Managing Vocabulary for a Centralized Clinical System

1
Managing Vocabulary for a Centralized Clinical
System

• James J. Cimino, MD Stephen B. Johnson, PhD
• George Hripcsak, MD Claire L Hill, MA
• Paul D. Clayton, PhD
• Department of Medical Informatics - Columbia
  University
• New York, New York, USA

2
Managing Vocabulary for a Centralized Clinical
System

• Automated Knowledge-Based Techniques
• Putting Theory into Practice

3
Distributed Systems
4
Central Clinical Repository(Fantasy)
5
Central Clinical Repository(Reality)
6
Central Controlled Vocabulary
7
Vocabulary Construction Issues

• Understanding
• Modeling
• Creation
• Maintenance

8
The Theory

• "A knowledge-based approach to vocabulary
  representation will improve maintenance and
  utility."

9
Medical Entities Dictionary (MED)

• Semantic network of concepts
• Multiple hierarchies
• Frame-based concept representation
• 46,000 concepts

10
MED Structure
Medical Entity
Substance
Laboratory Specimen
Event
Plasma Specimen
Anatomic Substance
Chemical
Diagnostic Procedure
Substance Sampled
Laboratory Test
Laboratory Procedure
Plasma
Has Specimen
Bioactive Substance
Carbo- hydrate
CHEM-7
Plasma Glucose
Part of
Glucose
Substance Measured
11
Retrieving Results Individually
K1
K2
K3
12
Retrieving Results by Class
K
K1
K2
K3
13
Maintenance Tasks

• New Vocabularies (Laboratory)
• Changing Vocabularies (Pharmacy)

14
New Vocabulary Laboratory

• Original lab 2533 terms
• New lab 5291 terms
• Vocabulary delivered June 15, 1994
• Go live date July 24, 1994

15
Changing Vocabulary Pharmacy

• Started with 2091 drugs
• In two years, added 2327 drugs
• Classification by
• Ingredients
• AHFS Class
• Allergy
• DEA
• Form

16
Adding New Terms

• Identify redundant terms
• Put new terms into existing classes
• Create new classes where appropriate

17
Put Terms into Existing Classes

• Theory The attributes of new terms can be used
  to identify classes
• Practice "Pushing" Terms

18
Pushing a Term
Medical Entity
Chemical
Laboratory Test
Carbo- hydrate
Bioactive Substance
Stat Glucose Test
Chemistry Test
Plasma Glucose Test
Glucose
Stat Glucose Test
Chem-7 Glucose Test
Chem-20 Glucose Test
Stat Glucose Test
19
Create New Classes

• Theory Attribute patterns can be detected which
  identify potential classes
• Practice Recursive partitioning of existing
  classes

20
Finding a New Class
Medical Entity
Medical Entity
Laboratory Test
Laboratory Test
Chemical
Chemical
Chemistry Test
Chemistry Test
Antigen
Antigen
Core Antigen
HBC
Hepatitis B Core Antigen Test
Hepatitis B Core Antigen
Hepatitis B Core Antigen
Core Antigen
HBC
21
screen shot MED Editor
22
screen shot MED Editor proposing a new class
23
Semi-Automated Maintenance

• Read formulary file
• Identify new drugs
• Link new drug to ingredient(s)
• Suggest classifying in preparation class
• Add new drug as per human reviewer

24
Interactive Classification
Adding "LASIX 20MG TAB" Generic Ingredient
"FUROSEMIDE" AHFS Class "DIURETICS" Add to
"FUROSEMIDE PREPARATION"? y Adding
"ZAROXOLYN 5MG CAP" Generic Ingredient
"METOLAZONE" AHFS Class "DIURETICS" Add to
"DIURETICS"? n Create
METOLAZONE PREPARATION" Class? y
25
Automated Classification
Medical Entity
Allergy Class
Chemical
Drug
Sulfa Allergy "S1"
Trimethoprim Allergy "65"
Antibiotic
Pharmacologic Substance
Trimethoprim/ Sulfamethoxizole Preparations
Sulfameth- oxizole
Trimeth- oprim
Bactrim "S1", "65"
Septra "S1"
26
Formulary Correction Statistics

• Among original 2091 drugs
• 334 unclassified drugs assigned classes
• 289 drugs assigned multiple classes
• 173 drugs discovered to be missing allergy codes
• Among additional 2327 drugs added
• 28 unclassified drugs assigned classes
• 141 drugs assigned multiple classes
• 57 drugs discovered to be missing allergy codes

27
Future DirectionsWeb-Browser

• Platform independence
• Available everywhere

28
screen shot Web
29
Future DirectionsX-Based Browser/Editor

• Runs directly off vocabulary server
• Multi-user environment
• Ready for use in real world

30
screen shot Accessmed
31
Future DirectionsK-Rep

• IBM product
• Knowledge-based approch built in
• Automated term subsumption
• Moving from research to real world

32
screen shot K-Rep
33
Impact of "Theory into Practice"Better
management

• Easier to merge new vocabularies
• Easier to automate change management
• Higher quality through better modeling

34
Impact of Better ManagementMore Useful
Vocabulary

• MED is up-to-date for ancillary systems
• Easier to find terms in the MED
• Support for multiple conceptual levels
• More accurate database queries

35
See for Yourself

• http//www.cpmc.columbia.edu/homepages/ciminoj