AJR 2001; 176:335-339
© American Roentgen Ray Society
Networked ICD-9 Coding System for a Radiology Department
Chris Sistrom1 and
Walter Drane
1
Both authors: Department of Radiology, P.O. Box 100374, University of Florida
School of Medicine, 1600 S.W. Archer Rd., Rm. G387, Gainesville, FL
32610.
Received May 17, 2000;
accepted after revision July 11, 2000.
Address correspondence to C. Sistrom.
Abstract
OBJECTIVE. We describe a networked database system to implement a
departmental list of International Classification of Diseases, 9th
edition (ICD-9) codes for use by reporting radiologists.
CONCLUSION. Our system is reliable, easy to use, and has increased
coding accuracy and speed. In-house development allowed designing a system to
specifically address the needs of the radiology group. The resulting list of
codes accurately reflects the current practice environment and grows
dynamically to suit changing needs.
Introduction
Documentation of radiology services for many medical payers (including
Medicare) requires that each report for a procedure or examination be appended
with one or more codes found in the International Classification of
Diseases, 9th edition (ICD-9)
[1]. These codes serve to
document the sign, symptom, or relevant clinical diagnosis needed to justify
the service. In many practices, the task of assigning these codes falls to
coding clerks employed by the physician group or hospital. In some practices,
radiologists themselves are given the job of coding cases as they dictate
them. This task adds an additional step to interpreting a radiologic
examination. After using printed lists of codes for several years, we devised
a computerized solution that has reduced the time required for coding and
increased accuracy and reproducibility.
Background
Our radiology practice is conducted in a 577-bed tertiary care hospital
with three satellite imaging centers. Approximately 220,000 examinations are
performed and interpreted per year. The radiology staff consists of 22
attending faculty, 12 fellows, and 24 residents. All studies except
angiography are read from a hybrid picture archiving and communication system
(PACS) developed in house [1,
2]. Reports are generated with
the PowerScribe (Lernout and Hauspie, Burlington, MA) voice-activated
reporting system (VRS). This system runs over a Windows NT (Microsoft,
Redmond, WA) network consisting of a server and 35 reporting clients. The
reporting system is interfaced with a radiology information system (RIS) and,
through it, is interfaced to the hospital information system (HIS). Both
systems are from SMS (Shared Medical Systems, Malvern, PA). The department
employs two coding managers and three quality coders. They are responsible for
generating all bills for radiology services, including both technical and
professional components, and are capable of coding all procedures. However,
the volume of work makes it impossible for them to do so. Additionally,
radiologists have access to information about the patient's condition not
available to coders who only have the radiology report and the ordering
information. Therefore, reporting radiologists are required to append one or
more ICD-9 codes to every case they dictate.
Before developing the ICD-9 database system, we posted printed lists of
relevant ICD-9 codes at each reporting station. Radiologists looked up codes
on the basis of information found on the examination request, clinical
information obtained from ordering physicians, or patient interviews. They
then dictated or typed them in the report. ICD-9 code books were kept at each
workstation for unusual circumstances. There were several problems with this
system. The most troublesome problem was keeping the printed lists up-to-date
and complete. A list of codes was kept by each divisional director in
disparate locations and different formats. To avoid finding the most recent
list and looking up a code, radiologists often relied on memory. This habit
resulted in many radiologists using a limited number of codes. Even with a
small set of codes committed to memory, radiologists made inevitable errors.
Furthermore, cases that should have been given more than one code were often
incompletely coded. Even if the radiologist found or remembered the correct
codes, the process of typing or dictating them into the report sometimes
resulted in errors. When these reports were parsed by the editing function of
our billing system, numerous cases were rejected for unknown or missing codes.
These reports were then copied and distributed to the dictating radiologist
for correction. In most cases, the billing office requested that the
radiologist dictate an addendum to the report containing the correct code.
When the billing office identified patterns of inaccurate coding, they would
distribute memos or send e-mail to all physicians detailing these problems,
but this effort often failed to completely correct such inconsistencies.
System Requirements, Architecture, and Design
The most inportant requirement of our system is a master list of relevant
and accurate departmental ICD-9 codes in a common location that is always
available for radiologists while dictating. The same list must be
simultaneously available to quality coders to modify as needed. There should
also be a mechanism whereby radiologists can add codes not contained in the
list. A radiologist should have to manually look up a code in the ICD-9 book
only once. After that, it should be available in the database for all to use.
