Introduction

The variety of available data standards/sources exhibits the entities' semantics in different manners. The most common types of semantics reflexion utilized by ontologies are Pre-coordination and Post-coordination.

Pre-coordination

When complex clinical expressions representing two or more concepts or a concept and modifiers occur commonly in clinical settings, it becomes more efficient to create one code that represents both concepts. The approach allows coding the complex multiaxial terms as 1 and the only explicit concept. Naturally not all the events that occurred in the real world may be reflected this way: the challenge of pre-coordination is a “terminology explosion” as additional codes are created from core concept and modifier codes to create long lists of concepts.

Post-coordination

Problem statement: A lot of source concepts are complex and are heavily pre-coordinated, often because they were not originally meant for describing a clinical fact but for billing purposes. Common examples are Codes from the ICD10 family of vocabularies or procedure codes. If that source concept combines its individual attributes with an AND combination, one solution is to split it up into multiple concepts and map the source concept to all of them as targets.

The approach allows coding the complex terms as a combination of several facts/concepts. The codes that were “post-coordinated” to create a single clinical expression demonstrate significantly greater value than when only the base concept is used. The approach may have several representations:

• a) the creation of pairs of originally independent facts, which occur at the same timepoint
• b) the creation of dependent pairs such as questions and answers (for laboratory test results, questionnaires).

However, this can also create problems for analysis as described here. If all targets are put in one concept set, one would not find only the actual source but all those which share one of the targets (because a concept set is by default treated as an OR combination). Finding them requires building a cohort with an AND combination and grouping them by timestamp.

Therefore knowing about these pitfalls is very important to achieve accurate results.

Source codes with attributes combined as OR statements almost always end up either as being mapped to less specific codes, becoming standard themselves (e.g. CPT) or being mapped to precoordinated concepts.

OMOP CDM Perspective

Both modeling methods are widely used in OMOP CDM and permit the end-user to define the phenotypes of interest in the most relevant ways. The pre-coordination is a prerogative of vocabularies enrichment, while the post-coordination is a modality available due to CDM Structure and ETL rules application. Classical CDM tables allowing the dependent facts to be stored in are Measurement and Observation. Tha systemic (CDM based) ways to post-coordinate are: FACT_RELATIONSHIP table population,modifier_of_event_id (measurement_event_id and observation_event_id) with mandatory population of modifier_of_field_concept_id ( meas_event_field_concept_id , obs_event_field_concept_id).

Examples of the Modeling approach

Most frequent issues may be stored in different ways described in the tables below. These include the laboratory finding representation, the coverage of historical facts, and independent simultaneous fact representation.

Example 1 - Laboratory Finding

Laboratory tests are usually specified as a combination of the sample, analyte, and its status. The status in different ontologies may be reflected as either part of a pre-coordinated entity or a dependent answer for the test.

Table 1 - Laboratory Finding

Source	concept_id	concept_code	concept_name	concept_class_id	standard_concept	invalid_reason	domain_id	vocabulary_id	modeling type	entity type	depedency type
CRP elevated	46234770	76485-2	C reactive protein [Moles/volume] in Serum or Plasma	Lab Test	Standard	Valid	Measurement	LOINC	post-coordination	fact#1	Q
CRP elevated	1620380	LA32146-5	Elevated	Answer	Standard	Valid	Meas Value	LOINC	post-coordination	fact#1.1	A
CRP elevated	37108742	119971000119104	Elevated C-reactive protein	Clinical Finding	Standard	Valid	Condition	SNOMED	pre-coordination

Example 2 - Historical Context

Historical findings, elements that exhibit details about previously diagnosed conditions, facts about family history, and usually dated only by the visit date. In case of the absence of pre-coordinated entities in standard OMOP vocabularies the possible solution is to post-coordinate as question (Q) and answer(A)

Table 2 - Historical Context

Source	concept_id	concept_code	concept_name	concept_class_id	standard_concept	invalid_reason	domain_id	vocabulary_id	modeling type	entity type	depedency type
h/o of MI	4214956	417662000	History of clinical finding in subject	Context-dependent	Standard	Valid	Observation	SNOMED	post-coordination	fact#1	Q
h/oof MI	4329847	22298006	Myocardial infarction	Clinical Finding	Standard	Valid	Condition	SNOMED	post-coordination	fact#1.1	A
h/o of MI	4163874	399211009	History of myocardial infarction	Context-dependent	Standard	Valid	Observation	SNOMED	pre-coordination

Example 3 - Complex Facts

The composition of several facts is a common scenario mentioned as post-coordination for primary multiaxial source expressions.

Table 3 - Multiaxial concepts representation

Source	concept_id	concept_code	concept_name	concept_class_id	standard_concept	invalid_reason	domain_id	vocabulary_id	modeling type	entity type	dependency type
Chronic r. iliac DVT	762256003	16026431000119101	Thrombosis of iliac vein	Clinical Finding	Standard	Valid	Condition	SNOMED	post-coordination	fact#1
Chronic r. iliac DVT	762422	350341000119102	Chronic deep venous thrombosis of right lower extremity	Clinical Finding	Standard	Valid	Condition	SNOMED	post-coordination	fact#2
Chronic r. iliac DVT	35616025	293451000119102	Chronic deep vein thrombosis of right iliac vein	Clinical Finding	Standard	Valid	Condition	SNOMED	pre-coordination