it
is the time at which the diagnostic statement of the ICD-9-CM code holds, and thus the
time at which the assertions implied by the ICD-9-CM code hold.
Finally, the expansion templates could refer to a particular already assigned an IUI in
the RTS. The problem of ensuring unique reference of IUIs and potentially merging
duplicates is not unique to this work: it is a problem for referent tracking in general. I
assume the existence of a solution in the RTS to this problem that the ESS may employ.
5
The complete set of expansions is in the same document with the codes.
4. Discussion
I have defended a view of ICD-9-CM codes as representing not diseases, but diagnostic
or even clinical statements. Using realist definitions of disease and its related entities,
and the referent-tracking paradigm, I have enabled representation of the entities in
reality referred to by 51 ICD-9-CM codes that imply the existence of hypertension.
This approach may help unlock EMR data for translational science.
Acknowledgements
This work was supported by award number 1UL1RR029884 from the National Center
for Research Resources.
References
1. Bodenreider O, Smith B, Burgun A. The ontology-epistemology divide: A case study in
medical terminology. Proceedings of FOIS 2004; 2004; Turin.
2. Ceusters W, Smith B, DeMoor G. Ontology-based integration of medical coding systems and
electronic patient records 2006 [cited 2010 Apr 16]. Available from:
http://ontology.buffalo.edu/medo/CodingAndEHCR.pdf
3. ICD-9-CM Official Guidelines for Coding and Reporting2009 [cited 2010 Apr 16]: Available
from: http://www.cdc.gov/nchs/data/icd9/icdguide09.pdf.
4. Ceusters W, Smith B. Strategies for referent tracking in electronic health records. J Biomed
Inform. 2006 Jun;39(3):362-78.
5. Cherry D, Woodwell D, Rechtsteiner E. National Ambulatory Medical Care Survey: 2005
Summary. Hyattsville, MD: National Center for Health Statistics,2007.
6. Chobanian AV, et al. The Seventh Report of the Joint National Committee on Prevention,
Detection, Evaluation, and Treatment of High Blood Pressure: the JNC 7 report. Jama. 2003
May 21;289(19):2560-72.
7. Cowley AW. The genetic dissection of essential hypertension. Nat Rev Genet. 2006
Nov;7(11):829-40.
8. Tomaszewski M, Zimmerli L, Charchar FJ, Dominiczak AF. Genetic information in the
diagnosis and treatment of hypertension. Curr Hypertens Rep. 2006 Aug;8(4):309-16.
9. Scheuermann RH, Ceusters W, Smith B. Toward an ontological treatment of disease and
diagnosis. AMIA Summit on Translational Bioinformatics; 2009.
10.Ontology for General Medical Science. 2009 [cited 2009 11/24]; Available from:
http://code.google.com/p/ogms/.
11. Lifton RP, Gharavi AG, Geller DS. Molecular mechanisms of human hypertension. Cell.
2001 Feb 23;104(4):545-56.
12. Malignant hypertension. 2009 [cited 2010 Apr 16]; Available from:
http://www.nlm.nih.gov/medlineplus/ency/article/000491.htm.
13.Manzoor S, Ceusters W, Rudnicki R. Implementation of a referent tracking system. Int J of
Healthcare Information Systems and Informatics. 2007 Oct-Dec;2(4):41-58.