What is SNOMED CT good for ? Ole Terkelsen MD Ph.D. Danish National Board of Health Why is there a need for a clinical terminology? Electronic Health Records (EHRs) will be introduced in the hospitals in this decade In the paper records there have always been a demand for precise and detailed documentation about e.g. the patient's diagnosis and procedures performed in relation to the patient The same demands exists for EHRs The mentioned information can be written in "free text" – but will in this case not be much easier to find than information in paper records Sandefjord 2005 2 Why is there a need for a clinical terminology? If possible, it would be rational to structure the information – i.e. use codes in order to ease retrieval The primary demands to a coding system that could meet the demands would be it will have to be highly granulated or detailed in order to capture the clinical situations it will have to reflect the terms used in the clinics it will have to contain some kind of definitions What coding systems can meet such demands? Sandefjord 2005 3 Why can't we use classifications like ICD-10? ICD-10 is a statistical classification that often aggregate information at code level e.g. C49.0 Malignant neoplasm of connective and soft tissue of head, face and neck It is therefore not granulated enough There are no definitions C80 Malignant neoplasm without specification of site probably means "cancer" It is out of date C85.0 Lymphosarcoma probably means "malignant lymphoma" Sandefjord 2005 4 What terminologies are available? Clinical Terms ver. 3 ("Read Codes" v.3) SNOMED RT SNOMED CT Open Galen UMLS (Unified Medical Language System) is not a terminology but a collection of approximately 130 classifications and terminologies Sandefjord 2005 5 What is SNOMED CT? SNOMED CT is a merge and further development of SNOMED RT and Clinical Terms ver. 3 The largest coherent terminology covering the clinical domain Sandefjord 2005 6 A quick journey from the sources of SNOMED Clinical Terms RCGP Oxmis 1984 Read Code 4-byte (10,000) 1983 Read Code Mnemonics (500) 1965 SNOP 1978 SNOMED 45,000 1988 Read Code Version 2 (30,000) 2001 SNOMED RT (150,000) 1993 SNOMED International 130,000 NHS Clinical Terms Version 3 (250,000) 1993 SNOMED International 3.5 156,000 2002 SNOMED Clinical Terms Sandefjord 2005 (350,000) 7 What is SNOMED CT? Contains approximately 300.000 active concepts approximately 1 million terms (incl. synonyms) 1.5 million relations between the concepts Languages: English (US and UK), Spanish, German In use in: USA, soon in England (NHS), trails in Denmark and Argentina Sandefjord 2005 8 SNOMED CT's Top-level Hierarchies Sandefjord 2005 9 SNOMED CT database tables Sandefjord 2005 10 Concepts – table – 350,000 entries CONCEPTID 74400008 80146002 233604007 3716002 CONCEPTSTATUS FULLYSPECIFIEDNAME 0 Appendicitis (disorder) 0 Appendectomy (procedure) 0 Pneumonia (disorder) 0 Goiter (disorder) CTV3ID Xa9C4 X20Wz X100E X76FB SNOMEDID D5-46100 P1-57450 D2-0007F DB-80100 ISPRIMITIVE 0 0 1 1 69536005 113276009 14742008 1236009 0 0 0 0 Head structure (body structure) Intestinal structure (body structure) Large intestinal structure (body structure) Duodenal serosa (body structure) Xa1gv Xa1Fr Xa1Fv XU5xL T-D1100 T-50500 T-59000 T-58230 1 1 1 1 41146007 0 9861002 0 113861009 0 Bacterium (organism) Streptococcus pneumoniae (organism) Mycobacterium tuberculosis (organism) X79pY X73GQ XU3Q2 L-10000 L-25116 L-21907 1 1 1 373270004 0 Penicillin (substance) XUWFk C-0021D 1 17369002 123603008 123604002 123609007 0 0 0 0 Spontaneous abortion (disorder) Acute focal hepatitis (disorder) Toxic cirrhosis (disorder) Subacute glomerulonephritis (disorder) L04.. XU5xO X307V XU5xW D8-04100 D5-80300 D5-80390 D7-12102 1 1 1 1 12361006 0 Osteotomy of radius and ulna (procedure) XU5xY P1-16187 1 Sandefjord 2005 11 Descriptions – table – ca. 1 mio. synonyms DESCRIPTIONID DESC-STATUS 814894010 0 123558018 0 CONCEPTID 74400008 74400008 TERM Appendicitis (disorder) Appendicitis