Representing clinical concepts Paul Taylor UCL
ICD-10 CRITERIA FOR "CHILDHOOD AUTISM" B. at least six symptoms from (1), (2) and (3), at least two from (1) and at least one from each of (2) and (3) 1. Qualitative impairment in social interaction in at least two of: a. failure adequately to use eye-to-eye gaze, facial expression, body postures, and gestures to regulate social interaction; b. failure to develop peer relationships that involve a mutual sharing of interests, activities and emotions; c. lack of socio-emotional reciprocity as shown by an impaired or deviant response to other people s emotions; or lack of modulation of behavior d. lack of spontaneous seeking to share enjoyment, interests, with other people (e.g. pointing out to other people objects of interest to the individual). 2. Qualitative abnormalities in communication in at least one of : a. delay in or total lack of, development of spoken language that is not accompanied by an attempt to compensate through the use of gestures or mime as an alternative mode of communication b. relative failure to initiate or sustain conversational interchange in which there is reciprocal responsiveness to the communications of the other person; c. stereotyped and repetitive use of language or idiosyncratic use of words d. lack of varied spontaneous make-believe play or (when young) social imitative play
Issues You can constrain a clinician to use a particular term you can t control what he or she uses it to mean Advantages of consistency undermined by loss of expressive power key issue is granularity another is the importance of context and process Systems exploit the logical structure of clinical concepts but the logic always has exceptions, clinical language is more systematic than natural language but it isn t completely systematic
Controlled clinical terminology Clinical terminology concerns the meaning, expression and use of concepts in statements in the medical record or other clinical information system a concept is a unit of thought, an idea expressed in language A controlled clinical terminology is a system which allows the regulated and constrained expression of clinical concepts
Taxonomy Top Domain Eurkarya Bacteria Kingdom Animals Plants Phylum Arthropoda Chordata Magnoliophyta Subphylum Vertebrata Class Order Family Genus Species Aves Mammalia Primates Hominidae Homo Homo sapiens
Taxonomy: Aristotelian categorization Categories are discrete mutually exclusive collectively exhaustive defined by a set of attributes that the members share symmetrical all members of a category are equally members of the category
Taxonomy: concepts in real life Categories are not discrete not exclusive not everything is classified often defined by family resemblance or proximity to a prototype not symmetrical some members of a category are further from the prototype
Different ways of cutting up things Different approaches to modelling the world
ICD-10 International Classification of Diseases sponsored by WHO originated in 19 th century as a classification of causes of death consists of a core and a set of ancillary tools Type: enumerative single hierarchy fixed depth
Kinds of coding system: single vs multiple hierarchies diseases cancers diseases of the respiratory system breast cancer lung cancer emphysema
Kinds of coding system: enumerative vs compositional enumerative list of everything that you might want to represent, with a separate symbol identified for each item compositional symbols corresponding to core concepts are combined with modifiers and qualifiers e.g. combine acute bacterial with septicaemia to represent acute bacterial septicaemia
ICD-10: 3 character core Twenty two chapters anatomical area, class of disease, related information Each chapter is assigned a set of codes Chapter 2 Neoplasms: C00-D48 Within chapters codes are divided into blocks Malignant neoplasms of specified sites: C00-75 Malignant neoplasms of digestive organs: C15-26 Optional 4 th character for further specification Slow viral infections of central nervous system: A81 Creutzfeld-Jacob disease: A81.0
Chapter Codes Description I A00-B99 Certain infectious and parasitic diseases II C00-D48 Neoplasms III D50-D89 Diseases of the blood, blood-forming organs and diseases of the immune system IV E00-E90 Endocrine, nutritional and metabolic disorders V F00-F99 Mental and behavioural disorders VI G00-G99 Diseases of the nervous system VII H00-H59 Diseases of the eye and adnexa VIII H60-H95 Diseases of the ear and mastoid process IX I00-I99 Diseases of the circulatory system X J00-J99 Diseases of the respiratory system XI K00-K93 Diseases of the digestive system XII L00-L99 Diseases of the skin and sub-cutaneous system XIII M00-M99 Diseases of the musculoskeletal system and connective tissue XIV N00-N99 Diseases of the genitourinary system XV O00-O99 Diseases of pregnancy, childbirth and puerperium XVI P00-P96 Certain conditions originating in the perinatal period XVII Q00-Q99 Congenital malformations, deformations and chromosomal abnormalities XVIII R00-R99 Symptoms, signs and abnormal clinical and lab findings not elsewhere classified XIX S00-T98 Injury, poisoning and certain other consequences of external causes XX V01-Y98 External causes of morbidity and mortality XXI Z00-Z99 Factors influencing health status and contact with services
ICD 10: 3 character core I21 Acute myocardial infarction Includes: myocardial infarction specified as acute or with a stated duration of 4 weeks (28 days) or less from onset Excludes: certain current complications following acute myocardial infarction ( I23.- ) myocardial infarction: old ( I25.2 ) specified as chronic or with a stated duration of more than 4 weeks (more than 28 days) from onset ( I25.8 ) subsequent ( I22.- ) postmyocardial infarction syndrome ( I24.1 )
ICD 10: issues Limited scope classification of diseases Constrained structure makes it hard to update Use of not otherwise specified/not elsewhere classified A81 Slow viral infections of central nervous system A81.0 Creutzfeld-Jacob disease A81.8 Other slow viral infections of central nervous system A81.9 Slow viral infections of central nervous system not otherwise specified J16: Pneumonia due to other infectious organism, not elsewhere classified P23: Congenital pneumonia
My favourite ICD 10 CODE Y34 Unspecified event, undetermined intent
