Linguistic String Project Format 5 Patient State

1
STRUCTURED HEALTH MARKUP LANGUAGE SHML
DAVID J. ROTHWELL M.D. HEALTH LANGUAGE CENTER
2
Major Issues Affecting Patient Medical Record
Information

• Continuing Barriers to Data Entry
• Increasing Role of Patient Choice
• Shift from Acute / Inpatient to Disease
  Prevention / Management
• Impact of Genomics
• Rapid Rise of Internet Standards

3
Health Language Center Approach

• Convergence / Maturity of Two Major Technologies
• Internet Standards - XML
• Natural Language Processing
• Potential to Keep Pace with Change

4

• Care Process
• ?Clinical Research
• ?Aggregation of Data
• ?Epidemiology
• ?Adherence to guidelines
• ?Eligibility for protocols

PMRI
Utility
Provider Patient Document
Findings
PMRI
NLP XML SHML

• Clinically Important Information (minus)
  verbiage
• ? Clinically Relevant Content Units (CRCU)

Text SDE
5
Natural Language Processing

• A system which turns natural language clinical
  documents into structured data for a variety of
  applications.
• The NLP method makes explicit the informational
  structure of texts, using linguistic method and
  the most advanced information technology
  available today.

6
NLP (contd)

• Language presents information linearly in
  strings--phrases, sentences, paragraphs,
  sections, documents, discourses,
• Information in these strings is carried by the
  semantic types of words, occurring in particular
  combinations.

7
(No Transcript)
8
NLP
Formats include

• Demographic Data
• Verbs
• Patient State Data
• Diagnosis
• S-S (H-INDIC)
• Patient Status Data
• Patient
• Anatomy (H-PTPART)

• Treatments
• Test and Result
• Time
• Uncertainty (H-MODAL)
• Negation
• Response
• Changes

9
(No Transcript)
10
Use of Medical Language Processing for HEDIS
Measures
"BETA BLOCKER TREATMENT AFTER A HEART
ATTACK" Hospital discharge -records in text
form (natural language) are prepared for
processing, analyzed, formatted for retrieval of
information, and the following queries submitted
by Aurum Medical Language Processor ? Did the
patient have an acute, myocardial infarction? ?
Was the patient given a beta blocker
medication? ? Did the patient have any
contraindications?
11
List of Beta Blockers LOPRESSOR, METOPROLOL,
NADOLOL, CORGARD, ATENOLOL, TENORMIN, PlNDOLOL,
VISKEN, PROPRANOLOL, INDERAL, INDURAL, BETA
BLOCKER, ACEBUTOLOL, BETAXOLOL, BISOPROLOL,
CARTEOLOL, CARVEDLOL, LABETALOL, PENBUT0LOL,
SOTALOL, TIMOLOL. List of Contraindications
HEART BLOCK, ATRIOVENTRICULAR BLOCK,
BRADYCARDIA, BRADYCARDIC, LV DYSFUNCTION,
VENTRICULAR DYSFUNCTION, DIASTOLIC DYSFUNCTION,
VENTRICULAR DIASTOLIC DYSFUNCTION, COPD, CHRONIC
OBSTRUCTIVE PULMONARY DISEASE, DIABETES
MELLITUS, ASTHMA, CONGESTIVE HEART FAILURE,
Results from database queries Beta Blocker
Beta Blocker Given Not Given Total With
contraindications 42 19 61 Without
contraindications 28 2 30 Total 70 2l 91

12
eXtensible Markup Language (XML)

• Streamlined subset of SGML
• XML is a language
• Document Type Definition
• DTD defines it's use (it's grammar)
• Designed for data exchange
• Data processing oriented rather than publishing
  (SGML)
• Create own 'tags' -what you need to know
• i.e. Annotate document with meaning

13
eXtensible Markup Language (contd)(XML)

• Data and document are combined
• Tagged text transforms it into data
• Tags are granular descriptive views of text
• Tags are metadata--information not explicit in
  text
• Puts meaning and interpretation on top of text
• Ability to catalogue all information in document
• Preserves fundamental structure of document
  (PMRI)
• Tagged data can fit into any data format or data
  model

