Transcript Medication Reconcilliation Using Natural Language

Computer-based Support for Improving
Patient Medication Management
James J. Cimino
Chief, Laboratory for Informatics Development
National Institutes of Health Clinical Center
Senior Scientist, Lister Hill Center for Biomedical Communications
National Library of Medicine
Informatics Grand Rounds
Dartmouth-Hitchcock Medical Center
May 16, 2008
Challenges to Medication Management
• Lack of information about the patient
– Patient’s condition
– Patient’s co-morbidities
– Medications the patient is supposed to take
– Medications the patient is actually taking
• Access to medical knowledge
– Knowing about availability of knowledge resources
– Knowing how to use knowledge resources
– Effort to use knowledge resources
Solutions
• Medication reconciliation
– Collect information from disparate sources
– Present information to support decision making
• Infobuttons
– Anticipate user’s information needs
– Automate access to appropriate resources
– Automate retrieval from these resources
The Challenge of Medication Reconciliation
Go
Stop
Stop
Stop
Stop
Stop
Stop
?
Go
Go
Many a Slip ‘Twixt the Cup and the Lip
Stop
Stop
Stop
Stop
Patient is
Supposed to Take
Patient is
Taking
Patient is not
Taking
Patient is not
Supposed to Take
Reports
Taking
Doesn’t
Report Taking
Reports
Taking
Doesn’t
Report Taking
Reports
Taking
Doesn’t
Report Taking
Reports
Taking
Doesn’t
Report Taking
Problems and Solutions
• Errors due to:
–
–
–
–
Not starting medications the patient should be taking
Starting medications the patient shouldn’t be taking
Not communication starts/stops to next caregiver
Not communicating changes to patients
• Beers, et al. J Am Geriatric Society 1990:
– 83% of hospital admission histories missed one or
more medications
– 46% missed three or more
• Problems occur at all transitions in care:
– “Continue all outpatient medications”
Electronic Health Records to the Rescue!
Go
Stop
Stop
Stop
Stop
Stop
Stop
?
Go
Go
Computer Assisted Medication Reconciliation
• Poon et al.: JAMIA 2006:
– Preadmission Medication List
– Grouped medications by generic names
•
•
•
•
•
Text sources
Multiple sources
Substitutions might occur
Confusing chronology
Information overload!
Our Approach to Medication Reconciliation
• Multiple inpatient and outpatient systems
• Natural language processing to get codes
• Medical knowledge base to group codes
• Chronological presentation
Methods
• All recent admissions for one physician (JJC)
• Multiple inpatient and outpatient resources
• Carol Friedman’s Medical Language Extraction and
Encoding (MedLEE)
• US National Library of Medicine’s Unified Medical
Language System (UMLS)
• Columbia’s Medical Entities Dictionary (MED)
• American Hospital Formulary Service (AHFS)
classification
• Evaluation of ability to capture, code and organize
Data Sources
Data Source
1. Prior Clinic Note
2. Prior Outpatient Medications
3. Admission Note
4. Admission Note Plan
5. Admission Orders
6. Admission Pharmacy Orders
7. Active Orders at Discharge
8. Discharge Pharmacy Orders
9. Discharge Instructions
10. Discharge Plan
11. Clinic Note after Discharge
12. Outpatient Medications after Discharge
System Data Type
WebCIS Narrative
Coded
WebCIS
WebCIS Narrative
WebCIS Narrative
Coded
Eclipsys
Coded
WebCIS
Coded
Eclipsys
Coded
WebCIS
Eclipsys Narrative
WebCIS Narrative
WebCIS Narrative
Coded
WebCIS
Results
• 70 patient records reviewed
• 30 hospitalizations identified
• 17 met inclusion criteria
• MedLEE found 623/653 (95.4%) medications
• Total of 1533 medications (444 unique) in MED
Medications by Source
Prior Clinic Note *
Prior Outpatient Medications
Admission Note *
Admission Note Plan *
Admission Orders
Admission Pharmacy Orders
157
211
102
41
88
152
Records
with Data
17
13
14
12
8
14
Active Orders at Discharge
93
8
11.6
Discharge Pharmacy Orders
Discharge Instructions *
Discharge Plan *
Clinic Note After Discharge *
171
60
123
140
14
7
16
16
12.2
8.6
7.7
8.8
Outpatient Medications after Discharge
225
13
17.3
