Mapping and matching concepts

From Discovery Data Service
Jump to navigation Jump to search

Managing codes and taxonomies

Main article Management of code based taxonomies. Describes how codes such as EMIS local codes, Read 2, OPCS4 etc are handled in the information model applying the maps as described in this article.

Concepts - background

Information consists of ideas. Another word for an idea is a 'concept' . A concept may be named,( in which case the meaning of the concept can usually be understood), or they may be an unnamed expression, which is made up of a set of interrelated named or unnamed concepts.

For example the term "chest pain" implies the idea of a pain in the chest. In Snomed-CT it is a named concept. "Chest pain, worsened by exercise" may be an example of an expression style concept made up from the concept of "chest pain", and the statement that it is "made worse by -> exercise". In this case “made worse by” and “exercise” are both different concepts but no author has yet created a single named concept for this expression.

The new generation of health record management systems tend towards the recording of concepts, with the objective being for the record entry to closely match the idea behind the entry. These types of concepts can be called term based concepts as the term is the thing that describes the idea.

A modern term based concept is defined in relation to other concepts by a set of assertions indicating whether the concept is equivalent to, or a subtype of, a set of other concepts. The standard approach to this is via the use of Description Logic (DL). By using DL, a computer can automatically classify a concept which can result in a computer deducing additional knowledge over and above the human who created the concept. Snomed-CT is the worlds largest ontology of healthcare term based concepts and is authored using DL. A collection of concepts defined in this way constitute an "Ontology" and there is a standard language OWL that is used to represent the definitions.

Codes originated from a different starting point. The intention of a coded entry is to pre-classify an entry before it is recorded. The code is designed for a particular set of business processes e.g. analytics or payment and it is important to understand the context in which a code has been used. A coded concept, being pre-classified, relies on categorisation of the codes, and that classification may or may not imply that one code is a subtype of another. Nothing can be inferred from a code other than its relation to another code as authored. Consequently, as the philosophy is different, code based concepts have to be dealt with differently from term based concepts, even if they seem to saying the same thing.

Because of their history, it is not always possible to assert the exact meaning of a code. However, it is often the case that meaning can be inferred or approximated from a coded entry. With preference to move to an ontology, this inference can be achieved via the use of a mapping process that matches coded concepts to term based concepts that are identified from a code.

There are two strategies to link codes to concepts.

1. A coded term may be stated confidently to be the same as, or a variation on, a concept. Typically code systems like Read2 or CTV3 can be dealt with in this way because they are designed to try and capture the idea in the clinicians mind, and they have been incorporated as concepts anyway. Likewise many system supplier codes have been created in this way. In this case the term code can be said to be a term code of the concept. Read2 G33 - Angina pectoris is a term code for the concept of angina pectoris.

2. A coded term might be the same term as a concept but may have been entered without the assertion that is a true representation of a state. Typically code systems such as ICD10 and OPCS fall into this category. E11 - Diabetes type 2, seems to be the same as the concept of diabetes type 2, but was entered without clinician attestation and may have been approximated for payment purposes. In this case a legacy concept is produced and a map between this concept and the similar clinical concept is generated.

A map is just another form of relationship, but unlike an ontological equivalent or subclass axiom it implies that the relationship is an approximation. It is a sort of statement that something is possibly or probably similar to something else and thus has much less weight than an asserted relationship.

Legacy Code based concepts can be mapped to Core concepts , and this enables the use of the vast volumes of data already recorded in systems. Maps must be used with care as it is almost always the case that the use of a mapped code in a query is dependent on the purpose of the query. This means that mappings are more of a guide to the things to include rather than a confident statement of meaning. When querying records the query author may need to determine which codes to include or exclude on a case by case basis.

Code relationships to term based concepts

As mentioned above the relationships are managed as mappings which state the type or degree of match.

Maps generally fall into 4 patterns. These are illustrated in the context of code based concepts as follows:

Simple match

A concept may be matched to one other concept, the match having a certain weighting or category. For example the ICD10 code for Angina may have a map which maps to the single term based Snomed-CT concept of angina, with a category indicating that the source concept is properly classified. Note that many coded concepts may be mapped to one single term based concept. The map is viewed from the perspective of the coded concept.

