dataset using a FHIR terminology server Matt Cor Cordell ll Ter - - PowerPoint PPT Presentation

An introduction to analysing a SNOMED CT coded dataset using a FHIR terminology server

Matt Cor Cordell ll Ter erminolo logy Spe Specialist

A quick introduction to SNOMED CT, FHIR & Ontoserver

SNOMED CT

• Much larger than most other code systems traditionally used in healthcare (ICD, ICPC etc.)
• Primary purpose is recording clinical notes, with the specificity required by clinicians, and interoperability

– Structure* supports secondary uses (analytics).

• Codes have no intrinsic meaning, simply identifiers. 278285008|Left hemiplegia| & 278284007|Right

hemiplegia|

• Concepts in the terminology are associated by range of relationships, forming an Ontology.
• Expression Constraint Language (ECL) – language that supports sophisticated queries against the terminology.

FHIR

• Latest Interoperability standard from HL7, supporting modern RESTful practices. (ValueSets)

Ontoserver

• Provides FHIR based access to terminology, including ECL support
• Made available for use throughout Australia via the National Clinical Terminology Service (NCTS)

ECL in 90 seconds

<396234004|Infective arthritis| All (Subtypes) of Infective arthritis <64572001|Disease|:116676008|Associate d morphology|=23583003|Inflammation| All Diseases associated with inflamation <928000|Musculoskeletal disorder|:246075003|Causative agent|=<<49872002|Virus| Musculoskeletal disorders with some Viral involvement

What might a SNOMED CT dataset look like?

Unique Conditions : 24647 Unique Medications: 10128 Rows : 500,000 * Randomly generated synthetic dataset

Index Sex DoB PostCode Condition Medication F 26/04/1998 B03 102930000 7086011000036102 1 F 24/01/1953 E00 49512000 1112071000168105 2 M 7/09/1943 E00 277627005 5604011000036100 3 M 1/01/1966 E00 3109008 3231000036108 4 F 14/02/1957 E00 723409007 6286011000036105 5 M 14/08/1961 E00 3272007 761951000168100 6 F 28/01/1986 C04 86225009 921045011000036104 7 F 15/06/1983 C04 163577001 NaN 8 F 23/05/1967 C04 191737008 927853011000036101 … … … … … … 499998 M 16/01/1984 B09 443919007 36227011000036103 499999 M 28/03/1995 B09 723913009 5081011000036108

Basic outline of approach to SNOMED CT analytics

• Define aggregation categories using SNOMED CT Expression Constraint Language (ECL)
• Identify all the codes that match our category, using Ontoserver to perform valueSet Expansions.
• Store the results of each expansion in a Hash Set for fast lookup.
• Use the Sets to filter our dataset, and optionally create human readable labels.
• Use standard analytic approaches to report and visualise the data.

Populate Set with ECL

• Create a GET request with the ECL parameter
• Parse the JSON response to a FHIR Value Set
• Iterate through the Value Set and populate

the Hash with just the codes.

• Return the Hash.

import requests #for Rest calls from fhir.resources.valueset import ValueSet def PopulateSetWithECL(ecl): endpoint= https://ontoserver.csiro.au/stu3-latest expandAPI="/ValueSet/$expand“ sctValueSetUrl='http://snomed.info/sct?fhir_vs=ecl/’ urlParam={'url':sctValueSetUrl+ecl} response=requests.get(endpoint+expandAPI,params=urlParam) j=response.json() vs=ValueSet(j) _set=set() for e in vs.expansion.contains: _set.add(e.code) return _set

Creating Health Condition Labels

• A list of tuples, each tuple consisting of an ECL definition and label
• Iterate through this list
• Create the Hash Set based of the ECL
• Create Boolean filter for concepts that match the Set
• Label accordingly in a new “Category” column.

healthCategories=[ ('<<106028002','Musculoskeletal problems’), ('<<106048009','Respiratory problems’), ('<<195967001','Asthma’), ('<<363346000','Cancer’), ('<<13645005','COPD’), ('<<73211009','Diabetes mellitus’), ('<<106063007','Cardiovascular problems’), ('<<249578005','Kidney problems’), ('<<74732009','Mental illness’), ('<<40733004','Infectious disease’), ('<<414022008','Blood disease’)] for category in healthCategories: categorySet = PopulateSetWithECL(category[0]) filter = codeSet["Condition"].isin(categorySet) codeSet.loc[filter,"Category"]=category[1]