Data Source
* Narrative text
Meds
Meds per
Patient
9.2
16.2
7.3
3.4
11.0
10.9
MedLEE Terms Found
48 Other Meds
(8%)
30 Non-Med,
(5%)
545 UMLS
(87%)
MED Terms
16 non-AHFS
(1.0%)
1517 AHFS
(99.0%)
Mapped to UMLS
8 Other Meds
(4%):
INH, MVI, asa,
Os-Cal,
darvocet, hctz,
niacin, toprol
4 Non-Med
(3%): cream,
antiinflammatory, lotion,
lozenge, po
169 UMLS
(93%)
Mapped to AHFS
2 non-AHFS
(0.5%):
oxygen,
medication
442 AHFS
(99.5%)
Transition from Outpatient to Inpatient
Patient #9
201204:
Anticoagulants
240400:
Cardiac
Drugs
240800:
Hypotensive Agents
Prior Clinic Note
coumadin
verapamil
cozaar
Prior Outpatient
Medications
Coumadin
5 mg Tab
Verapamil
180 mg
Extended
Release
Tablet
Losartan
Potassium
100 mg
Tablet
Admission Note
coumadin
verapamil
cozaar
Admission Note
Plan
coumadin
Verapamil
SR Oral
240 MG
Losartan Oral
50 MG
VERAPAMIL
SR TAB
240 MG
LOSARTAN
POTASSIUM
TAB 50
MG
Admission Orders
Warfarin
Sodium
Oral 10
MG
Admission
Pharmacy Orders
WARFARIN
TAB 5 MG
10
MILLIGRA
M
280000:
CNS
Agents
281604:
Antidepressants
cymbalta
Pregabalin
50mg Capsule
cymbalta
Transition from Outpatient to Inpatient
Patient #9
201204:
Anticoagulants
240400:
Cardiac
Drugs
240800:
Hypotensive Agents
Admission
Pharmacy Orders
WARFARIN
TAB 5 MG 10
MILLIGRAM
VERAPAMIL
SR TAB 240
MG
LOSARTAN
POTASSIUM
TAB 50 MG
Active Orders
at Discharge
Verapamil
SR Oral
240 MG
Losartan Oral
50 MG
Discharge
Pharmacy Orders
VERAPAMIL
SR TAB
240 MG
LOSARTAN
POTASSIUM
TAB 50 MG
280000:
CNS
Agents
281604:
Antidepressants
DULOXETINE CAP
20 MG
Discharge
Instructions
cymbalta
Discharge Plan
cymbalta
Clinic Note After
Discharge
Outpatient
Medications after
Discharge
coumadin
verapamil
Coumadin 5
mg Tab
Verapamil
180 mg Extended
Release Tab
cymbalta
Losartan
Potassium 100
mg Tablet
Pregabalin
50mg
Capsule
Discussion
• Data from multiple coded and narrative sources
can be coded automatically and merged into a
single form
• The UMLS and MED are both needed for coding to
a single terminology (AHFS)
• Further work on MedLEE and the MED are needed
• Drugs tend to group into one per class; allows for
change from one generic to another
• Chronology by drug class can highlight changes in
medication plans
• Changes can be intended or unintended, but
should not be ignored
• The next step is medication reconciliation
http://www.dbmi.columbia.edu/cimino/medrec/
Next Step: High-Quality Decision Making
• Providing patient information evokes additional
information needs
• These needs are stereotypical
• Resources exist to address these needs
• If we can predict the needs, we can provide links
• Information available in the context can be used to
target the resources
Health Knowledge for Decision Support
Health Knowledge for Decision Support
?
Infobuttons
Anticipate
Need and
Provide
Queries
i
Information Needs of CIS Users
• Common tasks may have common needs
• System knows:
– Who the user is
– Who the patient is
– What the user is doing
– What information the user is looking at
• We can predict the specific need
• User is sitting at a computer!
• We can automate information retrieval
First Attempt: The Medline Button
•
•
•
•
CIS on mainframe
BRS/Colleague (Medline) on same mainframe
Get them to talk to each other
Search using diagnoses and procedures
Admission Profile
----------------------------------------------------------------------------Admission Date: 01/03/95 Discharge Date: 02/16/95
Location:
Doctor: CIMINO, JAMES J
Discharge Summary: N
Primary Diagnosis: 410.71
ACUTE MI,SUBENDO INFARC, INITI
M6GS
Select Terms You Are Interested in:
X
X
_
410.71
780.3
507.0
426.0
415.1
453.8
428.0
Diseases:
ACUTE MI, SUBENDO INFARC, INITI
CONVULSIONS
FOOD/VOMIT PNEUMONITIS
ATRIOVENT BLOCK COMPLETE
PULMON EMBOLISM/INFARCT
VENOUS THROMBOSIS NEC
CONGESTIVE HEART FAILURE
F8 = for more information
----------------------------------------------------------------------------Help=F1
Search MEDLINE=ENTER
MEDLINE Queries from Admission Profile
----------------------------------------------------------------------------Select a question:
_
1.
