Crotty engaging patients in new ways from open notes to social media
Integrating evidence based medicine and em rs
1. Integrating Evidence-Based Medicine
and EHRs:
“Nextgen” Clinical Decision Support for
Genomically-enabled Healthcare
Daniel Masys, MD
Affiliate Professor
Biomedical & Health Informatics
University of Washington, Seattle
CMIO Leadership Forum
October 5, 2012
2. Topics
• The goal
• The challenges to be addressed on the way to
the goal
• Partial progress toward a decision-support
infrastructure for ‘precision healthcare’
• Currently unmet needs – a plea for your help
3. A deceptively simple goal
• Do the right thing, and only the right
thing, and do it every time for every
individual
–Therapeutic subset: right dose of the
right drug for the right patient at the
right time
4. The baseline:
The gap between what we know
and what we do
6,712 Individuals in 12 Cities
Only 54.9% received recommended care
Only 54.9% received recommended preventive care
Only 53.5% received recommended acute care
Only 56.1% received recommended chronic care
Examples: Hip Fracture 22.8% (Range 6.2-39.5%)
Atrial Fibrillation 24.7%
Depression 57.2%
Senile Cataract 78.7% (Best performance)
McGlynn, et. al., NEJM 2003;348:2635-45
5. Rising expectations
that systems should work
• IOM Reports
– 14 reports now in the Quality Chasm series
– Current mantra: “Learning Healthcare Systems”
• Persistent media attention
7. Process Errors
• Majority of errors do not result from individual
recklessness, but from flaws in health system
organization (or lack of organization).
• Failures of information management are common:
– illegible writing in medical records
– lack of integration of clinical information systems
– inaccessibility of records
– lack of automated allergy and drug interaction checking
9. If all of these studies on healthcare
quality are correct, the problems are
too big to name, and would not
leave the message-giver intact
Don Berwick
(Outgoing)
Administrator, Centers for
Medicare & Medicaid Services
10.
11. If airlines were run like healthcare...
• Pilots would build and maintain their own airliners
• Navigation instruments would be available but not used
(“just another case of Denver…”)
• There would be no ‘second pair of eyes’ (ATC) watching
each flight’s progress
• Ticket = seat only. Pilots would separately bill for piloting
services two months after the flight
• There would be no National Aerospace Course Guidance
infrastructure embedded in GPS’s and onboard navigation
computers linked to autopilots
• Pilots would leave the plane before all accidents
• The transportation system would be no safer or consistent
than any single practitioner participating in it
12. Systems Design Issues
in Healthcare
• Current practice largely
depends upon the clinical
decision making capacity and
reliability of autonomous
individual practitioners, for
classes of problems that
routinely exceed the bounds
of unaided human cognition
Masys DR. Effects Of Current And Future
Information Technologies On The Health
Care Workforce. Health Affairs, 2002 Sept-Oct;
21(5):33-41.
13. Why?
• In the absence of facts, opinion prevails
(85% of healthcare)
- T. Clemmer, M.D.
• “A Thousand Doctors, A Thousand Opinions”
- French proverb
• “Instead of teaching doctors to be intelligent map
readers, we have tried to teach every one to be a
cartographer.”
- L. Weed, M.D.
• “We need ‘just in time’ education. In medical school we
teach ‘just in case’ education.
- William Stead, M.D.
14. Inescapable Conclusion
• Health care in the 20th Clinical
Century and before was Events
hopelessly bound by reliance
on imprecise phenotypic
manifestations of disease and
a thousand year old Guild and
Apprenticeship model of
Molecular
medical reasoning and Events
education
15.
16. The Genome Sequence
is at hand…so?
“The good news is that we have the human genome.
The bad news is it’s just a parts list”
17. The Promise
• Molecular and clinical biomarkers for health
conditions individuals either have or are
susceptible to
• Includes traditional healthcare history, physical
findings, diagnostic imaging, standard clinical laboratories
• Increasingly: large volumes of molecular data
– Structural genomics: DNA in residence (~22,000 genes)
– Functional genomics: genes switched on (1-2% active)
– Proteomics (400,000 proteins from 22,000 genes)
18. The Promise, cont’d
• Precision Health Care
• Pharmacogenomics
– “The right dose of the
right drug for the right
patient at the right time”
– Drug development:
• Avoid drugs likely to cause
side effects
• Re-investigate “back-
burner” drugs
• Develop entirely new
drugs targeting
fundamental disease
processes "Here's my sequence...”
New Yorker, 2000
20. The need for patient-specific clinical decision support
in the era of precision medicine
1000
Facts per Decision
Proteomics and other
effector molecules
100
Functional Genetics:
Gene expression
profiles
10 Structural Genetics:
e.g. SNPs, haplotypes
Decisions by
clinical Human Cognitive
phenotype
Capacity
i.e., traditional
health care
1990 2000 2010 2020
21. A three step approach to managing
and using personal genomic data for
decision support
Step 1:
Get the data into Electronic Health
Records (EHRs) in a usable form
22. Most common current method for delivery of DNA analysis
into clinical operations
23. Problems with treating genomic
analysis in same fashion as other
professionally interpreted clinical data
• Lossy compression: many DNA features observed, only
a few clinically relevant reported, remainder discarded
• Interpretation inextricably bound together with
primary observations in a document format
• Document reporting format not amenable to parsing
for automated machine interpretation and decision
support
• Much more unknown than known about genomic
effects, and science changing rapidly
24. Output of workshop on “Integration of Genetic Test Results into Electronic Medical
Records” convened by the National Heart Lung and Blood Institute, Bethesda, MD
August 2-3, 2011
25. 7 desiderata for molecular variation
data in EHRs
1. Lossless data compression from (high volume) primary
observations to clinically relevant subsets.
