Transcript QuARK - CNS - Indiana University

Illuminating the Fine Print:
Visualizing Medication Side-Effects
in Complex Multi-drug Regimens
Jon D. Duke, MD
NLM Medical Informatics Fellow
Regenstrief Institute
Indiana University
The QuARK Project
Quantitative Adverse Reaction Knowledgebase
The Tao of QuARK
• The Concept
• Building the
Knowledgebase
• Clinical Applications
• Testing the Model
• Future Directions and
Research
QuARK: Quantitative Adverse Reactions Knowledgebase
Part I:
The Concept
QuARK: What is it good for?
The primary goal of QuARK is to
simplify the process of assessing
adverse drug reactions in patients
taking multiple medications.
QuARK: Quantitative Adverse Reactions Knowledgebase
Polypharmacy
Polypharmacy
Kaufman DW, Kelly JP, Rosenberg L, Anderson TE, Mitchell AA. Recent patterns of medication use in the ambulatory
adult population of the United States: the Slone survey. JAMA 2002;287: 337-44.
Polypharmacy
• Has increased significantly over past 20 years
• Increases risk for adverse drug reactions
• Known risk factor for overall morbidity and
mortality
• Estimated cost $76B annually
1
2
1
3
1.
2.
3.
Hajjar ER, Cafiero AC, Hanlon JT. Polypharmacy in elderly patients. Am J Geriatr Pharmacother 2007;5: 345-51.
Nguyen JK, Fouts MM, Kotabe SE, Lo E. Polypharmacy as a risk factor for adverse drug reactions in geriatric
nursing home residents. Am J Geriatr Pharmacother 2006;4: 36-41.
Tam-McDevitt J. Polypharmacy, Aging, and Cancer: Growing Risks. Oncology 2008;9.
Side-Effect Complexity
Number of Drugs X
SE Complexity
Physician Time
= The Problem
Side-Effects
Interactions
Current Solutions
Goals
• Look up multiple medications simultaneously
• Rapidly get to side-effect of interest
• Show the relative strength of association
between a drug and its side-effects
• Well-integrated into clinical workflow
Origins of QuARK
Hmmmmm….
Zocor
Metformin
Norvasc
Lisinoprol/HCTZ
Azithromycin
Nausea
Dizziness
Edema
Fatigue
Cough
Palpitations
Part II:
Building the Knowledgebase
Which Medications to Include?
• By prescribing volume
• By formulary
QuARK
Wishard Top 500
Clarian Top 500
U.S. Top 300
Coding the Medications
•
•
•
•
RxNorm
UNI
NDC
Regenstrief Dictionary
QuARK
RI Dictionary
RxNorm
Sources of Adverse Reaction Data
•
•
•
•
FDA Label
MedWatch / AERS
Clinical Repository (eg. RMRS)
Social Networks (eg. patientslikeme.com)
QuARK
FDA Label
Coding the Side-Effects
•
•
•
•
•
MedDRA
CTCAE
SNOMED-CT
ICD-9
UMLS CUI
QuARK
MedDRA
UMLS CUI
SNOMED-CT
Which Side-Effects to Include?
Must select a single unique
representation of each
medication / side-effect pair
Which Side-Effect Data to Include?
•
•
•
•
Which treatment indication?
Which dose?
Which trial duration?
Pre- / Post-marketing data?
QuARK
If duplicate data:
Most common
indication
preferred
Aggregate dose
data preferred,
otherwise most
common dose
Larger trials with
longer duration
preferred
Post-marketing
data included if
not present in trials
Side-Effect Quantification
The assignment of a numeric score to represent
the relative frequency at which a particular
medication causes a particular side-effect.
Types of Frequency Data
• Drug vs Placebo
– 34% of Neurontin patients experienced nausea vs 12% of
placebo patients
• Frequency Range
– Between 3% and 9% of patients taking Lipitor experienced
dizziness
• Qualitative Frequency Descriptor
– Diarrhea occurred infrequently in patients taking Lisinopril
• Statement of Occurrence
– Thrombocytopenia was reported in patients taking
Norvasc.
Drug vs Placebo
eg. 34% of Neurontin pts experienced nausea vs 12% of placebo pts
• Optimal data format
• Applied “Absolute Risk
Reduction” approach (ie.
treatment incidence – placebo
incidence)
• ex. Score = 34 - 12 = 22
• Database would include both the
original raw data in addition to
the calculated score
QuARK
Drug vs Placebo
Score =
Treatment
Incidence
Placebo
Incidence
Frequency Range
eg. Between 3% and 9% of Lipitor patients experienced dizziness
• No placebo data given
• Study size and duration not
available
• Patient population unknown
• Conservative score calculation:
Score = x+(y-x)/3 = 3+(9-3)/3 = 5
• Original data range preserved in
database
QuARK
Frequency Range
Scoring
Between X% and
Y% of patients
taking {drug}
experienced
{effect}
Score =
X+(Y-X)/3
Qualitative Frequency Descriptor
eg. Diarrhea occurred infrequently in patients taking Lisinopril
• No placebo or population data
• Wide range of terms used (eg.
rarely, occasionally, often)
• Quantitative mappings may be
provided (Rarely = “< 1/100”)
• Where mappings unavailable,
conservative scores assigned
based on interpretation of terms
(sometimes = occasionally >
infrequently)
QuARK
Qualitative
Scoring
Occasionally 0.75
Infrequently 0.5
Rarely
0.3
Statement of Occurrence
eg. Thrombocytopenia was reported in patients taking Norvasc.