Does Myocardial Infarction cause Convulsions?
2.
Is Myocardial Infarction caused by Convulsions?
3.
Does Myocardial Infarction occur with Convulsions?
----------------------------------------------------------------------------Help=F1
Search MEDLINE=ENTER
BRS Query from Admission Profile
----------------------------------------------------------------------------(Myocardial Infarction WITH (ET OR SC)) AND (Convulsions WITH CO)
----------------------------------------------------------------------------Help=F1
MEDLINE Queries=ENTER/F3
Session
Edit
Commands
Options
Help
^F File ^E Edit ^A Search ^L Limit ^V View
1
^T Tools
^O Options
Myocardial infarction/et,sc and convulsions/co
Ovid – Medline <1973-1983>
[To select option hold Ctrl and letter indicated.
Enter subject, then press <Enter>
_: _
Press ^Y for Help.]
^Y Help
1
First Attempt: The Medline Button
•
•
•
•
•
•
CIS on mainframe
BRS/Colleague (Medline) on same mainframe
Get them to talk to each other
Search using diagnoses and procedures
Technical success
Practical failure
Education at the Moment of Need
i
Education at the Moment of Need
i
1
Understand
Information
Needs
Education at the Moment of Need
2
Get Information
From EMR
i
1
Understand
Information
Needs
Education at the Moment of Need
2
Get Information
From EMR
i
1
Understand
Information
Needs
3
Resource
Selection
Education at the Moment of Need
4
2
Get Information
From EMR
Resource
Terminology
i
1
Understand
Information
Needs
3
Resource
Selection
Education at the Moment of Need
4
Resource
Terminology
5
2
Automated
Translation
Get Information
From EMR
i
1
Understand
Information
Needs
3
Resource
Selection
Education at the Moment of Need
4
Resource
Terminology
6
Querying
5
2
Automated
Translation
Get Information
From EMR
i
1
Understand
Information
Needs
3
Resource
Selection
Education at the Moment of Need
4
Resource
Terminology
5
2
Automated
Translation
Get Information
From EMR
i
6
Querying
1
Understand
Information
Needs
3
Resource
Selection
7
Presentation
Infobuttons vs. Infobutton Manager
Resource s
Clinical System
Infobutton
Query
Knowledge
Base
Context
Infobutton
Manager
Page
of
Hyperlinks
Au
g0
N 2
ov
-0
Fe 2
bM 03
ay
Au 03
g0
N 3
ov
-0
Fe 3
bM 04
ay
Au 04
g0
N 4
ov
-0
Fe 4
bM 05
ay
Au 05
g0
N 5
ov
-0
Fe 5
bM 06
ay
Au 06
g0
N 6
ov
-0
Fe 6
bM 07
ay
Au 07
g07
Usage in Lab Contexts
2500
2000
1500
HR-LabDetail
IM-LabDetail
1000
500
0
Aug-07
May-07
Feb-07
Nov-06
Aug-06
May-06
Feb-06
Nov-05
Aug-05
May-05
Feb-05
Nov-04
Aug-04
May-04
Feb-04
Nov-03
Aug-03
May-03
Feb-03
Nov-02
Aug-02
Usage in In-Patient Drug Contexts
700
600
500
400
HR-InPatientDrugs
IM-InPatientDrugs
300
200
100
0
Aug-07
May-07
Feb-07
Nov-06
Aug-06
May-06
Feb-06
Nov-05
Aug-05
May-05
Feb-05
Nov-04
Aug-04
May-04
Feb-04
Nov-03
Aug-03
May-03
Feb-03
Nov-02
Aug-02
Usage in Diagnosis Context
900
800
700
600
500
HR-Diagnoses
400
IM-Diagnoses
300
200
100
0
Au
g0
N 2
ov
-0
Fe 2
bM 03
ay
Au 03
g0
N 3
ov
-0
Fe 3
bM 04
ay
Au 04
g0
N 4
ov
-0
Fe 4
bM 05
ay
Au 05
g0
N 5
ov
-0
Fe 5
b0
M 6
ay
Au 06
g0
N 6
ov
-0
Fe 6
bM 07
ay
Au 07
g07
Usage in Lab Order Entry Context
140
120
100
80
HR-LabOrder
60
IM-LabOrder
40
20
0
Aug-07
May-07
Feb-07
Nov-06
Aug-06
May-06
Feb-06
Nov-05
Aug-05
May-05
Feb-05
Nov-04
Aug-04
May-04
Feb-04
Nov-03
Aug-03
May-03
Feb-03
Nov-02
Aug-02
Usage in InPat Drug Order Entry
350
300
250
200
HR-DrugOrder
IM-DrugOrder
150
100
50
0
The Coumadin Story
• Chair of Medicine wants link to Coumadin
protocol
• First, I have to find the guidelines
The Coumadin Story
• Chair of Medicine wants link to Coumadin
protocol
• First, I have to find the guidelines
• Then I have to add the question to the IM table
The Coumadin Story
• Chair of Medicine wants link to Coumadin
protocol
• First, I have to find the guidelines
• Then I have to add the question to the IM table
• Finally, I link the question to the context
The Coumadin Story
• Chair of Medicine wants link to Coumadin
protocol
• First, I have to find the guidelines
• Then I have to add the question to the IM table
• Finally, I link the question to the context
• Voilá!
