Transcript Case Study: As Is Model - National Center for Ontological Research

Peter L. Elkin, MD, MACP, FACMI, FNYAM
Professor and Chair
Department of Biomedical Informatics
Professor and Vice-Chair for Quality and Patient Safety
Department of Internal Medicine
UB Jacobs School of Medicine and Biomedical Sciences
University at Buffalo
State University of New York

Bill Kneivel is a 49 w male on vacation with his
family in Florida. Driving conditions are wet and
slippery and after being cut off his car impacts
an embankment. Unconscious he is taken to
FGH ED. He routinely gets his care at St.
Elsewhere in Boston MA. Records are requested
by FGH and text records are sent which show
Mr. K to have a PMHx of hypertension and Type
II DM. He is currently taking Lisinopril, HCTZ,
Metformin, and ASA.

In ED he is found to have an MI with a
Troponin-T of 0.42, his CBC was wnl,
he had a creatinine of 2.7, normal
electrolytes and a glucose of 187,
Beta-Hydroxybuterate was not
elevated, and his Bicarb was 24. He
was found to have ketones in his
urine.

Mr. K is continued on his present
medication. A beta-blocker is
started. However, two days later his
serum K+ was 7.2 and pts
Bicarbonate was 14. Mr. Kneivel’s
rhythm degenerates into ventricular
tachycardia, followed by a cardiac
arrest and in spite of an aggressive
resuscitation attempt he passes
away.

Data from St. Elsewhere was transmitted
to FGH in a standards-based codified
interoperable format. The Dr. Overworked
(admitting physician) uses an order entry
system that provides decision support and
incorporates into its logic engine the data
transferred from St. Elsewhere. The
system alerts the admitting physician that
“with a creatinine of 2.7 (>2.0) the
metformin should be discontinued due to
the risk of a serious metabolic acidosis.”
Dr. O d/cs the Metformin and orders a SS
Insulin regimen.


Two days later Mr. K’s K+ is 4.7 and his bicarbonate is
24, his creatinine is 1.7 and trending toward his
baseline of 1.3. On the fourth hospital day he is
discharged in good condition and his family decides to
plan a repeat trip to Disney World next year when dad
feels a little better.
For Mr. K, Informatics standards and FGH’s clinical
decision support system made an immeasurable
difference for him and his family.

Michael, Mr K’s son, is inconsolable. His pediatric records
from St. Elsewhere shows that he is considered healthy, but
due to his behavior and separation from his parents, he is
visited by child psychiatry. It is determined that Michael is
suffering from an acute stress reaction and they wish to
prescribe a selective serotonin reuptake inhibitor, namely
escitalopram (Lexapro). However, Michael’s St. Elsewhere
records show his DNA sequence data to have a polymorphism
associated with nonfunctioning CYP3A4 enzyme indicating
that Michael would likely be a poor metabolizer of
escitalopram. Therefore, he is placed on sertraline (Zoloft)
which is metabolized by the p450 CYP2D6 and CYP2C19
pathways where he is known to have a *1/*1 phenotype. This
medication decision avoids subjecting Michael to potentially
serious medication side-effects that may have occurred
through inadvertent overdosing of escitalopram.
Basic Formal Ontology (BFO)
Defines the high-level structures common to all domains
Connects → Health – Basic Science – Finance & Engineering
• Cell Ontology (NHGRI, NIAID)
• eagle-i and VIVO (NCATS)
• Environment Ontology (GSC)
• Gene Ontology (NHGRI)
• IDO Infectious Disease Ontology
(NIAID)
• Nanoparticle Ontology (PNNL)
• Ontology for Risks Against Patient
Safety (EU)
• Ontology for Pain, Mental Health
and Quality Of Life (NIDCR)
• Plant Ontology (NSF)
• Protein Ontology (NIGMS)
Ceusters W, Elkin P, Smith B. Negative findings in electronic health records and
biomedical ontologies: a realist approach. Int J Med Inform. 2007 Dec;76 Suppl
3:S326-33.
• Translational Medicine Ontology
(W3C)
• US Army Biometrics Ontology
(DOD)
• Vaccine Ontology (NHBLI)
EHR architecture (Level 2 Ontology: Healthcare Specific)
*
Audit Log
Best Practice Rules
Temporal Model
Clinical Reminders
Genomics/Proteomics
Core Classes
*
*
1
1
*
*
*
1
1
1..*
1..*
*
1
0..*
1
*
Patient
Clinical State
Episode of Care
Documents
1
1
1..*
1
1..*
0..*
1
1
* 0..*
1..*
1
0..*
0..*
0..*
0..*
Encounter
Non-Encounter Event
*
1
Quality Assurance
0..*
Abstraction
*
*
0..*
1..*
1..*
1
0..* 0..*
1
1
Reports
1..*
1
1
1
*
1
1..*
0..*
1
Surveillance
1
Adverse Reactions
Medications
-Allergic
-Non-allergic
*
*
1
*
*
0..*
1..*
1
*
0..*
0..*
Diagnosis
1
0..*
*
Result
*
Order
Procedure
*
0..*
*
1 1 1..*
1
0..*
*
Symptoms
0..*
*
1
0..*
1
0..*
1..*
0..*
*
1
*
Privileged Clinician
Prescription
0..*
1
*
*
Security management Information Base
1
-Authenticate
-Authorization
-Confidentiality
-Integrity
-Non-reputability
1
1
*
Security Policies
Authentication
1
0..*
1
Privileges
*
Clinician
-Identity
*
1
1 *
1 1
1





