django-snomed-ct

A Django App for storing, managing, and querying the SNOMED Clinical Terms (SNOMED CT) system - an international medical terminology system and standard - by itself or as part of Django application.

Installation

django-snomed-ct can be installed via :

pip install django-snomed-ct

It currently works with Django versions 3.2+ but not 4+ and has been tested with SNOMED CT United States Edition version 20230301 (released on March 1, 2023).

As it is a separate, installable Django app that doesn't come with a Django project, you will either have to use it with an existing Django project or create one to use its Django management commands (load_snomed_ct_data and query_snomed_ct_data).

In either case, once you have it working with a Django project you will need to set up a database (in your project's settings.py) into which the SNOMED-CT content will be loaded.

For performance reasons, it is recommended that you use a database management system capable of handling queries against data at the size of the current SNOMED-CT distribution. In addition, you can also use django-snomed-ct with Django’s cache framework to reduce the cost of repeatedly looking up the fully-specified or preferred names of SNOMED-CT concepts. This is not required, but is advised.

See Setting up the cache for more information about how to configure Django's cache framework, but you will basically need the CACHES setting configured to have an entry for 'snomed_ct'.

Loading data

Once your Django project is configured to use a database and the database has been initialized, you can use the load_snomed_ct_data custom django-admin command to load a release into a configured database.
Below is an example of running the load_snomed_ct_data command (using the manage.py of a Django project) to load the 3/1/2023 US release, its ICD10 mappings and ISA transitive closure files, and assumes you have downloaded the SNOMED-CT distribution files into a ~/SCT-distribution/ directory.

• SnomedCT_ManagedServiceUSTransitiveClosure_PRODUCTION_US1000124_20230301T120000Z.zip
• SnomedCT_ManagedServiceUS_PRODUCTION_US1000124_20230301T120000Z.zip
• SnomedCT_ManagedServiceUSTransitiveClosure_PRODUCTION_US1000124_20230301T120000Z.zip

$ python manage.py load_snomed_ct_data --snapshot \
   --icd10_map_location=~/SCT-distribution/SNOMED_CT_to_ICD-10-CM_Resources_20230301.zip \
   --transitive_closure_location=~/SCT-distribution/SnomedCT_ManagedServiceUSTransitiveClosure_PRODUCTION_US1000124_20230301T120000Z.zip \
   ~/SCT-distribution/SnomedCT_ManagedServiceUS_PRODUCTION_US1000124_20230301T120000Z.zip

The --icd10_map_location option can be excluded if you don't want to load the ICD10 SNOMED-CT mappings

The --international option should be specified if working with an international distribution (otherwise, it defaults to assuming it is a US distribution). The --snapshot option should be used with a snapshot distribution (otherwise, it defaults to Full)

Simple functionality

from snomed_ct.models import Concept, ICD10_Mapping, TextDefinition, Description, ISA

You can fetch Concept instances by their SNOMED-CT identifiers:

>>> from snomed_ct.models import Concept, ICD10_Mapping, TextDefinition, Description, ISA
>>> from snomed_ct.models import Concept
>>> Concept.by_id('59820001')
<Concept: 59820001|Blood vessel structure (body structure)>

The custom manager for the Concept Model provides a by_ids method which takes a list of SNOMED-CT identifiers and returns a corresponding set of Concepts:

>>> Concept.objects.by_ids(['371627004', '194984004'])
<ConceptQuerySet [<Concept: 194984004|Aortic stenosis, non-rheumatic (disorder)>, <Concept: 371627004|Angiotensin converting enzyme inhibitor-aggravated angioedema (disorder)>]>

In addition to retrieving concepts by their identifiers, you also can fetch Concept instances via various types of string matching against their name:

• Regular expression matching (case sensitive or insensitive), using Django's regex and iregex field lookups respectively
• Text substring matching (case sensitive or insensitive), using Django's contains and icontains field lookups respectively
• PostgreSQL full text search, using Django's search lookup for PostgreSQL

The Concept custom manager also provides a by_fully_specified_name method. This method returns a queryset of Concepts matched by their fully specified name. The first argument is a search string pattern and the second optional keyword argument (search_type) can be one of the following attributes on the TextSearchTypes class in snomed_ct.models:

• CASE_INSENSITIVE_CONTAINS
• CASE_SENSITIVE_CONTAINS
• CASE_SENSITIVE_REGEX
• CASE_INSENSITIVE_REGEX
• POSTGRES_FULL_TEXT_SEARCH

Each of these corresponds to the 5 ways the string search will be performed. By default, and if not specified, the search will be performed using the case insensitive (CASE_INSENSITIVE_CONTAINS) substring matching method:

>>> Concept.objects.by_fully_specified_name('vessel structure')
<ConceptQuerySet [<Concept: 281471001|Abdominopelvic blood vessel structure (body structure)>, <Concept: 42586008|Large blood vessel structure (body structure)>, <Concept: 397018003|Blood vessel structure of skin (body structure)>, <Concept: 59820001|Blood vessel structure (body structure)>, <Concept: 306954006|Regional blood vessel structure (body structure)>]>

There is also a by_fully_specified_names method defined on the custom manager that can be used in a similar way to return a query set of Concept objects except that the first argument is a list of search string patterns to match against using the specified string search method.