Codes should be listed several times if the description has more than one
common nomenclature. For example, "hepatic insufficiency,"
"elevated LFTs," and "liver function abnormality"
would all refer to the same code but would require separate entries. The list
should be available in at least two orderings: descriptive text and code
number. The system should provide a simple and robust keyword search method
for finding codes.
The optimal solution meeting these requirements centers around a networked
front end and back end database system, with the master code list (back end)
residing on a centralized server. The database and client software were
developed with Access 97 (Microsoft). The front end consists of clients for
viewing, searching, and adding to the list of codes. This information is
installed on all dictation workstations and on many faculty office computers.
Each quality coder and the billing managers have a second client application
for listing, sorting, modifying, and replicating codes. The preexisting
departmental computer network provided the ideal infrastructure for this
system. The server complex already dedicated to the voice-activated reporting
system had ample processing and storage capacity to handle the relatively
small-scale task of hosting the database.
The database design is straightforward. Each record in the departmental
code table represents an individual ICD-9 code. The field structure is
outlined in Table 1. The first
field contains the numeric code; another holds the description. A single code
may be duplicated any number of times with different descriptors to allow
users to find the code by synonymous terms and phrases. There are eight
Boolean fields labeled with practice areas that allow a single code to be
related to more than one practice area without having to duplicate the record.
A date field is assigned the current date when a new code is entered so that
our quality coders can review recently added codes and check them for
accuracy. A second table is similar to the first except that it does not have
the practice area fields or the date field. It contains the full ICD-9 code
set consisting of some 15,000 entries, downloaded from the United States
Health Care Financing Administration Web site
(http://www.hcfa.gov/stats/pufiles.htm#icd), parsed, and converted into a
database table.
ICD-9 Code Finder Functionality
ICD-9 code finder functionality is provided by a client application program
available on all reporting workstations, physicians' office computers, and
many other departmental computers. The user works from a single screen
(Fig. 1). From it, subsets of
codes can be selected by practice area, the entire departmental list, or the
full code set. Drop-down lists sorted alphabetically by text descriptor and
numerically sorted by code number can be activated from the main screen. Most
useful is a keyword search function. Users type in a single search term, and
all the codes matching the key word or string are displayed
(Fig. 2). The search function
allows partial keywords to be searched. For example, searching
"carcino" will find all records containing words such as
"carcinoma," "carcinomatosis," or
"adenocarcinoma."
View larger version (49K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 1. —Main screen of code finder client. Case has been coded for
"ascites" and "liver failure" with appropriate code
string shown at bottom. String "ICD-9 Code: 571.5, 789.5" is
automatically placed in Windows copy buffer for pasting in dictation.
|
|
Our billing system requires that the ICD-9 code be listed at the top of the
dictation in a specific format. Before using our system, radiologists had to
look up codes from paper lists or a book and then type or speak the numbers
into the VRS. Errors frequently resulted from mistyping or misrecognition of
the spoken digits by the VRS. To correct this problem, we added a feature to
the code finder that allows users to build a code string. After they find a
relevant code, users can click a button on the form to append it to any
previously identified codes for the case at hand. This string is automatically
placed in the Windows copy buffer every time it is modified. When ready, the
user simply activates the VRS window and pastes the code string in the report.
This feature completely eliminates errors in transferring codes to the
report.
A function to allow radiologists to add new codes provides the means for
the database to be continuously updated. A single button brings up a small
screen (Fig. 3) on which the
radiologist can type in a new code number with the text descriptor and check
the relevant practice areas. Another function allows the radiologist to move a
selected entry from the full set of codes into the departmental set
(Fig. 4). Whenever a new entry
is added to the departmental code set database by either method, the current
date is appended to the record.
View larger version (35K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 3. —Adding new code. "ADD A NEW CODE" button on the
finder screen activates this screen. Code for "fecal incontinence"
(787.6) is being added. It has been tagged as useful in body imaging and
fluoroscopy (gigu) areas. Data checking ensures that user has entered code
number, description, and checked at least one practice area.
|
|
View larger version (73K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 4. —Moving code from full code set to departmental list.
"MOVE THIS CODE" button on the finder screen activates this
screen. Code for "secondary malignant neoplasm of kidney" (198.0)
will be moved. It has been tagged as useful in body imaging and fluoroscopy
(gigu) areas. If code already exists in departmental set, user is notified.
Radiologist can alter descriptive text if desired. At least one practice area
is required.
|
|
ICD-9 Code Manager Functionality
By design, the code finder client blocks the user's ability to modify or
delete existing codes. These functions are relegated to the code manager
client. Although the code finder is widely available, the code manager client
is only installed on a few specific workstations. Physician managers, quality
coders, and billing managers have exclusive access to the code manager
functions. When activated, this application brings up the master list of codes
in a scrollable data sheet format similar to a spreadsheet
(Fig. 5).