Fully Encoded Health Record
Consistent with the Level One and Two
Ontologies for Health
Compositional Expressions are
assigned Automagically
Information is gathered through the
usual documentation of patient care.
Example…………..
Unstructured Text Converted to
Indexed & Q.A.’d Electronic Health Record
Unstructured
Medical Text
Parsed Electronic
Health Record
Indexed Electronic
Health Record
Q.A.’d Electronic
Health Record

-
Cellulitis of the left foot with Osteomyelitis
of the Third metatarsal without Lymphangitis
[AND]
-
-
[WITH]
Cellulitis (disorder) [128045006]
[has Finding Site]
Entire foot (body structure) [302545001]
[has Laterality]
Left (qualifier value) [7771000]
Osteomyelitis (disorder) [60168000]
[has Finding Site]
Entire third metatarsal (body structure) [182134006]
[WITHOUT]
Lymphangitis (disorder) [1415005]
-
-
-
.
-
-
.
-
.
Case One
Case Two
Multi-Center Data
Sharing and
Interchange
Intelligent Agents
Method
Clustered Genes
Unclustered
p-value
Genes
Traditional
4,100 had
4,178 had
P=0.228
Articles
Articles
Semantic
38,820 had
28,839 had
P<0.001
Articles
Articles
Semantic Method True Cluster
Negative Cluster
Predictive Value
Clustered
1284
0
100%
Unclustered
6
734
99.2%
Sensitivity – 99.5% Specificity – 100%
Traditional Method True Cluster
Clustered
861
Unclustered
429
Sensitivity – 66.7%
Negative Cluster
0
3325
Specificity – 100%
Predictive Value
100%
88.6%
English
French
1) The ultimate health care record - (Sep 10 2007)
Mayo Clinic researchers are working on ways to make electronic health
care records more intelligent. But can they get too smart for everyday
providers?
2) Editor's Letter: To the edge and back - (Sep 10 2007)
The United States is no doubt one of the most innovative countries in its
use of health IT.
3) iEHRs await federal action (Oct 22, 2008)
An expert predicts that federal standards for electronic quality monitoring
will drive the market for intelligent EHRs.
4) Health IT success: How cool is that? (Nov 20, 2008)
Energizing public-sector health care organizations should be near the top
of Obama’s management agenda. Dr. Peter Elkin at the Mount Sinai
School of Medicine is working on electronic health record technology that
would automate clinical data gathering. Using so-called intelligent EHRs,
clinicians would get swift feedback about the quality of the care their
patients receive.