14
XML Tags
ltelementTypegt Content lt/elementTypegt Pneumonia ltD
iagnosisgt Pneumonia lt/Diagnosisgt Pneumonia,
right lower lobe ltDiagnosisgt Pneumonia ltlocationgt
right lower lobe lt/locationgt lt/Diagnosisgt Pneumon
ia, right lower lobe, superior, due to
Klebsiella. ltDiagnosisgt Pneumonia ltlocationgt
right lower lobe lt/locationgt ltpositiongt superiorlt/
positiongtltlinkgtdue tolt/linkgt ltorggtKlebsiella
lt/orggt lt/Diagnosisgt PRESENTATION FORMAT (one of
many) Diagnosis Pneumonia Location RLL,
superior Organism Klebsiella
15
XSL
Demographic Data Node positive carcinoma of the
left breast, treated by mastectomy, chemotherapy
and radiotherapy Radiotherapy treatment summary
the left breast and draining nodal areas received
a dose of 42.9 Gy in 13 fractions treating three
times a week with 6 Mv photons. Treatment started
on the 15th of September and was completed on the
14th October, 1995.
Radiotherapy Treatment Summary Status before
radiotherapy Diagnosis Carcinoma
left breast Spread Left axillary nodes
Previous treatment Mastectomy,
chemotherapy Radiotherapy given Treatment
type 6 Mv photons Site Left chest wall
and draining nodes Total dose given 43 Gy
Schedule 3 fractions/week 15 September 1995 to
... Follow up plan One visit . . .
16
Structured Health Markup Language

• Utilizes Linguistic Model not Coding
• Utilizes XML
• Designed to Integrate both Structured Data Entry
  and Text

17
SHML and MLP
Medical documents
MLP in HLC/SHML Dictionaries
standardization
Documents with SIDs

• GENERATORS
• SHML/DTD
• SHML/XSL
• SHML/XQL

Documents in CRCUs with SHML and MLP tags
MLP
Documents in rows of standard dBMS
Other Applications
18
Mapping --NLP Dictionary / SHML
Term NLP Class SHML
Altered awareness H-INDIC--N fs
neuro Alternate H-TMREP--TV tmr Alternating
between H-CONN--P lprep Altogether
H-AMT--D mamt Aluminosis H-INDIC--N fd
tox Augmentations H-CHANGEMORE--N mcha Augus
t NTIME1--N tme Aunt
H-FAMILY--N per kin Auricle
H-PTPART--N as ear Auriculectomy
H-TTCHIR--N pr Auriculo-osteodysplasia
H-INDIC--N fd cong Auriculotemporal
H-PTPART--ADJ areg Auscultate
H-TXCLIN--TV pr
19
SHML TAGS Traditional Set

• Demographic data ltdemgt (subclasses)
• Anatomic Structure ltA.S.gt (Digital Anatomist)
• Medication ltdrgt (Multum, First Data)
• Organisms ltorgt
• Chemical ltchgt (nonphysiologic)
• Devices ltdevgt (ECRI)
• Occupation ltoccgt (National, international)
• Procedures ltprgt (CPT, other)
• Diagnosis ltfdgt (ICD-9, ICD-O, Medcin)