View larger version (44K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 5. —Code Manager client. Data sheet type view is presented on
startup. Double-clicking on descriptor text field activates modify, delete,
and duplicate functions (Fig.
6). Double-clicking on code field activates sorting and search
functions (Fig. 7).
|
|
View larger version (49K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 6. —"Edit Single code." This feature allows changing
code, descriptor, and practice area designation. Entry can be duplicated or
deleted. Duplication facilitates giving single numeric code more than one
meaningful description by altering text string in duplicate. This code might
be duplicated and changed to "Muscular pain" to allow same code to
be found under either designation.
|
|
View larger version (38K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 7. —"Recorder and Search." This feature allows
ordering by code, descriptor, or date of entry. Search function positions
record pointer to first or next matching code in data sheet. User has entered
both code number (123.1) and text (Cysticercosis) search strings. Typically,
either code number or text search is done.
|
|
Double-clicking over the descriptor field activates a pop-up form that
allows the user to work with an individual code
(Fig. 6). Typing in the
appropriate fields serves to modify the code, and check boxes can be clicked
on or off to change the practice area designations. When finished, the user
can accept or discard the changes. Other buttons allow deleting, replicating,
and adding new codes.
Double-clicking over the numeric field brings up a second subform that
allows ordering the code set by number, descriptive text, or date of entry
(Fig. 7). Sorting by date of
entry is particularly valuable because it allows the coders to regularly check
newly added codes for accuracy and relevance. This check is done by ordering
the codes by date and scrolling to the bottom of the list. A search feature
allows a specific entry to be found by numeric code, descriptive text, or
both.
Initial Experience with the System
We have used the ICD-9 code-finding application in full production for
about 1 year. The code list management functions were activated 6 months ago.
Both radiologists and quality coders consider the system mission critical and
use it many times per day. Because the client programs are small, they can run
all the time and be minimized or restored as needed. The key to successful
implementation was having the program installed at all reporting stations. The
instant and efficient feedback provided by the radiologist's capability to add
new entries to the departmental list and the ability of quality coders to
identify and check those entries for accuracy on a daily basis have proved
invaluable. Figure 8 shows how
the database system is integrated into the departmental workflow.
View larger version (30K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 8. —Integration of code finder and manager clients into
departmental work flow. As shown by dotted connecting lines, we plan to
eventually integrate coding functions with our order entry, online medical
record, and reporting systems.
|
|
Shortly after implementation, quality coders and management personnel
observed that the number of coding errors decreased. More cases were given
appropriate multiple codes. We have considered turning off the system to allow
quantifying changes in coding time, completeness, and accuracy; however, it is
now so ingrained in daily practice that shutting it down would be impractical.
The full set of more than 15,000 codes turned out to be less useful for
routine cases and more helpful as a backup source for unusual cases. The
descriptive nomenclature in the large code set is awkward, many of the
descriptions are abbreviated, and many irrelevant codes are included. In
contradistinction, the abbreviated departmental code list changes dynamically
with ongoing experience, fits our unique practice environment, and uses
familiar nomenclature. In some ways, the codes not included in the
departmental list are as important to its use as those that are.
Discussion
There is no doubt that radiologists will become increasingly involved in
coding their work for the medical record. Radiology is data rich and lends
itself particularly well to the application of information technology.
Correctly coding all patient encounters is of critical importance not only to
meet payer and regulatory requirements but also for effective patient-oriented
clinical research, technology assessment, and organizational planning
[2,
3]. Traditional approaches that
rely on after-the-fact coding by nonphysician personnel result in significant
errors [4]. Active and direct
physician involvement in coding enhances accuracy and completeness
[5,6,7].
There are several controlled medical vocabularies available for encoding
medical diagnoses, and though the ICD-9 system has some shortcomings, it
represents the prevailing standard
[8,
9]. Computerized aids for
finding appropriate ICD-9 codes are certainly not new, and several commercial
versions are available. Ample evidence reveals that computerized coding
assistance allows health care workers to produce significantly more accurate
and complete codes in less time than with manual methods
[5,6,7,
10,11,12,13,14,15,16,17,18].
Hohnloser et al. [10] points
out that physicians forced to code manually may shift as many as 84% of
diagnoses from the appropriate section to other sections, thus avoiding the
need to encode them. This tendency is significantly reduced when they use
computerized methods [15].