• No frequency information
• No placebo or population data
• Commonly seen with postmarketing reports or class
effects
• Conservative scoring applied
• “Post-Marketing” status noted
in database
QuARK
Occurrence
Scoring
Occurs in drug 0.8
Occurs in class 0.7
Occurs more often
in placebo
0.1
RI Term
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
18175
IngredientName
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Sertraline HCl
Effect
Nausea
Ejaculation Disorders
Diarrhea
Insomnia
Dry Mouth
Nervousness
Somnolence
Tremor
Dizziness
Fatigue
Dyspepsia
Decreased Libido
Anorexia
Flatulence
Paresthesias
Vomiting
Suicidal Ideation
Flushing
Headache
Agitation
Drug
25
14
20
21
14
6
13
8
12
12
8
6
6
3
3
4
4
2
25
5
Placebo Score Code
11
14 CALC
1
13 CALC
10
10 CALC
11
10 CALC
8
6 CALC
0
6 CALC
7
6 CALC
2
6 CALC
7
5 CALC
7
5 CALC
4
4 CALC
2
4 CALC
2
4 CALC
0
3 CALC
0
3 CALC
2
2 CALC
2
2 CALC
0
2 CALC
23
2 CALC
3
2 CALC
QuARK Part III:
Applications
Rxplore
• Interactive visualization of QuARK data
• Allows quick retrieval of most common sideeffects of complex drug regimens
• Highlights potential causal agents in the
setting of an adverse drug event
• Allows “virtual swapping” of a medication to
assess impact on patient’s side-effect profile
QuARK & Gopher
• Goal: Allow QuARK visualizations to be
retrieved directly from Gopher order entry
• Created prototype running on Gopher Dev
• Auto-populates medication list directly from
Gopher patient chart
QuARK Bubble Map
Medication Heat Map
Adverse Effects by Organ System
Diazepam
Highly Affected
Minimally Affected
Dizziness 24% vs 3%
Headache 11% vs 2%
Insomnia 6% vs 3%
QuARK & Clinical Reminders
• Chief Complaint-driven
• Trigger Event-driven
QuARK & Clinical Reminders
• Chief Complaint-driven
– Which of a patient’s medications are associated
with the Chief Complaint?
– At what frequency?
“ The patient’s complaint of Dizziness has been
associated with use of:
Gabapentin (28% vs. 7% Placebo)
Atenolol (13% vs 6% Placebo)
Omeprazole (Less than 1%) ”
QuARK & Clinical Reminders
• Trigger Event-driven
– Laboratory / EKG change generates reminder
– Offers suggestions for possible causal agents
“ Neutropenia (WBC 1.4 10/22/08) has been
associated with use of:
Valsartan (1.9% vs 0.8% placebo)
Amiodarone (Has been reported)
Lisinopril (Occurs rarely) ”
QuARK Part IV:
Testing the Model
Garbage In / Garbage Out?
•
•
•
•
Limitations of the Data
Algorithmic Considerations
Does a Gold Standard exist?
An Approach to Validation
Comparison with AERS
•
•
•
•
Adverse Event Reporting System
Captures over 400,000 reports a year
Allows for listing of multiple medications
Records Adverse Reaction and Suspected
Cause
• Subset includes “Dechallenge” Data
QuARK vs AERS
• Dechallenge data from 2008 Q2
• Evaluated reports of four common reactions
(nausea, edema, insomnia, hyponatremia )
• Limited to cases where patient was taking at
least 5 medications
• Compared the QuARK “suspected drug” with
actual reported cause
Accuracy of QuARK Ranking for AERS Reports Q2 2008
90
80
% Accuracy
70
60
50
Exact Match
40
Top Two
30
20
10
0
Nausea
n=31
Edema
n=21
Insomnia
n=14
Reported Adverse Reaction
Hyponatremia
n=25
Sources of Error
• <10% missed cases due to algorithm error
• >90% missed cases due to complete absence
of the adverse reaction from the drug label
• Delays in drug label updating
• Reflects nature of adverse event reporting
– Known side-effects often not reported
– New drug mandatory reporting predominates
AERS
QuARK Part V:
Future Directions and Research
Evaluation Studies
• Laboratory study of “decision velocity”
• Survey of User Satisfaction / Efficiency
Clinical Reminder Study
• Generate QuARK-based reminders for
laboratory triggers (eg. LFT’s)
• Intervention group receives reminder noting
potential causal agents / frequency data
• Compare drug discontinuation rates as well as
time between trigger and discontinuation
Build a Better QuARK
•
•
•
•
•
Additional medications
Expansion of AERS-QuARK analysis
Optimization of scoring algorithm
Additional visualization methods
Potential use in consumer health
Summary
• QuARK is a knowledgebase containing
quantitative frequency data for adverse drug
reactions
• Potential applications include:
– visualization of side-effect data
– simplified lookup of multidrug regimens
– clinical reminders targeted at adverse drug events
• Opportunities for research collaboration
Thanks!
NLM, Steve Downs, Mike McCoy, Marc
Overhage, Shaun Grannis, Gunther Schadow,
Siu Hui, Martin Were, Marc Rosenmann, Linas
Simonatis, Atif Zafar, Paul Dexter, Mike Weiner,
Paul Biondich, Burke Mamlin, Anne Belsito
Pop the QuARK!