N.B. In line with use of the health information modelling language based on RDF, Turtle syntax is used with the IRIs expanded by use of their name.

icd10:I209 |Angina Pectoris (ICD10 I20.9)| 
  :hasMap [ 
      :matchedTo [
        rdf:type sn:194828000 |Angina (disorder);
        :mapCategory sn:447637006 |Map source concept is properly classified;
       im:assuranceLevel im:NationallyAssuredUK

i.e. The ICD10 code I20.9 is matched to a single Snomed-CT concept.

Complex option match

A concept may be matched to a number of alternative concepts

icd10:E140| Unspecified diabetes mellitus with coma
           //This maps to a number of potential  target concepts
  :hasMap [
                 sn:26298008|Ketoacidotic coma due to.....,
                 sn:421725003|Hypoglycemic coma due to...,
                  sn:267384006 |Coma due to hypoglycemia]

Complex source match

A combination of concepts may be matched to a single target concept (e.g. A and B matches C) implying that the meaning of C should include all of the source concepts.

sn:6025007 |Laparoscopic appendectomy (procedure)]
:hasMap [
  :combinationOf [
     :oneOf opcs:H029| Unspecified other excision of appendix (OPCS49 H02),
            opcs:H021 | Interval appendicectomy (opcs49 H02.1),
            opcs:H028  Other specified other excision of appendix(opcs49 H02)
       opcs:Y752 | Laparoscopic approach to abdominal cavity NEC (opcs49 Y75.2),
       opcs: Y755 |Laparoscopic ultrasonic approach to abdominal cavity (opcs49 Y75.5)]

In other words a combination of one of the appendix excision OPCS codes and laparoscopic codes matches to the Snomed-CT concept of laparoscopic appendectomy. The matching objects also contain advice.

Axiom based target match.

A concept may be matched with a high level of confidence to an intersection of target concepts i.e. a concept expression. If the level of confidence is high enough and the context known, this could also be asserted as an axiom.

It should be noted that in this case, in the knowledge that the original code was authored with an ontological definition in mind that the above could be represented as an equivalent i.e.

emis:ALLERGY6183BRIDL | Adverse reaction to Mercilon
               [rdf:type sn:281647001 |Adverse reaction (disorder)],
               [rdf:type : owl:Restriction;
                owl:onProperty sn:246075003 | Causative agent (attribute);
                owl:someValuesFrom sn:9491701000001106|Mercilon (product)


Source resources properties and local codes

In the above examples, coded concepts were considered as context independent in the sense that the same code used by many providers and many systems would generally mean the same thing and can be treated the same way.

It is equally common to find provider and system specific constructs, including coded items whose meaning depends on the table or field within the source system. A similar approach to mapping of standard code schemes can be taken except that the source properties of the source concept must be explicitly described in order to provide context.

In the same way that codes can be mapped, so can source resource types such as tables or fields, message types or message segments. Mapping may involve functional transformation

Defining source context

The first step in managing source concepts is to define the concept in the context of the originator of the data. This employs the use of a context object.

         [:organisation :organisation/12345|Barts NHS Foundation Trust;
         :system :system/92223 | Cerner Millenium ;
         :resource :table/cds_type_130;
           :field :field/admin_cat_code| administrative category code on admission
       owl:equivalentClass nhsdm:administrative_category_code_on_admission.

Mapping nodes

A second step is to identify whether the source concept is equivalent to another source concept. This is done in order to rationalise the number of mappings steps needed between a source concept and the final target concept. For example:

         [:organisation :organisation/12345|Barts NHS Foundation Trust;
         :system :system/92223 | Cerner Millenium ;
         :resource :table/cds_type_130;
           :field :field/admin_cat_code| administrative category code on admission
       owl:equivalentClass nhsdm:administrative_category_code_on_admission.

Matching to ontological concept

In the above example we may have a number of different providers each providing different files or concepts, whose context suggests a match with the NHS Data model. As the Discovery data service includes the NHS data model attributes as part of its core model, the NHS datamodel is then mapped to the Discovery model.

              [rdf:type im:administrativeCategory| administrative category on admission]

The information model has fully defined the administrative category property as a property of a subclass of encounter dealing with hospital stays. Consequently the source system's table and field can be fully mapped to the common model field.