DESCRIPTIONTYPE 3 1 LANGUAGECODE en en 21274010 132967011 132973012 132972019 0 0 0 0 80146002 80146002 80146002 80146002 Appendectomy (procedure) Appendectomy Appendicectomy Excision of appendix 3 1 1 2 en en-US en-GB en 621810017 350049016 0 0 233604007 233604007 Pneumonia (disorder) Pneumonia 3 1 en en 3716002 3716002 3716002 3716002 3716002 3716002 3716002 3716002 3716002 3716002 Goiter (disorder) Goiter Goitre Struma - goiter Struma - goitre Swelling of thyroid gland Thyroid enlargement Enlargement of thyroid Struma of thyroid Thyromegaly 3 1 1 2 2 2 2 2 2 2 en en-US en-GB en-US en-GB en en en en en 768995016 0 7261017 0 7267018 0 486646013 0 486645012 0 486643017 0 486644011 0 7263019 0 7264013 0 Sandefjord 2005 7265014 0 12 Relationships – table – 1.5 million entries RELATIONSHIPID 521526024 556899029 462569022 1045543021 405306026 1800183029 1939511022 707803022 136924025 78981022 1752936025 372287021 2038091027 152634025 1919793025 859420029 20869021 210013026 181749023 Sandefjord 2005 CONCEPTID1 236209003 247994001 191910002 190570008 147235008 129709009 206126004 15410007 309574009 257819000 315369003 172363006 64614001 122210004 122279008 74319002 106424006 38169004 20628002 RELATIONSHIPTYPE 363704007 363714003 123005000 363698007 116680003 363714003 246075003 363704007 116680003 116680003 363714003 116680003 116680003 116680003 260686004 123005000 116680003 116680003 116680003 CONCEPTID2 181422007 47078008 362012001 77637002 363662004 278844005 373266007 30291003 118246004 129304002 302147001 172359004 39981009 104172004 129265001 361714009 236312003 106424006 106424006 Is a 13 The architecture of SNOMED CT ! Disorder A concept based terminology Tumor "Is a" relation Throat disease Inflammation Tonsillitis Lung disease Cancer Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 14 SNOMED CT as a multilingual terminology Fully specified name Appendectomy (procedure) Appendektomie (Verfahren) Apendicectomía (procedimiento) Appendektomi (procedure) All with the same conceptid: 80146002 Modified from David Markvell Sandefjord 2005 15 SNOMED CT as a multilingual terminology Preferred term Synonym Appendectomy Excision of appendix Appendicectomy Entfernung des Wurmfortsatzes Appendektomie Operative Entfernung des Appendix Apendicectomía Escisión del apéndice Appendectomi Operativ fjernelse af blindtarm Sandefjord 2005 16 SNOMED CT - relations Attribute relations Associated morphology (attribute) Has specimen (attribute) Specimen source morphology (attribute) Specimen source topography (attribute) Specimen source identity (attribute) Specimen procedure (attribute) Part of (attribute) Has active ingredient (attribute) Subject of information (attribute) Causative agent (attribute) Associated finding (attribute) Component (attribute) Onset (attribute) Severity (attribute) Occurrence (attribute) Episodicity (attribute) Revision status (attribute) Access (attribute) Approach (attribute) Method (attribute) Priority (attribute) Sandefjord 2005 Course (attribute) Using (attribute) Laterality (attribute) Finding site (attribute) Direct device (attribute) Direct morphology (attribute) Direct substance (attribute) Has focus (attribute) Has intent (attribute) Procedure site (attribute) Has definitional manifestation (attribute) Temporally follows (attribute) Indirect morphology (attribute) Has interpretation (attribute) Interprets (attribute) Associated etiologic finding (attribute) Access instrument (attribute) Recipient category (attribute) Specimen substance (attribute) Pathological process (attribute) 17 SNOMED CT – relations Appendectomy Bacterial meningitis is-a Operation on appendix is-a Infective meningitis is-a Partiel excision of large intestine is-a Bacterial infection of central nervous system procedure-site Appendix structure method Excision - Action finding-site Meninges structure associated-morphology Inflammation pathological process Infectious disease Causative-agent Bacterium (fully defined) The use of attribute