20
SHML TAGS (contd)
Verbs ltlvgt Subsets defined by
MLP Preposition ltprepgt All except time
prep Dietary/Food ltdietgt Nutrition Time lttmgt
Exact, begin, end, frequency Amount ltmamtgt Cha
nge ltmchagt Less, more Certainty ltmcergt Uncert
ainty, modal, certainty Negation ltmneggt Stage
grade ltms-ggt Dimension ltmdimgt Adjective ltmadgt A
ppearance ltmappgt Color, shape, clarity
Odor ltmsmellgt Position ltppogt Top to bottom,
laterality Person ltpergt Masc, Fem,
pronouns Environment ltenvgt Physical
locations Transparent lttranspgt Derived from
entity, modified
21
Additional SHML TAGS
Findings ltFindinggt Vital Signs
Signs/symptoms by organ system Lab
Image Behaviors
Living Functional status
Injury Disability Exposures Compliance
Travel Dental ADL Alternative
care Exercise etc. Leisure,
sports Immunization Allergy,
tolerances Education, counseling
22
SHML / XML TAGGINGltelementTypegtContentlt/elemen
tTypegt
Clinical Table
Vocabulary Table
Chest Pain Location Substernal Onset
--hrs ago --days ago Brought on by
jogging walking Relieved by rest
nitroglycerin
Terms Taxonomic knowledge Hierarchical knowledge
(classificatory) Synonyms/Equivalent
terms Linguistic knowledge Definitional
knowledge Non-unique term knowledge Tag knowledge
23
SHML Vocabulary Table (Outline)
Digestive Tract (syn G.I. Tract, Alimentary
Tract) Upper G.I. Tract Mouth Tongue
Teeth Gums Pharynx Oropharynx Hypoh
arynx Esophagus Stomach Lower G.I.
Tract Small Intestine Duodenum Jegun
um Large Intestine Cecum Appendix
Colon Rectum Anus
24
Term/phrase Relation/Attribute
Term/Phrase Relation Relation TAG
(has
member) Anatomic Structure as Integumentary
System is a Anatomic StructureAnatomic Structure
as Breast is a Anatomic
StructureAnatomic Structure as
Musculoskeletal System is a Anatomic
StructureAnatomic Structure as Digestive
System is a Anatomic StructureDigestive
System as Digestive Tract is a Anatomic
StructureDigestive System as Digestive Organs is
a Anatomic StructureDigestive Tract as Upper GI
Tract is a Anatomic StructureDigestive
Tract as Lower GI Tract is a Anatomic
Structure Upper G. I. Tract as Mouth is
a Anatomic StructureOral syn Mouth BLANK BLANK
Upper G. I. Tract as Pharynx is a Anatomic
StructureUpper G. I. Tract as Esophagus is
a Anatomic Structure Upper G. I.
Tract as Stomach is a Anatomic
Structure Mouth as Lips is a Anatomic
StructureMouth as Tongue is a Anatomic
Structure Mouth as Palate is a Anatomic
StructureRoof of mouth syn Palate BLANK BLANK To
ngue as Posterior third is a Anatomic
StructureTongue as Anterior 2/3 is
a Anatomic Structure
25
Vocabulary Attribute Table Design---Findings
Term/phrase Relation Term/phrase
Relation Tag (has member) Findi
ng is a Source Finding
Finding Finding constitutional (fc) Finding
Finding Finding tissue(ft) Finding
Finding Finding integumentary (fs integ)
Finding Finding Finding musculoskeletal
(fsmss) Finding Finding Finding
respiratory (fs resp) Finding
Finding Finding neurologic (fs neuro) Finding
neurological Finding Dyslexia is a
Finding Finding neurological
Finding Aphasia is a
Finding Finding neurological Finding Phobia
is a Finding Phobia
Finding Acrophobia is a
Finding Phobia Finding Claustrophobia
is a Finding
26
SHML and NLP Example 2
SIDNE1519 017A.11.01 ON THE 16TH HOSPITAL DAY
, AN ELECTROCARDIOGRAM SHOWED PROBABLE ATRIAL
FIBRILLATION AT A VENTRICULAR RATE OF 100 , WITH
PREMATURE VENTRICULAR CONTRACTIONS AND POSSIBLE
OLD INFERIOR AND ANTEROSEPTAL MYOCARDIAL INFARCTS
.
WITH
AND
ON THE 16TH HOSPITAL DAY , AN ELECTRO-CARDIOGRAM
SHOWED PROBABLE ATRIAL FIBRILLATION AT A
