SNOMED and “phenotypes”

SNOMED and “phenotypes”Signs, symptoms, findings, etc

Signs, Symptoms and Findings: Steps Toward an Ontology of Clinical Phenotypes

Sept 3-4, 2008Dallas TX

Symptoms & signs

• Conclusions of the SNOMED RT group (1996):– There is no point attempting to put all

observations or findings or manifestations into the mutually-exclusive categories “symptom” and “sign”

– But: it is ok to classify and observation by the source of information (subject, observer, carer, etc) when that is explicit, clear, and necessarily so.

• Pain is not necessarily reported by the patient.• Tenderness is not necessarily reported by the clinician.

Findings & disorders

• Conclusion after (literally) years of attempting to find a clean reproducible boundary:– It is pointless to attempt to pursue a boundary

between “findings” and “disorder”– It probably is both possible and useful to

differentiate “disease” (as a potentiality), from qualities, the results of observations

• “Diagnosis” is not the same as disease, and is used as a (variable-across-sites) administrative category

“lab” tests & results• A clinical laboratory is an environment in which tests are performed on patient

specimens, and which is specifically designed and configured for that purpose

• Many “lab” observations are moving out of the laboratory and into point-of-care testing environments, e.g. operating room (and an OR is not a lab).

– Trend is likely to continue

• It is not important to (ontologically) categorize observations as being necessarily “lab tests”

– And they MUST NOT be categorized as lab tests if they are not necessarily performed in a laboratory setting (otherwise what is a lab test?)

• A prothrombin time (PT) is NOT a laboratory test (sorry).

• It may be useful to produce a subset of observations that are “ordinarily thought of in our institution as lab tests”, but being a lab test is not a necessary definitional aspect of the observations.

Terminology vs Information model:Balance, overlaps, gaps

• Record the fact that “malignant mesothelial cells were found in a pleural fluid aspirate”:

Field or question Terminology value

Pleural fluid finding Malignant mesothelial cells

Site of malignant mesothelial cells

Pleural fluid

Lab test result Malignant mesothelial cells in pleural fluid

Type of mesothelial cells in pleural fluid

Malignant

Type of malignant cells in pleural fluid

Mesothelial

Terminology vs Information model:Balance, overlaps, gaps

• There is no single best way to split assertions between the information model and the terminology model (or between the observables and the other values in the terminology!)

• The best we can do is recognize equivalence

• The best tools for recognizing equivalence (by machine) are logic-based

• Therefore, a logic-based model of semantics is the foundation not just for the terminology but also for the combination

Observations / observables

• An observation is an act (a procedure)• The result of the observation may be a

statement about a phenotype or finding• An observable: what is it?

– Incomplete finding

Observables: examples

• Head circumference• Blood hemoglobin concentration• MRSA POC test result• Contents of urine on microscopy

Yet another draft model for observables (1)

– Define a model for observables that makes a distinction between the inherent quality that is being observed and any aspects of the actual observation

– Two parts of the model: the property part, and the observation part.– The property part deals with real properties that exist independent of observation;

the observation part deals with how we know about the quality/property (it is ontological with respect to acts of observation)

– For the inherent property, use a role group instead of nesting; this would allow more than one property per observation

– Need to validate whether it is necessary to have more than one property; if not, we can eliminate the role group

– For the observation part, no role group is needed because we assume a different code for each observation

– If a concept (finding) involves the results of multiple observations, then assume it is a situation. E.g. “hyponatremia with hypokalemia”

– Expand observable model to make a model for observation FINDINGS and a model for observation PROCEDURES:

– observation PROCEDURES add the attribute: METHOD = observation action.– observation FINDINGS add attributes HAS INTERPRETATION and HAS VALUE


• What happens to existing measurement procedure attributes?– HAS SPECIMEN

– Replaced by a DIRECT SITE attribute, which is the direct object of the observation action and is to be used when the entity that is being observed is different from the entity in which the property inheres.

– Replace MEASUREMENT METHOD with TECHNIQUE– Create a new value set for techniques

– Revise and change the configuration of COMPONENT, PROPERTY, SYSTEM– COMPONENT and SYSTEM

– Replace by INHERES-IN and TOWARDS, to get better reproducibility– PROPERTY

– Retain current properties and add values from PATO (ontology of qualities)– TIME ASPECT, and SCALE TYPE

– Retain in observation part of model, and add UNITS to coordinate with IFCC-IUPAC

– Move to their own hierarchies, separate from observables:• functions• processes • activities

– Allow functions, processes, activities also to be values of TOWARDS (in addition to substances, etc)

– Do not allow observables to be values of INHERES-IN or TOWARDS.


– Differentiate between a property per se and a property type.– Concentration is a property type.– The concentration of sodium in serum is a property.– In the current model, the values of the PROPERTY attribute are

property types.– Consider what happens by avoiding the use of “presence” as

a property type. (This also presumes avoiding the use of “absence” as a property type.)– Asserting absence requires negation (possibly using the situation