2. Since methods will change, molecular lab results carry
observation methods with them (LOINC model)
3. Compact representation of clinically actionable subsets for
optimal performance (clinician thinkspeed = 250msec)
4. Simultaneously support for human-viewable formats (with
links to interpretation) and formats interpretable by decision
support rules.
5. Separate primary sequence data (remain true if accurate)
from clinical interpretations of them (will change with rapidly
changing science)
6. Anticipate the boundless creativity of Nature: multiple
somatic genomes, multiple germline genomes for each
individual over their lifetime.
7. Support both individual care and discovery science
26. Structured keywords for clinical decision Interpretive
Intrepretations support (e.g., C*2*2CLM fires decision codes
of primary data rule for CYP2C19*2 homozygotes at time
of clopidigrel prescribing). Diagnostic
(expect rapid
change) A few tens of bytes each. Interpretations
(PDF reports). A few
kilobytes each.
Layered classes of Personal molecular differences
represented in EHR as computed
EHR-relevant data offset from a Clinical Standard
Reference Genome (CSRG)
=~1% of genome/ proteome.
Primary
A few megabytes.
Observations.
If
accurate, keep Consensus full personal germline and somatic sequence(s)
forever
and metadata: a few gigabytes each
30x+ nextgen reads:
hundreds of gigabytes/few terabytes
27. A three step approach to managing
and using personal genomic data for
decision support
Step 2:
Create a people and technology
infrastructure to use the data for
decision support
28. Example of rule-based clinical decision support (CDS) today
(version 2.0 CDS: present problem along with solution to problem)
29. Chem7 Panel (BUN, Creat, Lytes, Gluc)
Effect of
New orders/day decision
support:
show
provider
most
recent
value of
same
Stopped orders/day test
Neilson, EG, et al Ann Intern Med. 2004; 141: 196-204
30. Effect of Computerized Provider Order Entry (CPOE) with CDS
at Vanderbilt
35
30.1
30
Errors per 100 Orders
25
20
15
10
6.8
5
2.2 1.3 0.2 0.1
0
pre-CPOE post-CPOE
Potential ADE's Medication Prescribing Errors Rule Violations
Potts, A. et al. PEDIATRICS 2004;:113:59-63
31. An example of progress towards
operational genomically enabled
clinical decision support
32. Vanderbilt PREDICT project
Pharmacogenomic Resource for Enhanced Decisions In Care and Treatment.
Go-live date: September 2010
Replicate
literature DNA
association Evidence review
Guidance:
effect in by P&T sub- implementation
Professional
local committee
societies, FDA
biobank
• Drug-genotype
Prospective pair in EMR Follow outcomes
Genotyping: • Other genotypes • Is dose changed?
(e.g. Illumina outside EMR • Are outcomes affected?
ADME panel) • Point of care • What do patients think?
decision support
Pulley JM et al. Operational Implementation of Prospective
Genotyping for Personalized Medicine: The Design of the Vanderbilt
PREDICT Project. Clin Pharmacol Ther. 2012 May 16
33. As seen by providers at the moment of
prescribing:
38. A three step approach to managing
and using personal genomic data for
decision support
Step 3:
Scale the decision support up to
enable all providers and all patients
and families to benefit, via a public
information infrastructure
41. A Systems Approach to Scaling-up
• A National Healthcare Course Guidance
infrastructure (analogous to FAA course guidance
database for aviation)
1. A continuously updated Public Library of clinical
decision support ‘packages’ (a federally supported
information commons): Wikipedia for decision
support.
2. Event monitors embedded in EHR systems:
healthcare autopilots and “guardrails”
3. System-generated alerts at the “teachable moment”
of diagnostic testing and therapy ordering
4. Automated tracking of outcomes vs. provider
decisions: a learning healthcare system
42. Your help needed:
Currently Unmet Infrastructure Needs
• Standards for electronic “decision support packages” containing:
1. Recognition logic for conditions of interest as represented in EHR
systems
2. Guidance for target users (clinician, patient, family)
3. Recognition logic for “closed loop decision support”: process or
outcome measure to monitor, along with record of whether clinician
accepted or rejected guidance
• A Decision Support Public Library of clinical decision support
‘packages’ representing best practice
• Decision support authoring systems to enable local ‘best practice
rules committees’ to easily import, review, and implement
decision support packages received from the Decision Support
Public Library
43. It Takes a Village…
Thanks to:
• Vanderbilt PREDICT and • NHLBI EHR technical
EHR team desiderata co-authors
– Dan Roden, MD – Gail Jarvik, MD, PhD,
– Josh Denny, MD, MS Nick Anderson PhD, Neil
– Jim Jirjis, MD Abernethy PhD – UW
– Kevin Johnson, MD, MS – Isaac Kohane, Harvard
– Jill Pulley, MBA – Marc Hoffman, Cerner
– Dana Crawford, PhD – Howard Levy, Johns
Hopkins
– Dina Paltoo, George
Papanicolau, NIH