In designing computerized coding aids, several features may be useful. One
is using common terminology for code descriptors and frequency-based
presentation of codes [6,
17]. Our solution incorporates
common descriptive terms for codes but does not yet use frequency-based
ordering. Perhaps most useful is a feature that allows users to add or change
text descriptors to suit their own personal preferences
[16,
18]. With over 50 radiologists
empowered to add new entries and expert coders checking additions, the list of
codes in our database grows dynamically and retains its accuracy. We have made
the system available via file transfer protocol
(ftp://webster.xray.ufl.edu/pub/sistromc/) to interested readers for their own
use and modification. The distribution will include a sample departmental code
set with standard nomenclature from the full set (also included) and the code
finder and manager clients.
References
-
ICD-9-CM, 1999: International classification of
diseases, 9th ed., vols. 1 and 2, Dover,
DE: American Medical Association, 1999
-
Elevitch FR. Prospecting for gold in the data mine. Clin
Lab Med 1999;19:373
-384[Medline]
-
Groves WE. Storage and retrieval of coded patient diagnoses and
data on a clinical laboratory computer system. Comput Programs
Biomed 1980;12:225
-229[Medline]
-
Chewning SJ, Nussman DS, Griffo ML, Kiebzak GM. Health care
information processing: how accurate are the data? J South Orthop
Assoc 1997; 6:8
-16[Medline]
-
Gibby GL, Paulus DA, Sirota DJ, et al. Computerized pre-anesthetic
evaluation results in additional abstracted comorbidity diagnoses.
J Clin Monit
1997;13:35
-41[Medline]
-
Deimel D, Hesselschwerdt HJ, Lusznat A. Computer-assisted
classification of ICD-9/10 and IKPM in compliance with the new federal health
care regulation: practice-oriented solution for trauma surgery and orthopedics
[in German]. Unfallchirurg
1995;98:545
-550[Medline]
-
Yarnall KS, Michener JL, Broadhead WE, Hammond WE, Tse CK.
Computer-prompted diagnostic codes. J Fam Pract
1995;40:257
-262[Medline]
-
Chute CG, Cohn SP, Campbell KE, Oliver DE, Campbell JR. The content
coverage of clinical classifications: for The Computer-Based Patient Record
Institute's Work Group on Codes and Structures. J Am Med Inform
Assoc 1996;3:224
-233[Abstract/Free Full Text]
-
Stitt FW. The problem-oriented medical synopsis: coding, indexing,
and classification sub-model. (abstr) Proc Annu Symp Comput Appl
Med Care 1994;1:964
-
Hohnloser JH, Puerner F, Soltanian H. Improving coded data entry by
an electronic patient record system. Methods Inf Med
1996;35:108
-111[Medline]
-
Hohnloser JH, Kadlec P, Puerner F. Coding clinical information:
analysis of clinicians using computerized coding. Methods Inf
Med 1996;35:104
-107[Medline]
-
Hohnloser JH, Puerner F, Soltanian H. Improving clinician's coded
data entry through the use of an electronic patient record system: 3.5 years
experience with a semiautomatic browsing and encoding tool in clinical
routine. Comput Biomed Res 1996;29
: 41-47[Medline]
-
Hohnloser JH, Kadlec P, Puerner F. Experiments in coding clinical
information: an analysis of clinicians using a computerized coding tool.
Comput Biomed Res
1995;28:393
-401[Medline]
-
Hohnloser JH, Purner F, Dusek R, Zitek M. 4.5 years of experience
with an electronic patient record system at the University of Munich: PADS
(patient archiving and documentation system). (abstr)
Medinfo
1995;8:1661
-
Hohnloser JH, Soltanian H. Positive efficiency findings using
computer assisted ICD-encoding: 3.5 years of experience with the computerized
patient record system PADS (patient archiving and documentation system).
(abstr) Proc Annu Symp Comput Appl Med Care
1994;1:965
-966
-
Krall MA, Chin H, Dworkin L, Gabriel K, Wong R. Improving clinician
acceptance and use of computerized documentation of coded diagnosis.
Am J Manag Care
1997;3:597
-601[Medline]
-
Michel PA, Lovis C, Baud R. LUCID: a semi-automated ICD-9 encoding
system. (abstr) Medinfo
1995;8:656
-
Rossi CR, Alberti V, Mancino G, et al. Comparison between manual
and automatic coding of medical record statistical cards at a university
hospital. Med Inform 1993;18
: 53-59
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
|
|
 
C. D. Phillips and B. J. Hillman
Coding and Reimbursement Issues for the Radiologist
Radiology,
July 1, 2001;
220(1):
7 - 11.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