relations follow specific rules (description logics) Sandefjord 2005 anatomical man 18 Do SNOMED CT meet the demands? It is highly granulated and detailed and can capture the clinical situations It do reflect the terms used in the clinics conclusion from clinical trail it does contain formal definitions Sandefjord 2005 19 What about statistics and DRG? Sandefjord 2005 20 Handling legacy systems Is it possible to map? what are the use cases? mapping from SNOMED CT to classifications? mapping from classifications to SNOMED CT? Is it possible to use EHR data directly? for statistics? for DRG/HRG? etc. Sandefjord 2005 21 Is it possible to map? what are the use cases? mapping from SNOMED CT to classifications? mapping from classifications to SNOMED CT? New EHR EPJ based on EPJ baseret på BEHR baseret BERH på BERH national patient registry (continuity care based) SNOMED CT codes Sandefjord 2005 mapning, converting and explicitreporting national patient registry (based on contact registration) statistics DRG quality research etc. (based on contact registration) Classification codes 22 Mapping from SNOMED CT to classifications Questions to be asked In the following slides ICD-10 is used as an example How is the structure/architecture of SNOMED CT ? How is the structure/architecture of ICD-10 ? Can they be aligned ? Sandefjord 2005 23 The architecture of SNOMED CT ! Disorder A concept based terminology Tumor Throat disease Inflammation Tonsillitis Lung disease Cancer Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 24 The architecture of ICD-10 The basic building blocks are categories Groups of up to 10 entries The two last mentioned are often XNN.8 Other . . . XNN.9 . . ., unspecified The categories are grouped under “headings” The headings are assembled in chapters Sandefjord 2005 25 The architecture of ICD-10 - Examples CODE A03 A03.0 A03.1 A03.2 A03.3 A03.8 A03.9 TEXT Shigellosis Shigellosis due to Shigella dysenteriae Shigellosis due to Shigella flexneri Shigellosis due to Shigella boydii Shigellosis due to Shigella sonnei Other shigellosis Shigellosis, unspecified Apparent rule: ICD-10 becomes “less specific” the higher the code number The three-character code is never reported to registers (at least not in Denmark) The XNN.8 and/or XNN.9 therefore appears as “top-level concepts” Sandefjord 2005 26 The architecture of ICD-10 – More examples Do this ”rule” hold ? A49.8 Other bacterial infections of unspecified site A49.9 Bacterial infection, unspecified B26 B26.0 B26.1 B26.2 B26.3 B26.8 B26.9 Mumps Mumps Mumps Mumps Mumps Mumps Mumps orchitis meningitis encephalitis pancreatitis with other complications without complication B34.8 Other viral infections of unspecified site B34.9 Viral infection, unspecified B99.9 Other and unspecified infectious diseases Again – apparently Sandefjord 2005 27 Proposed mechanism for mapping Disorder Tumor Inflammation Throat disease Lung disease C80.9 Cancer C80.9 Tonsillitis Pneumonia Benigne tumor in throat Throat cancer C80.9 C39.9 C34.9 Malignant neoplasm of bronchus or lung, unspecified C80.9 Malignant neoplasm without specification of site Lung cancer Create a 1:1 input mapping table Read ICD-10 backwards – and assign every ICD-10 code (map) to the concept and all its decendents Sandefjord 2005 28 The architecture of ICD-10 – More examples However, for some reason ICD-10 breaks its own rule A49.8 Other bacterial infections of unspecified site A49.9 Bacterial infection, unspecified A53 Other and unspecified syphilis A54 Gonococcal infection Solution: Identify the areas and re-run the algorithm for these selected areas Sandefjord 2005 29 Mapping from SNOMED CT to ICD-10 The “algoritm” was implemented on an Oracle database (program written in PL/SQL) Temporary result: Over 70.000 concepts – mainly disorders mapped This result can be refined When new versions of the terminology and/or the classification are released the program can be reexecuted Sandefjord 2005 30 Mapping from classifications to SNOMED