Similarly, concepts can be fetched via matching definitions (the TextDefinition model) made about them by the Concept.by_definition class method, which has the same method signature and arguments as by_fully_specified_name .

for c in Concept.by_definition('aort.+stenosis', search_type=TextSearchTypes.CASE_SENSITIVE_REGEX):
..  print(c)

Concept query sets have a has_definitions method which filters the concepts to only include those that have textual definitions:

>>> Concept.objects.by_fully_specified_name('aortic stenosis').has_definitions()
<ConceptQuerySet [<Concept: 783096008|Subaortic stenosis and short stature syndrome (disorder)>]>

You can get direct access to the fully specified name (via the fully_specified_name property) of a Concept as well as the name without the parenthesized type suffix (fully_specified_name_no_type property)

concept = Concept.by_id(194733006)
concept.fully_specified_name_no_type
concept.fully_specified_name

Concept query sets also have a fully_specified_names_no_type property that can be used to return the list of fully specified names of its concepts:

>>> Concept.objects.by_fully_specified_name('vessel structure').fully_specified_names_no_type
['Blood vessel structure', 'Blood vessel structure of skin', 'Abdominopelvic blood vessel structure', 'Large blood vessel structure', 'Regional blood vessel structure']

Relationships about a given concept (concept definitions relating another concept to the one given) can be returned as an iterable of Relationship instances using the outbound_relationships() method and the same can be done in the other direction via the inbound_relationships() method on Concept model instances.

for rel in snomed_concept.outbound_relationships():
..  print("\t- {} -> {}".format(rel.type, rel.destination))

Hierarchies and Descriptions

You can navigate the ISA relationships of a given concept (General Concept Inclusion) via the isa Concept property, which returns an iterator of Concept instances related to the starting concept via ISA:

for general_concept in c.isa:
..  print("\t", general_concept)    

Conversely, there is also a specialization property that returns an iterator over other Concept instances that have an ISA relationship with the given concept:

for specialized_concept in c.specializations:
..  print("\t", specialized_concept)    

Similarly, you can iterate over the finding site concepts (via the finding_site property) as well as the associated morphologies (of disorders via the morphologies property) and everything a given concept is 'part of' using the part_of property, which returns an iterator of concepts related in this way. The has_part property returns everything that is 'part of' a given concept (the opposite relationship).

All the definitions for a concept object can also be returned as an iterator of TextDefinition model instances using the definitions() method.

for defn in c.definitions().values_list('term', flat=True):
..  print("\t", defn)    

All the descriptions for a concept object can be returned with the descriptions relation name:

for description in c.descriptions().values_list('term', flat=True):
..  print("\t", description)    

There are a few convenience properties available on QuerySets of Description and TextDefinition instances. In particular, the terms property will return a list of the term vales for each instance in the query set, the synonyms property will filter the query set to only include Description or TextDefinition instances that are synonyms:

description_text = list(c.descriptions().terms)
definition_text = list(c.definitions())
synonyms_text = list(c.definitions().synonyms)

Relationships

SNOMED-CT's Relationship relation is implemented in the Relation Django model. Every instance of this model has a source, destination, and type attribute, each of which corresponds to a Concept. These correspond, semantically, to OWL 2 (existential) Property Restrictions, where type is the property being restricted, source is the Concept whose definition restricts the use of the property, and destination is the Concept that the object of the property must be an instance of.

Each Concept instance has an outbound_relationships() and inbound_relationships() method that can be used to return a query set of Relationship model instances that have the concept as its source in the case of the former and its destination for the latter. Query sets of Relation model instances can be filtered in the usual manner but also define the following useful methods:

• sources()
• destinations()
• types()

They return a query set of Concepts that correspond to the source, destination, and type respectively of each Relationship instance in the query set

To learn more about the how SNOMED-CT uses the reasoning capabilities of Description Logic and OWL for its semantics, please read the SNOMED CT OWL Guide.

ICD 10 mapping

Once you have loaded ICD 10 <-> SNOMED-CT mappings via the --icd10_map_location option of the load_snomed_ct_data manage command, you can start finding mappings by ICD10 codes, iterating over just those SNOMED-CT concepts with ICD 10 mappings, etc.

Each ICD10_Mapping model represents a mapping from SNOMED-CT to ICD 10, where the reference_component attribute is the Concept being mapped to ICD-10, map_target is a string representation of the linked ICD 10 code(s), and map_rule is a string representation of a boolean that has a value of "TRUE" for a 'correct' mapping (see NLM's SNOMED CT to ICD-10-CM Map and ICD-10 Mapping Technical Guide .