model).– Consider a separate pre-coordinated “property” hierarchy as

a work-around for the lack of nested expressions in definitions

observable

PROPERTY Properties

TIME ASPECTSCALE

Time aspects

Scale types

DRAFT model of observables

independent continuantINHERES IN

TOWARDS Functions, substances

UNITS units

TECHNIQUE techniques

GPRECONDITION Body states

DIRECT SITE Body structures, specimens

observable

PROPERTY Properties

TIME ASPECTSCALE

Time aspects

Scale types

DRAFT model of observables

independent continuantINHERES IN


UNITS units




IFCC-IUPAC NPU elementsLOINC elements

Kind-of-property property

System/specimen system

component component

Time aspect

scale

units

method method

System/specimen

component

observable

PROPERTY Substance concentration

TIME ASPECTSCALE

Single point in time

quantitative

LOINC Example:Sodium:SCnc:PT:Ser/Plas:Qn

PlasmaINHERES IN

TOWARDS Sodium ion

UNITS units



DIRECT SITE Serum/Plasma

Kind-of-property

component

Time aspect

scale

System/specimen

observable

PROPERTY Substance concentration

TIME ASPECTSCALE


quantitative

IFCC-IUPAC NPU Example:P—Sodium ion; subst.c. = ? mmol/l

PlasmaINHERES IN

TOWARDS Sodium ion

UNITS mmol/l




property

system

component

units

observable

PROPERTY concentration

TIME ASPECTSCALE

Time aspects

Scale types

Blood hemoglobin concentration

Intravascular bloodINHERES IN

TOWARDS hemoglobin

UNITS units




OBS TARGET independent continuant

finding

PROPERTY Mass concentration

TIME ASPECTSCALE


quantitative

VALUE 14

HAS INTERPRETATION Incr, decr, normal, abnormal

Blood hemoglobin 14.0 gm/dL

Intravascular bloodINHERES IN

TOWARDS hemoglobin

UNITS gm/dL





observable

PROPERTY circumference

TIME ASPECTSCALE

Time aspects

Scale types

Surface of headINHERES IN


UNITS units




Head circumference


finding

PROPERTY circumference

TIME ASPECTSCALE

Time aspects

Scale types

VALUE 28


Surface of headINHERES IN


UNITS cm




Head circumference 28 cm


observable


TIME ASPECTSCALE

Time aspects

Scale types

Serum concentration of Borrelia antibody (observable)

plasmaINHERES IN

TOWARDS Borrelia antibody (substance)

UNITS units

TECHNIQUE Light microscopy


DIRECT SITE Serum specimen


procedure


TIME ASPECTSCALE

Time aspects

Scale types

VALUE Numeric, ordinal, nominal


Measurement of serum Borrelia antibody by ELISA(procedure)

plasmaINHERES IN

TOWARDS Borrelia antibody (substance)

UNITS units

TECHNIQUE ELISA


DIRECT SITE serum specimen

METHOD Observation action


finding

PROPERTY serotype

TIME ASPECTSCALE

Time aspects

Scale types

VALUE O157


E coli (organism)INHERES IN


UNITS units

TECHNIQUE Bacterial serotyping


DIRECT SITE Microbial culture

Cultured organism serotype is O157

The information model is used to link this finding to the Culture and specimen that came from the patient.


What about abilities?

• Able, unable, etc. with respect to normal functions and activities

observable

PROPERTY ability

TIME ASPECTSCALE

Time aspects

Scale types

Auditory systemINHERES IN

TOWARDS To hear (function)

UNITS units




Ability to hear


finding

PROPERTY Ability

TIME ASPECTSCALE

Time aspects

Scale types


HAS INTERPRETATION able



UNITS units




Able to hear


What about negation?

• Option 1: use different values of HAS INTERPRETATION and deal with opposites outside the model of meaning (not recommended)

• Option 2: use the situation model

finding

PROPERTY Ability

TIME ASPECTSCALE

Time aspects

Scale types


HAS INTERPRETATION unable



UNITS units




Unable to hear (1)


finding

PROPERTY Ability

TIME ASPECTSCALE

Time aspects

Scale types





UNITS units




Unable to hear (2)

situation INCLUDES¬

finding

PROPERTY Ability

TIME ASPECTSCALE

Time aspects

Scale types





UNITS units




able to hear (2)

situation INCLUDES

observable

PROPERTY Briskness of response

TIME ASPECTSCALE

Time aspects

Scale types

Neuromuscular structures of the left knee deep tendon reflexINHERES IN

TOWARDS Deep tendon reflex

UNITS units




Left knee deep tendon reflex - briskness


finding

PROPERTY Briskness of response

TIME ASPECTSCALE

Time aspects

Scale types

VALUE 2+ out of 4+


Neuromusc. Struc. L knee DTRINHERES IN

TOWARDS Deep tendon reflex

UNITS units




Left knee jerk reflex 2+


Technological and scientific foundation

• Collaborative development (“social computing”)– SNOMED RT and CT were built using collaborative tools and techniques

• Campbell KE, Cohn SP, Chute CG, Shortliffe EH, Rennels G. Scalable methodologies for distributed development of logic-based convergent medical terminology. Methods Inf Med. 1998 Nov;37(4-5):426-39.

– IHTSDO is serious about re-energizing collaborative development• Open standards-based technology platform (Open Health Tools)• Multilingual workbench RFP

• Description logic– Semantics is understood and being used and studies by the

DL scientific community (DL is called EL+)• a number of papers at DL2008 use SNOMED CT as test bed• Stated form to be officially part of the release as of July 2008

– A variety of different classifiers (CEL, FaCT++) have been used to verify and validate (e.g. no post-processing, etc)

IHTSDO welcomes participation and collaboration

• Single world-wide affiliate license, with no charge for research & evaluation– http://www.ihtsdo.org/our-standards/licensing/

• Open publication of the concept model, style guide, technical specifications:– http://www.ihtsdo.org/about-ihtsdo/snomed-ct-publications/

• Collaborative web site – open to participation without charge– https://thecap.seework.com/login– Email to: [email protected] for free registration

• Working groups and Committees: all open– Project groups, Special interest groups

Documents

SNOMED and “phenotypes”