New York Presbyterian Hospital (Eclipsys)
NY Office of Mental Health (Psykes)
NY Office of Mental Health (Psykes)
Regenstrief Medical Record System
Cryststal Run Healthcare (NextGen)
AMIA 2007 Demo Participants
• Health care & academic institutions
– Intermountain Healthcare, Columbia University,
Partners Healthcare
• Content providers
– Wolters Kluwer Health, ACP, Micromedex,
UpToDate, Ebsco, Lexicomp
Institution-Specific Requirements
•
•
•
•
•
What are the users’ information needs?
In what contexts do those needs arise?
What resources will resolve the needs?
How do we deal with terminology?
How can the Infobutton Manager be
integrated into the clinical information
system?
• The institution’s librarian is the best person
to resolve most of these issues
Institution Customization Tasks
Infobutton Manager
Infobutton Manager
Maintenance Tool
System
Maintainer
Functions:
Browse
Add
Update
Delete
Translation
Table
Term
Translation
Context
Table
Context
Matching
Infobutton
Table
Query
Construction
Clinician
Page of
Links
Librarian Infobutton Tailoring Environrment (LITE)
•
•
•
•
•
Specify user contexts
Identify terminology in each context
Information needs in each context
Resources for resolving information needs
Automating translation and querying
Institution Customization Tasks
Infobutton Manager
Infobutton Manager
Maintenance Tool
System
Maintainer
Functions:
Browse
Add
Update
Delete
Translation
Table
Term
Translation
Context
Table
Context
Matching
Infobutton
Table
Query
Construction
Clinician
Page of
Links
LITE Tasks
Librarian Infobutton
Tailoring Environment
(LITE)
Context
Definition
Terminology
Specification
Institution
Librarian
Question
Construction
Infobutton Manager
Translation
Table
Term
Translation
Context
Table
Context
Matching
Infobutton
Table
Query
Construction
Page of
Links
Resource
Selection
Resource
Utilization
Infobutton
Manager
Log File
LITE Auditing
Infobutton Manager
Monitoring
LITE Monitoring
Clinician
LITE
Log File
LITE Research Plan
•
•
•
•
•
•
•
•
•
Conduct community assessment
Refine LITE features
Establish forum for feedback from librarians
Develop LITE in an iterative manner
Develop a user manual and tutorial
Evaluate usability of LITE by librarians
Evaluate the use of LITE
Disseminate the results of the project
Promote the use of the IM and LITE
Status Report
•
•
•
•
•
•
•
Drupal site
Community of users
Clear through Institutional Review Board
Enroll “subjects”
Make each draft a forum topic
Collect feedback
Iterate
www.infobuttons.org
lite.dbmi.columbia.edu
Conclusions
• Diverse medication data can be automatically
integrated
• Organizing data by time and drug class can
highlight possible errors
• Infobuttons can anticipate and resolve clinicians’
information needs
• Institution-specific tailoring is required
• International standard will stimulate wider adoption
• Librarian Infobutton Tailoring Environment will put
the Infobutton Manager on autopilot
Acknowledgments
• Medication Reconcilliation
– Carol Friedman for use of MedLEE
– Jianhua Li for programming
– Tiffani Bright for background research
– US National Library of Medicine
• Infobuttons
– Jianhua Li for programming
– Many student contributors
– Guilherme Del Fiol
– Noemie Elhadad
– National Library of Medicine