Concept instances have an icd10_mappings attribute which will return a related manager for instances of ICD10_Mapping that can be filtered. For example:

>>> Concept.by_id('712832005').icd10_mappings.filter(map_rule='TRUE')
<ICD10_MappingQuerySet [<ICD10_Mapping: I10 -> 712832005|Supine hypertension (disorder)>]>

Query sets of ICD10_Mapping can be filtered to only those that are mapped to SNOMED-CT concepts with textual definitions with the has_definition() method and the get_icd_codes() method returns the ICD 10 codes (or code patterns) mapped by each instance in the query set. Their concepts property will return a query set of each Concept mapped in the query set.

Finally, the custom manager for the ICD10_Mapping model (ICD10_mapping.objects) has the following methods that each return a query set:

• by_icd_codes(codes) - Takes a list of ICD 10 codes and returns a query set of mappings whose ICD 10 value begins with any of the given codes (supporting ICD 10 code prefix matching)
• by_icd_names(search_strings, search_type=TextSearchTypes.CASE_INSENSITIVE_CONTAINS) - Returns mappings where the label of the corresponding ICD 10 code matches an item in the first argument (a list of string patterns) using the search method specified by the second argument
• by_fully_specified_name(search_string, search_type=TextSearchTypes.CASE_INSENSITIVE_CONTAINS) - Behaves similarly to the by_fully_specified_name method defined on the Concept custom manager except it returns a query set of mappings from concepts whose fully_specified_name matches the given pattern and search method

ISA Transitive Closure

Once the transitive closure file has been loaded (as described above), you can return all the concepts related to a given concept via the transitive_isa method which returns a TransitiveClosureQuerySet.

The TransitiveClosure model is the transitive closure of the 'subsumption' relationship (the inverse of ISA) materialized by the SNOMED CT Database Scripts.

Each TransitiveClosureQuerySet instance, has a general_concepts property that returns a ConceptQuerySet of all the general concepts (those that generalise the given concept) and a specific_concepts method which returns all the specific concepts from the transitive closure. So the following will return all the more general SNOMED-CT concepts that a given concept is derived from that also have ICD mappings:

icd_mapped_concepts = c.transitive_isa.general_concepts.has_icd10_mappings()

Once you have loaded the transitive closure and ICD mappings, given a concept instance, you can use the mehcanism for both to find all the ICD-10 codes mapped to concepts in the transitive closure of its ISA relation (i.e., all more general SNOMED-CT concepts)

This is a very useful way to discover ICD-10 codes relevant to a problem identified (possibly from its name or definition) by leveraging the logical expressiveness of the mathematics used to capture the meaning of SNOMED-CT terminology

query_snomed_ct_data Django management command

Below is the raw documentation of this command (the inherited options have been removed for brevity):

$ ./manage.py query_snomed_ct_data --help
usage: manage.py query_snomed_ct_data [-h] [-qt {ICD,ICD_CODE,SNOMED,SNOMED_CODE}] [-rx] [-d] [-r] [-o {relations,english,name}] 
                                      search_terms [search_terms ...]

Query SNOMED CT Release from the database.

positional arguments:
  search_terms

options:
  -h, --help            show this help message and exit
  -qt {ICD,ICD_CODE,SNOMED,SNOMED_CODE}, --query-type {ICD,ICD_CODE,SNOMED,SNOMED_CODE}
                        What the search is matching against: 'ICD' (name), 'ICD_CODE', 'SNOMED_CODE', or 'SNOMED'
  -rx, --regex          The query is a REGEX, otherwise it is a case-insensitive substring to match
  -d, --def-only        Ony show concepts with textual definitions
  -r, --logically-related
                        Show concepts logically related to those matched
  -o {relations,english,name}, --output-type {relations,english,name}
                        How to print the matching concepts: 'relations', 'english', 'name'

For example:

$ ./manage.py query_snomed_ct_data -qt SNOMED -o english "systolic essential hypertension"
--------------- 429457004|Systolic essential hypertension (disorder) ------------------------------
Systolic essential hypertension is an essential hypertension (59621000) and a systolic hypertension (56218007).  It is an interpretation of blood pressure (75367002) as increased (35105006).  It is located in some systemic circulatory system structure (51840005) and entire cardiovascular system (278198007)
ICD 10 Mappings: I10 (Essential (primary) hypertension)

Controlled Natural Language

The rendering of SNOMED-CT concepts the --output-type option is set to 'english' is done by the snomed_ct.controlled_natural_language.ControlledEnglishGenerator class. This class takes a Concept instance as the only argument to its constructor and has a get_controlled_english_definition() method with an embed_ids boolean keyword argument that defaults to False. It returns a controlled natural language representation of the concept's SNOMED-CT definition, embedding SNOMED-CT identifiers to referenced concepts when embed_ids is True and excluding these otherwise.

The rendering process is mostly based on Attempto Controlled English but also implements its own syntactic mechanisms specific to SNOMED-CT's semantics, such as part-whole reasoning to infer that "everything located in a part is located in the whole," morphological phrases, clinical drugs (dose units), pathophysiology, specimen, situations with explicit context ("a situation involving [..] and [..], a finding in [..]"), and dose forms ("is administered via, given by, administered as [..]"). ,