SNOMED CT SNOMED: Systematized Nomenclature of Medicine begun by American College of Pathologists in 1965 SNOP evolved into SNOMED in 1974 in 1999, existing SNOMED terminology merged with Read Codes V.3 to create SNOMED CT
SNOMED CT The basic unit in SNOMED CT is the concept SNOMED CT contains over 450,000 concepts Each concept has : unique identifier set of descriptions definition
SNOMED CT: unique identifer a unique numerical code assigned to each concept concept Myocardial Infarction has unique identifier 22298006
SNOMED CT: descriptions The descriptions provided for concept id. 22298006 are: Fully specified name: Myocardial Infarction (disorder) Preferred term: Myocardial Infarction Synonym: Cardiac infarction Synonym: Heart attack Synonym: Infarction of heart Synonym: Infarto de miocardio
SNOMED CT: concept definition Allergic rhinitis due to pollen IS-A allergic rhinitis HAS-FINDING-SITE nose HAS-ASSOCIATED-MORPHOLOGY inflammation HAS-CAUSATIVE-AGENT pollen
SNOMED CT: concept definition allergic rhinitis IS-A nose HAS-FINDING-SITE allergic rhinitis due to pollen inflammation HAS-ASSOCIATED- MORPHOLOGY HAS-CAUSATIVE- AGENT pollen
My favourite SNOMED CODE 216246008 sucked into aircraft jet, without accident to aircraft, occupant of spacecraft injured
Match the Code to the Picture
Match the Code to the Picture 300928006 vegetarian 159385002 office manager 256311001 apple 76799001 homemade bomb 257593005 cricket ground 13197004 contraception 247870000 fear of the bogey man 312635000 provision of telephone 24851008 DNA 37763007 parasite 160634009 competitive athlete 46930001 constructional steel erector
Issues with SNOMED Pre and post coordinated concepts Situations with explicit context Representing parts and wholes
The exploding bicycle 1972 ICD-9 (E826) 8 READ-2 (T30..) 81 READ-3 87 SNOMED CT 287
Some examples: pedal cycle accident involving fall from pedal cycle, occupant of animal-drawn vehicle injured non-traffic accident involving breakage of part of motor-driven snow vehicle, pedal cyclist injured motor vehicle traffic accident involving collision between motor vehicle and tramcar, pedal cyclist injured
Defusing the exploding bicycle: 287 codes in pieces things to hit pedestrian / cycle / motorbike / car / HGV / train / unpowered vehicle / a tree / other roles for the injured driving / passenger / cyclist / getting in / other activities when injured resting / at work / sporting / at leisure / other contexts In traffic / not in traffic
A different system would allow needed terms to be constructed as required Structured Data Entry File Edit Help Cycling Accident What you hit Your Role Activity Location
Combinatorial explosion to build an application clinicians can use to find the right codes, we want access to the logic used to create these codes SNOMED defines a compositional logic that could allow this but: most SNOMED concepts are asserted not defined many more are pre coordinated
SNOMED CT: pre and post co-ordinated concepts Not all concepts that an application might require are pre-coordinated, even in SNOMED In theory applications developers can define new concepts by post-coordinating pairs of concepts using SNOMED relationships to create new concepts breast cancer has location upper outer quadrant Some SNOMED concepts are associated with qualifiers for sanctioned refinement
Issue: situations with explicit context A concept includes context if: the definition of the concept is composed of a kernel concept that is modified in some way and the particular modification has the specific effect that the resulting concept is not a specialisation of the kernel concept
Issue: situations with explicit context History of fracture is not a subtype of fracture so history represents context
Issue: situations with explicit context Planned hip replacement is not a subtype of hip replacement so planned represents context
Issue: situations with explicit context A disorder might be recorded as a: possible diagnosis, part of a differential diagnosis diagnosis applied to some person specifically excluded diagnosis diagnosis now known to be incorrect but which justified a course of action which has to be recorded diagnosis that a patient believes they might have How much context can and should we record around a term? How much of this is the role of the terminology, and how much is to do with the application?
Taxonomies and Partonomies disorder of nasal cavity IS-A rhinitis IS-A allergic rhinitis IS-A PART-OF PART-OF PART-OF nose nasal cavity head body allergic rhinitis due to pollen
Issue: representing parts and wholes SNOMED represents anatomical structures as SEP triples S: structure E: entire structure P: part of the structure
Issue: representing parts and wholes SEP triples allow an IS-A hierarchy to include facts we want to represent about parts and wholes
Conclusion 1: coded data still need to be interpreted In last year s HES summary statistics 785,263 inpatient episodes were coded under Obstetrics, of which 16,992 were recorded for male patients
Electronic Patient Record: the container metaphor clinical codes are static, unitary tokens that we drop into a passive receptacle, to be retrieved later A record of work that is done elsewhere
Hospital Standardised Mortality Ratio From NHS Hospital Episode Statistics calculate risk of dying based on: Age group (<1, then 5-year bands to 90+) Sex Admission method/type (emergency, elective etc) Admission source (home, transfers etc) Deprivation quintile (based on postcode) Diagnosis subgroup Comorbidity Emergency admissions in previous 12 months Palliative care (any episode that has a treatment function code 315 or any Z515 code) Month of admission Year of discharge Can then calculate the Expected Mortality for a hospital, using data about the patients it cares for
Problems with HSMR 3: gaming the metric Coding of palliative care at Mid-Staffs
Conclusion 2: dangers of ontological thinking Terminology systems involve the creation of conceptual models or ontologies Design independent of a use case Terminology is an awkward match between ontology and language Problems generally occur where language is illogical
Worked example This nice gentleman was brought into A&E having been scythed down by his foolhardy son at the velodrome. Extreme pain on right shoulder, bruising and minor cuts. Radiology confirmed a displaced fracture of the medial third of the right clavicle. Prescribed ibuprofen, codeine and paracetamol. Discharged with a sling and referred to fracture clinic.