Web 1.0
◦
◦
◦
◦
HTTP
HTML
Hyperlinks
Multimedia
• Web 2.0
•Web 1.0 Plus
•Social Networking
• Web 3.0
•Web 2.0 Plus
•Semantic Search
•Knowledge based Applications
eStudies:
The Semantic Biome:
Diabetic patients who had an
Acute Myocardial Infarction
and did not have Chest Pain
“Five Minute” eStudy . . .
1
2
3
4
5
6
7
8
9
10
11
12
13
pneumonia 233604007 [P] [Explode] [A]
pneumovax 333598008 [P] [Explode] [W]
(1 AND 2)
influenza vaccine 46233009 [P] [Explode] [W]
(1 AND 4)
influenza 6142004 [P] [Explode] [W]
influenza 6142004 [P] [Explode] [A]
(1 AND 7)
(4 AND 8)
streptococcus pneumonae 58800005 [P] [Explode] [W]
(1 AND 10)
(2 AND 11)
(3 AND 4)
469
167
24
77
3
37
6
8
Pneumonia Records
Pneumonia and Pvx
Pneumonia and Influenza Vaccine
Pneumonia and Influenza
Pneumonia and Influenza and Influenza Vaccine
Pneumonia and Strep Pneumonae
Pneumonia and Strep Pneumonae and Pvx
Pneumonia and Influenza vaccine and Pvx
44
3
Pneumonias and Smokers of 469 Pneumonias
Pneumonias and Smokers who were counseled to Quit smoking
319
202
Smokers
Smokers who were counseled to quit smoking
Exam
Quality Indicator
Sens
Spec
Does report note subjective complaints?
100%
-N/A-
Does report describe need for assistive devices?
97%
75%
74%
86%
92%
77%
TP
TN
FP
FN
Sens
Spec
Audio
Eye
Feet
1.
989
63
41
106
90%
61%
2.
820
40
22
112
88%
65%
3.
943
58
35
104
90%
62%
Does the report describe the effects of the condition on the
veteran’s usual occupation?
4.
GenM
iPTSD
Joints
Mental
rPTSD
Skin
Spine
1096
104
70
303
78%
60%
Does report describe effects of the condition on the
veteran’s routine daily activities?
1150
41
24
245
82%
63%
1086
84
40
100
92%
68%
1093
61
35
251
81%
64%
1131
92
60
187
86%
61%
917
58
39
126
88%
60%
1068
66
42
89
92%
61%
Total
10293
667
408
1623
86%
62%
5.
Does report provide the active range of motion in degrees?
100%
25%
6.
Does the report state whether the joint is painful on motion
82%
63%
91%
68%
7.
Does the report address additional limitation following
repetitive use?
8.
Does the report describe flare-ups?
90%
65%
9.
Does report address instability of knee?
86%
50%
98%
0%
92%
68%
10. Does the report include results of all conducted diagnostic
and clinical tests?
totals

Brown SH, Elkin PL, Rosenbloom T, Fielstein EM, Speroff T. eQuality for
All: Extending Automated Quality Measurement of Free Text Clinical
Narratives. AMIA Annu Symp Proc. 2008 Nov 6:71-5.
Comparison of natural language processing biosurveillance methods for
identifying influenza from encounter notes.
Elkin PL, Froehling DA, Wahner-Roedler DL, Brown SH, Bailey KR.
Ann Intern Med. 2012 Jan 3;156(1 Pt 1):11-8.
DISTINCT (Gene)
DISTINCT (Disorder)
DISTINCT (Gene-Disorder Pairs)
GENE [HUGO HGNC Database] –
DISORDER (14,293,089)
GENE [20889005] - DISORER (16,135)
GENE [67271001] - DISORDER
(151,329)
GENE [8116006] - DISORDER (72,082)
GENE[HUGO] (2,244)
GENE[20889005] (3)
GENE[67271001] (24)
GENE[8116006] (13)
Total
Sort the Gene – Cancer (and
Protein – Cancer) linkages by how
many Tissue Types of Cancer is
each Gene or Protein Linked
10 Genes linked to 30 or more
Cancers
72 Genes linked to 20 or more
Cancers
191 Genes linked to 10 or
more Cancers