Index Sex Condition Medication Category F 102930000 7086011000036102 Other Condition 1 F 49512000 1112071000168105 Mental illness 2 M 277627005 5604011000036100 Cancer … … … … … 499998 M 443919007 36227011000036103 Mental illness 499999 M 723913009 5081011000036108 Mental illness

codeSet.groupby(['Category','Sex']).size()

Category Sex Count Blood disease F 7741 M 3295 Cancer F 1909 M 3298 Cardiovascular problems F 13716 M 10481 Diabetes mellitus F 18463 M 10362 Infectious disease F 1435 M 368 Kidney problems F 531 M 356 Mental illness F 106980 M 104910 Musculoskeletal problems F 1817 M 1400 Other Condition F 107163 M 105340 Respiratory problems F 230 M 205

Category Overlap

Overlap managed by:

• Categories ordered by priority
• Later categories overwrite; or
• Only label unlabled
• Build disjointness into ECL

<<106048009|Respiratory| Minus ( <<363346000|Cancer| OR <<106028002|Musculoskeletal| OR <<40733004|Infectious ) Use case dependent, especially where double counting

Counting Opioids

• Again, iterate through this list as before, adding an “Opioid”

label

• pioids= [('<34841011000036108','dihydrocodeine'),

('<21821011000036104','codeine'), ('<21705011000036108','pholcodine'), ('<21232011000036101','buprenorphine'), ('<21357011000036109','methadone'), ('<135971000036102','tapentadol'), ('<21258011000036102','fentanyl'), ('<21259011000036105','oxycodone’), … ('<21252011000036100','morphine'), ('<21486011000036105','tramadol'), ('<21901011000036101','dextropropoxyphene'), ('<34839011000036106','pethidine’), ('<1247191000168104','sufentanil')] for opioid in opioids: OpioidSet = PopulateSetWithECL(opioid[0]) filter = codeSet[“Medication"].isin(OpioidSet) codeSet.loc[filter,"Opioid"]= opioid[1]

Index Sex Medication Opioid 65 M 7349011000036100

• xycodone

219 M 1070441000168107 codeine 648 F 1048081000168105 buprenorphine ... ... ... ... 499738 F 34022011000036100 methadone 499802 M 785911000168101 fentanyl 499951 M 36062011000036104 dextropropoxyphene

Opioids

Using AMT’s “Concrete domain” in ECL

/*High Dose, 200mg or greater*/ <30497011000036103|medicinal product|: { 30364011000036101|has Au BoSS|=1817011000036100|aspirin|, 700000111000036105|Strength| >= #200, 177631000036102|has unit|=700000801000036102|mg/each| }, [1..1] 700000081000036101|has intended active ingredient|=ANY

53798011000036101|Ecotrin 650 mg enteric tablet|

/*Low Dose <200mg */ <30497011000036103|medicinal product|: { 30364011000036101|has Au BoSS|=1817011000036100|aspirin|, 700000111000036105|Strength| < #200, 177631000036102|has unit|=700000801000036102|mg/each| }, [1..1] 700000081000036101|has intended active ingredient|=ANY

/*Combination Aspirin Products*/ <21719011000036107| aspirin (MP)|: [2..*] 700000081000036101|has intended active ingredient|=ANY

“Concrete Domain” expansions

High Dose – 28 concepts

• Solprin 300 mg dispersible tablet
• Disprin Direct 300 mg chewable tablet
• Alka-Seltzer Lemon-Lime 324 mg effervescent tablet

Low Dose – 27 concepts

• Spren 100 mg tablet
• Cardasa 100 mg enteric tablet
• Aspirin Low Dose (Nyal) 100 mg enteric tablet

Combination Products – 54 concepts

• Clopidogrel/Aspirin 75/100 (AN) tablet
• Duoprel 75/100 tablet
• Action Cold and Flu effervescent tablet

Additional Resources

snomed.org/ecl

SNOMED CT ECL Specification

• ntoserver.csiro.au/shrimp

Shrimp Browser

github.com/AuDigitalHealth/ecl-examples

Agency ECL examples

bit.ly/SNOMED_HDA19

Supplementary Jupyter Notebook

Contact us

1300 901 001 help@digitalhealth.gov.au healthterminologies.gov.au twitter.com/AuDigitalHealth

Help Centre Website Twitter Email

OFFICIAL