VENTRICULAR RATE OF 100
AND
PREMATURE VENTRICULAR CONTRACTIONS
POSSIBLE OLD INFERIOR MYOCARDIAL INFARCTS
POSSIBLE OLD ANTEROSEPTAL MYOCARDIAL INFARCTS
27
SHML and NLP Example 2.1
SIDNE1519 017A.11.01 ON THE 16TH HOSPITAL DAY
, AN ELECTROCARDIOGRAM SHOWED PROBABLE ATRIAL
FIBRILLATION AT A VENTRICULAR RATE OF 100 , WITH
PREMATURE VENTRICULAR CONTRACTIONS AND POSSIBLE
OLD INFERIOR AND ANTEROSEPTAL MYOCARDIAL INFARCTS
. ltCONNECTIVEgtltCONJOINEDgtltCONNgtltPgtWITHlt/Pgtlt/CONNgt
lt/CONJOINEDgt ltPATIENT-STATEgt
ltMETHODgtltPROCEDUREgtltTgtANlt/TgtltNgtltprgtELECTROCARDIOGR
AMlt/prgtlt/Ngt
lt/PROCEDUREgtlt/METHODgt ltVERBgtltTV
tensePASTgtltshowgtSHOWEDlt/showgtlt/TVgt
ltEVENT-TIMEgtltPgtONlt/PgtltTgtTHE
lt/TgtltADJgt16THlt/ADJgt
ltNgtltenvgtHOSPITALlt/envgtlt/NgtltNgtlttmlocgtDAYlt/t
mlocgtlt/Ngt ,
lt/EVENT-TIMEgtlt/VERBgt ltPSTATE-DATAgt
ltS-SgtltNgtltfnsgtFIBRILLATIONlt/fnsgtlt/NgtltPgtATlt/Pgt
ltMODSgtltMODALgtltADJgtltmcergtPROBABLElt/
mcergtlt/ADJgtlt/MODALgt
ltPTPARTgtltADJgtltansgtATRIALlt/ansgtlt/ADJgtlt/PTPART
gtlt/MODSgtlt/S-Sgt ltPTFUNCgtltTgtAlt/TgtltNgtltfungtRA
TElt/fungtlt/NgtltPgtOFlt/Pgt
ltMODSgtltPTPARTgtltADJgtltansgtVENTRICULARlt/ansgtlt/ADJgtlt/P
TPARTgtlt/MODSgt lt/PTFUNCgt
ltQUANTgtltQ-NgtltNUMgtltQgtltnumbgt100lt/numbgtltQgt,lt/NUMgtlt/QN
gtlt/QUANTgtlt/PSTATE-DATAgt lt/PATIENT-STATEgt
next...
28
SHML and NLP Example 2.2
SIDNE1519 017A.11.01 ON THE 16TH HOSPITAL DAY
, AN ELECTROCARDIOGRAM SHOWED PROBABLE ATRIAL
FIBRILLATION AT A VENTRICULAR RATE OF 100 , WITH
PREMATURE VENTRICULAR CONTRACTIONS AND POSSIBLE
OLD INFERIOR AND ANTEROSEPTAL MYOCARDIAL INFARCTS
. WITH ON THE 16TH HOSPITAL DAY , AN
ELECTROCARDIOGRAM SHOWED PROBABLE ATRIAL
FIBRILLATION AT A VENTRICULAR RATE OF
100 ltCONNECTIVEgtltCONJOINEDgtltCONNgtANDlt/CONNgtlt/CONJ
OINEDgt ltPATIENT-STATEgt ltPSTATE-DATAgt ltS-SgtltADJ
gtltH-INDICgtlttmlocgtPREMATURElt/tmlocgtlt/H-INDICgtlt/ADJgt
lt/S-Sgt ltPTPARTgtltADJgtltH-PTPARTgtltansgtVENTRICULARlt/
ansgtlt/H-PTPARTgtlt/ADJgtlt/PTPARTgt ltPTFUNCgtltNgtltH-PTF
UNCgtltfungtCONTRACTIONSlt/fungtlt/H-PTFUNCgtlt/Ngtlt/PTFUNC
gt lt/PSTATE-DATAgt lt/PATIENT-STATEgt next
...
29
SHML and NLP Example 2.3
SIDNE1519 017A.11.01 ON THE 16TH HOSPITAL DAY
, AN ELECTROCARDIOGRAM SHOWED PROBABLE ATRIAL
FIBRILLATION AT A VENTRICULAR RATE OF 100 , WITH
PREMATURE VENTRICULAR CONTRACTIONS AND POSSIBLE
OLD INFERIOR AND ANTEROSEPTAL MYOCARDIAL INFARCTS
. ltCONNECTIVEgtltCONJOINEDgtltCONNgtANDlt/CONNgtlt/CONJOI
NEDgt ltPATIENT-STATEgtltPSTATE-DATAgt
ltDIAGgtltNgtltH-DIAGgtltftgtINFARCTSlt/ftgtlt/H-DIAGgtlt/Ngt
ltEVENT-TIMEgtltADJgtltH-TMLOCgtlttmls
gtOLDlt/tmlsgtlt/H-TMLOCgtlt/ADJgtlt/EVENT-TIMEgt
ltMODSgtltMODALgtltADJgtltH-MODALgtltmcergtPOSSI
BLElt/mcergtlt/H-MODALgtlt/ADJgt
lt/MODALgtltMODSgtlt/DIAGgt
ltPTPARTgtltADJgtltH-PTAREAgtltppogtINFERIORlt/ppogtlt/H-PTAR
EAgtlt/ADJgt
ltADJgtltH-PTPARTgtltas_cvgtMYOCARDIALlt/as_cvgtlt/H-PTPART
gtlt/ADJgt lt/PTPARTgtlt/PSTATE-DATAgtlt/PATI
ENT-STATEgt ltPATIENT-STATEgtltPSTATE-DATAgt
ltDIAGgtltNgtltH-DIAGgtltftgtINFARCTSlt/ftgtlt/H-DIAGgtlt/N
gt ltEVENT-TIMEgtltADJgtltH-TMLOCgtlt
tmlsgtOLDlt/tmlsgtlt/H-TMLOCgtlt/ADJgtlt/EVENT-TIMEgt
ltMODSgtltMODALgtltADJgtltH-MODALgtltmcergtP
OSSIBLElt/mcergtlt/H-MODALgtlt/ADJgt
lt/MODALgtltMODSgtlt/DIAGgt
ltPTPARTgtltADJgtltH-PTPARTgtltppogtANTEROSEPTALlt/ppogtlt
/H-PTPARTgtlt/ADJgt
ltADJgtltH-PTPARTgtltas_cvgtMYOCARDIALlt/as_cvgtlt/H-PTPART
gtlt/ADJgt lt/PTPARTgtlt/PSTATE-DATAgtlt/PATI
ENT-STATEgt lt/CONNECTIVEgtlt/CONNECTIVEgtlt/CONNECTIVEgt
30
(No Transcript)
31
SHML TAGGING
32
SHML Tagging of Encounter