Elkin PL, Tuttle MS, Trusko BE, Brown SH.
(2009) BioProspecting: novel marker discovery
obtained by mining the bibleome. BMC
Bioinformatics. 10 Suppl 2: S9.
-
DISORDER
DISORDER
DISORDER
DISORDER
14,532,635
(8,076)
(7,728)
(8,064)
(8,059)
B
Modeled or experimental
protein structure
Ebola Proteome
Protein-drug ligand
Drug or other ligand
Ligand Library
A
Matrix of binding affinities
Top novel predictions
Compound(s)
ConfidenceInteraction Protein target(s)
enfuvirtide
score (min)score
(max)
7
2.0
GP2, VP35, 1ebo-F
vancomycin, bleomycin
10
2.0
GP1,2, pre-sGP, SGP, SsGP
octreotide, lanreotide,
somatostatin
Ubidecarenone (CoQ10)
10
2.0
GP1,2, pre-sGP, SGP, SsGP
7
1.6
unoprostone
10
1.3
GP1,2, GP2, VP24, VP35, VP40,
1ebo-F
GP1,2, VP35, VP24, 1ebo-F
Recommendation: Enfuviritide AND Ubidecarenone (CoQ10) AND
Octreotide With or Without Unoprostone



Technology and standards have progressed
to the point where NLP is a viable solution
High Throughput Phenotyping
◦ Recall (Sensitivity) of 99.7%
◦ Precision (PPV) of 99.8%.
Ontological tagging using NLP can support
translational research using EHR data in the
Era of Big Data, electronic quality
monitoring (eQuality) and clinical decision
support.
Peter L. Elkin, MD, Steven H. Brown, MD, Casey Husser, MD, Brent A. Bauer, MD,
Dietlind Wahner-Roedler, MD, S. Trent Rosenbloom, MD, Ted Speroff, PhD;
“An Evaluation of the Content Coverage of SNOMED-CT for Clinical Problem
Lists“, Mayo Clin Proc. 2006 Jun;81(6):741-8.

That our patients deserve the highest quality, safest
care that we can provide.
◦ Requires the use of all of the patient’s relevant data.
◦ Utilizing that data in the context of best practice.
◦ Our hypothetical Mr. Kneivel and his son Michael are
representative of people who have put their trust in
us to provide for them the very best care. This
requires systems engineering that can help us to
integrate and analyze patient data in order to
provide clinicians with just-in-time point-of-care
best practice advice, in support of their medical
practice.

A Strong Ontology based Informatics Program is
Essential to the Success of Modern Research
◦ Data is more Complex
◦ Data is seen in Greater Volume (Petabytes not Gigabytes or
Terabytes) – Big Data to Knowledge (BD2K)
◦ Ontology based EHRs can be implemented in such a way
that they can automatically populate observational
databases enhanced by Ontology such as i2b2 and OMOP
and others.
◦ Informatics tooling has been shown to be Reusable and
Practical
◦ Translational Research requires the combination of
multiple datatypes crossing the clinical and basic science
realms the link is BFO conformant Ontologies
◦ Informatics Pipelines can improve the speed of discovery
and increase research productivity.
“…there is nothing more difficult to take in hand,
more perilous to conduct, or more uncertain in its
success, than to take the lead in the introduction of
a new order of things. Because the innovator has
for enemies all those who have done well under the
old conditions, and lukewarm defenders in those
who may do well under the new. “
Nicolo Machiavelli c. 1505