CT Why map backwards? to get the primary table for mapping from SNOMED CT to classifications (the input table for the algorithm) to demonstrate a terminology's capability as an aggregation tool Sandefjord 2005 31 The architecture of a concept based terminology Disorder A polyhierarchal terminology One concept can have more than one supertype Tumour Throat disease Lung disease Cancer Inflammatory disorder Acute tonsillitis Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 32 The architecture of a concept based terminology Disorder The "is a" relations always points "upwards" Tumour Throat disease Lung disease Cancer Inflammatory disorder Acute tonsillitis Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 33 The architecture of a concept based terminology Disorder If the "is a" relation is used in "reverse" you can aggregate information (count) from any point (concept) downwards Tumor Throat disease Lung disease Cancer Inflammatory disorder Acute tonsillitis Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 34 The architecture of a concept based terminology Disorder If the "is a" relation is used in "reverse" you can aggregate information (count) from any point (concept) downwards Tumour Throat disease Lung disease Count cancers Cancer Inflammatory disorder Acute tonsillitis Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 35 The architecture of a concept based terminology Disorder If the "is a" relation is used in "reverse" you can aggregate information (count) from any point (concept) downwards Count "tumours" Tumour Throat disease Lung disease Cancer Inflammatory disorder Acute tonsillitis Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 36 The architecture of a concept based terminology Disorder If the "is a" relation is used in "reverse" you can aggregate information (count) from any point (concept) downwards Count "lung diseases" Tumour Throat disease Lung disease Cancer Inflammatory disorder Acute tonsillitis Pneumonia Benigne tumor in throat Sandefjord 2005 Throat cancer Lung cancer 37 There are several possibilities for selection of entry ("aggregation") points The mentioned terminologies contains many levels (they are "deep" not "flat") Each concept can be used as an "aggregation point" You can extract the list of concepts "below" a chosen point for review or "control" You can add or subtract chosen "subtrees" You can select via aggregation points in supporting hierarchies (e.g. anatomy or microbiology) Sandefjord 2005 38 While we are waiting for data recorded with codes from clinical terminologies The best way of showing the described mechanism is by collecting fine granulated coded information via an EHR Such information is currently not available However, disease - and procedure classifications have been in use for decades The classification codes can be mapped to terminologies "At the end of the day, a code is a code" Margo Imel Sandefjord 2005 39 Mapping of classification codes to a terminology Each classification code (in this example ICD-10 codes) is mapped to the corresponding terminology concept Disorder Tumour Throat disease Inflammation J39.0 Acute tonsillitis Sandefjord 2005 J18.9 When the ICD-10 codes are mapped to the terminology concept codes the terminology framework can be used as an aggregation tool Lung disease Cancer Pneumonia Benigne tumor in throat Throat cancer Lung cancer C34.9 40 Mapping of classification codes to a terminology Each classification code (in this example ICD-10 codes) is mapped to the corresponding terminology concept Disorder Tumour Throat disease Inflammation J39.0 Acute tonsillitis Sandefjord 2005 J18.9 This mechanism also works with concepts that only exists in the terminology – e.g. the concept "lung disease" that are not found in ICD-10 Lung disease Cancer Pneumonia Benigne tumor in throat Throat cancer Lung cancer C34.9 41 Mapping of classification codes to a terminology If a corresponding concept for a ICD-10 code does not exist this particular code mapped or linked