• Demographic
• Symptoms
• History
• Allergies

• 28 y/o female
• Stabbing, aching, burning pain, back of neck
• Pain radiating to right side into scapula
• Pain occurs occasional, usually end of day
• Numbness of left triceps occasional
• No urine problems
• No bowel problems
• No gait problems
• No drug allergies

33
SHML Tagging (contd)
Medications Physical Exam Impression Plan
Advil, 400 mg.hs,prn Neck, normal position Neck
supple Neck full range of motion Neck freedom of
movement Right Trapezoid mildly tense Spine no
point tenderness Arms, full strength Fingers,
full strength Arms 1 reflexes Legs 1
reflexes Musculoskeletal pain Herniated disc,
C-5 level, small symptoms mild Surgery not an
option at this time MRI deferred until/should
bowel findings bladder findings focal
weakness (which persists) focal numbness (which
persists) point tenderness(neck) Anvil 800
mg.tid, watch for G.I. side effects Physical
therapy
34
Potential of SHML Approach

• Ability to Resolve Ambiguity
• Ability to Deal with Multiple Hierarchies

35
NLP / SHML
Depression ltfs psygt psychological ltftgt depres
sion of surface, shape ltmchagt depression of
WBC, platelet ST segment depression - idiom
(phrasal term)
36
NLP / SHML
EKG revealed sinus
bradycardia
ltpr cv ekggt
ltshowgt
ltfs cvgt
ltas cvgt Heart ltas respgt Repiratory ltas
mssgt Within bone ltftgt Rectal sinus, (fistula)
37
Use with Structured Data Entry
Chest Pain
Onset hrs ago days ago Duration 20 min 1
day Location Laterality Character Brought on
by Associated with Aggravated Relieved
by Severity Radiating to Trend
tmbeg tmd A.S. ppo-lat fi-SS- fac fi-SS li li
mamt li fres
38
Structured Health Markup LanguageSHML /XML

• Adopt rules, notation that are in place for
  SGML/XML for the medical record (PMRI)
• Create an architecture for data types
• Structure the EMR
• Utilize XML rules, notation for content
  (semantics)
• eXtensible Markup Language (XML)
• SHML/XML works with language it does not
  reinvent it!!!
• XML provides structure and contextual meaning to
  a document
• XML is a self describing data structure

39
STRUCTURED HEALTH MARKUP LANGUAGE SHML

• Subcomponent of XML
• Health DTD for validation tags and their
  rendering
• Tags assigned to terms/phrases and CRCUs
• Tags specific for health
• Tags specific for NLP
• Structured
• Defines a syntax of tags
• Rules of well-formedness
• XSL eXtensible Style Language for rendering

40
SHML vs MLP elementTypes

• MLP (syntactic) part-of-speech elementTypes are
  based on major word classes, e.g. nouns (N),
  adjectives (ADJ), tensed verbs (TV), adverbs
  (D), E.g. shortness of breath, ltNgt
• MLP co-occurrence semantic elementTypes are
  based on word usage (context), e.g. shortness of
  breath, ltH-INDICgt
• SHML semantic elementTypes are based on medical
  knowledge (classification), e.g. shortness of
  breath, ltfs_respgt.

41
SHML DTD for CRCUs
lt? XML VERSION1.0 ?gt lt!DOCTYPE STRUCTURED
HEALTH MARKUP LANGUAGE shml.dtdgt lt!ELEMENT
PATIENT-STATE (PARAGR, PT-DEMOG, METHOD,
SUBJECT, VERB, TENSE, PSTATE-DATA,
PRECISIONS, TIME, TEXTPLUS)gt lt!ELEMENT
PATIENT-TREATMENTS (PARAGR, PT-DEMOG,
TREATMENT, STATE-SUBJ, PRECISIONS, TIME,
TEXTPLUS)gt lt!ELEMENT LABTEST (PARAGR,PT-DEMOG,
INST, PT, TEST-INFO, VERB, TIME,
TEXTPLUS)gt lt!ELEMENT PARAG (PCDATA)gt lt!ELEMENT
PT-DEMOG (AGE, RACE, SEX, FAMILY)gt lt!ELEMENT AGE
(PCDATA)gt lt!ELEMENT RACE (PCDATA)gt lt!ELEMENT
SEX (MALE, FEMALE)gt lt!ELEMENT MALE
(PCDATA)gt lt!ELEMENT FEMALE (PCDATA)gt lt!ELEMENT
METHOD (PROCEDURE, EXAMTEST, DEVICE)gt lt!ELEMENT
TREATMENT (GEN, CHIR, MED, COMP)gt lt!ELEMENT
PROCEDURE (PCDATA)gt lt!ELEMENT EXAMTEST
(PCDATA)gt ...etc.
42
Presentation of Data (Reformatting)
An echocardiogram performed in the Coronary
Care Unit shows dilated left atrium,
moderate global LV dysfunction, ejection
fraction of 30, moderate global RV dysfunction,
severe mitral regurgitation. pr
cv Echocardiogram Env (place) CCU Finding
cv Dilated left atrium Moderate global LV
dysfunction Ejection fraction 30 Moderate
global RV dysfunction Severe mitral
regurgitation
43
Mission of HLC / SHML

• Define a granular representation of terms and
  phrases that within a given language (domain)
  unambiguously define clinical concepts
• Provide for an adequate representation of these
  terms and concepts in a simple and easily
  understood architecture
• Provide for discrete mapping to any other
  nomenclature and/or code set
• Utilize easily available, inexpensive and widely
  supported tools for authoring, maintenance and
  use
• Provide this as a non-proprietary standard under
  the auspices of a private not-for-profit entity

44
(No Transcript)