to the concept in the terminology that corresponds to the nearest supertype Disorder Tumor Throat disease Lung disease Cancer Inflammatory disorder Abscess of pharynx J03.9 Retropharyngeal and parapharyngeal abscess Acute Tonsillitis J18.9 Pneumonia Benigne tumor in throat Throat cancer Lung cancer J39.0 Sandefjord 2005 42 Examples from the National Danish Patient Registrar On the following slides a few examples of aggregation of coded information based on the described method is shown The information is drawn from all outpatients and admitted patients in Denmark 2002 The information is recorded with ICD-10 codes partially mapped to SNOMED CT The aggregation points are SNOMED CT concepts shown in italics Sandefjord 2005 43 Data from NPR – ”aggregated” with SNOMED CT SNOMED CT concept in italics All types of pneumonia and viral pneumonia all admitted patients 2002, distributed by age 1400 1200 number 1000 800 600 400 200 Age 10 4 96 92 88 84 80 76 72 68 64 10 0 Sandefjord 2005 60 56 52 48 44 40 36 32 28 24 20 16 12 8 4 0 0 44 Data from NPR – ”aggregated” with SNOMED CT SNOMED CT concept in italics All types of complications to diabetes Outpatients and admitted patients 2002, distributed by age 1000 900 800 700 number 600 500 400 300 200 100 92 88 84 80 76 72 68 64 60 56 48 44 40 36 32 28 24 20 16 52 age 96 10 0 Sandefjord 2005 12 8 4 0 0 45 Data from NPR – ”aggregated” with SNOMED CT SNOMED CT concepts in italics Hereditary and congenital disease admitted patients 2002, distributed by age 2500 2000 number 1500 1000 500 age 96 10 0 92 88 84 80 76 72 68 64 60 56 52 48 44 40 36 32 28 24 20 12 8 4 0 Sandefjord 2005 16 0 46 Terminology as an aggregation tool Terminologies can be used as statistical aggregation tools It can be questioned if the mapping from a clinical terminology to a classification with the purpose of using the classification as the aggregation tool is practical in the future It is possible to link e.g. ICD codes into the terminology – and use this as an aggregation tool – both for analysing present day information and in the future for comparison of structured information collected from an EHR with present day coded registrar information Sandefjord 2005 47 Is it possible to use EHR data directly? - for statistics? apperantly! - for DRG/HRG? - etc. . . Sandefjord 2005 48 Can DRG/HRG groupings be found in SNOMED CT? 134 05 MED HYPERTENSION 46742003 skin ulcer* 127 05 MED HEART FAILURE & SHOCK 60168000 osteomyelitis* 271 09 MED SKIN ULCERS * include subtypes 238 08 MED OSTEOMYELITIS 38341003 hypertensive disorder* heart failure* + shock* Again apperantly However, the possibility of direct mapping from SNOMED CT to DRG/HRH should be analysed further 232 08 SURG ARTHROSCOPY 13714004 arthroscopy* Sandefjord 2005 49 Does a terminology give all the answers? Sandefjord 2005 Decision support 50 The Danish EHR model The steps of the clinical process Process Diagnostic Evaluation consideration Information Diagnosis Version The model 2.2isisa modified just released problem and solving is documented or quality in assurance text, use cases circle with UML and with health care terms and a "goal" added Sandefjord 2005 (Condition) Goal Planning Outcome Executing Plans 51 Model and Terminology Clinical Terminology The model requires highly structured input i.e. data types such as numbers, dates etc. and structured (preferably coded) clinical information e.g. from a terminology (including drugs) BEHR Sandefjord 2005 including information about location (hospital, department, etc.) user access (logging) 52 Can we use SNOMED CT? Process Diagnostic Evaluation consideration Information Diagnosis (Condition) Clinical finding A model is needed as a container for the information Sandefjord 2005 Goal Outcome (result) Observable entity Substance Planning Executing Plans Procedures 53 Comparing HL7 v3 with BEHR Diagnostic Evaluation consideration Diagnosis (Condition) Goal Planning Outcome (result) Executing Plans Sandefjord 2005 54 Sandefjord 2005 55