User Tools

Site Tools


Mapping of Concepts

Mapping is the process to transform one Concept to another. The clinical data tables of the CDM allow only Standard Concepts. All other codes used in the source databases have to be translated to Standard Concepts.

Mapping is done through records in the CONCEPT_RELATIONSHIP table. They connect each Concept to a Standard Concept through a number of special relationship_id:

Relationship IDPurpose
Maps toMapping Concept to Standard Concept to be placed in the *_concept_id
Maps to valueSpecial mapping between a Concept and a Standard Concept to be placed into the value_as_concept field of the MEASUREMENT and OBSERVATION tables

"Maps to" Relationships

Concepts participating in the “Maps to” mapping are Source Concepts and Standard Concepts. The mapping attempts to map to the equivalent target Concept. Equivalent means it carries the same meaning, and, importantly, the children in the hierarchy (if there are any) are equivalent as well or cover the same semantic space. If an equivalent Concept is not available, the mapping attempts a match to a broader Concept. This ensures that a query in the target vocabulary will retrieve the same records as if they were queried in the original source vocabulary.

Generally, source Concepts and Standard Concepts are mapped:

  • Source Concepts concepts are mapped to one or several Standard Concepts. If they are mapped to more than one Standard Concept, then in the resulting CDM table more than one record is written for each record in the source.
  • Standard Concepts are also mapped to Standard Concepts, usually this is a map to itself.

Classification Concepts (standard_concept='S') do not have a mapping to a Standard Concept.

"Maps to value" Relationships

These relationships are designed to distinguish between Observation and Measurements, and their results. For example, ICD9CM V12.71 concept_id 44820383 “Personal history of peptic ulcer disease” has a relationship “Maps to” to SNOMED 4214956 “History of clinical finding in subject” (Observation Domain) and another relationship “Maps to value” to SNOMED 4027663 “Peptic ulcer” (Condition Domain).

"Losses" due to Mapping

There is significant concern about the effect of mapping from one vocabulary to another on the quality of the data and the ability to reliably identify patients for a cohort given inclusion and exclusion criteria. This effect can have a variety of reasons:

  • Mapping is missing. Please report missing mappings to the OHDSI Forum, so they can be added.
  • Source code is ill-defined, for example ICD9CM 799 “Other ill-defined and unknown causes of morbidity and mortality”
  • Negative assertion, for example ICD9CM V64.0 “Vaccination not carried out”
  • Code irrelevant to the patient, for example ICD9CM V65 “Other persons seeking consultation”
  • Source and target hierarchies incongruent

The latter effect is not uncommon for heavily pre-coordinated Concepts (complex and fine grained Concepts that are combinations of different dimensions), the coordination depends on the structure of topology of the vocabularies from and to which the Concept is mapped. If they are not equivalent, no direct mapping can be established. But in order to maintain the ability to retrieve these Concepts when searching using hierarchical Concepts, two or more mappings are provided instead.

For example, ICD10 Concept 45755355 “Non-insulin-dependent diabetes mellitus with coma” (code E10.0) cannot directly map to SNOMED:

Mapping of ICD10 E10.0

The target hierarchy in SNOMED has a different topology to the organizational structure in ICD10. Here, type 1 (insulin-dependent, E10) and type 2 (insulin-independent, E11) diabetes are the primary distinctive features. The combination of type 2 diabetes with its combination coma E11.0 defines type 2 diabetes with its complication coma. In SNOMED, diabetic complications has its own hierarchy, which branches into the Concept of diabetes with coma, or type 1 or 2 diabetes with complication. On the next level the distinctive forms of coma (hypoglycemic, hyperosmolar and ketoacidotic) are distinguished. However, there is no combination of diabetes type 2 with coma.

The solution is to map E11.0 to both 44054006 Type 2 diabetes mellitus and 420662003 Coma associated with diabetes mellitus. Even though there is no distinct equivalent in SNOMED, hierarchical queries for either diabetes 2 in combination with coma will retrieve the correct records.

Rules for Condition Mapping

In general, Conditions are records suggesting the presence of a disease or medical condition stated as a diagnosis, a sign or a symptom. However, there are a lot of conditions that are not originating in the biological function of the human body, but still require medical attention. Or they are true Conditions, but not at the time of the recording (History of) or in the person itself (Family history of). The mapping rules for these are as follows:

Aftercare after procedures. Those are mapped to a single concept 413467001 “Aftercare”, and with a second “Maps to value” relationship to the adequante procedure. For example, ICD10CM Z47.1 “Aftercare following joint replacement surgery” maps to both 413467001 “Aftercare” and to the procedure 4189532 “Implantation of joint prosthesis”. Note that “Aftercare” is itself not a condition, but belons to the Observation Domain.

Late effects or sequelae of other conditions. If possible, these are mapped to single concepts describing them as such late effects. For example, ICD10CM S82.874S concept_id 45589204 “Nondisplaced pilon fracture of right tibia, sequela” is mapped to 197150 “Late effect of fracture of lower extremities”. The Domain is Condition.

History of a condition. These are conditions, but not at the time of the condition_start_date. Therefore, they are mapped to the Concept 4214956 “History of clinical finding in subject”, which is in the Observation Domain. The condition itself is recorded through the “Maps to value” relationship. For example, ICD10CM Z87.820 concept_id “Personal history of traumatic brain injury” maps to 4214956 “History of clinical finding in subject” and with “Maps to value” 4132546 “Traumatic brain injury”.

Family history of a condition. These, too, have a dual mapping to 4167217 “Family history of clinical finding” and a to the actual condition through a “Maps to value” relationship. For example, ICD10/ICD10CM Z80.0 concept_id 45542462 “Family history of malignant neoplasm of digestive organs” maps to 4167217 “Family history of clinical finding” as well as 443568 “Malignant neoplasm of gastrointestinal tract”.

History of medical treatment. Similary, these are mapped to Observation Concept 4207283 “History of drug therapy”, and through a “Maps to value” link to the actual therapy. For example, ICD10 Z92.0 concept_id 45605174 “Personal history of contraception” goes to 4207283 “History of drug therapy” history of drug therapy and 4027509 “Contraception”. If the exact Ingredient or Clinical/Branded Drug is known, the map would direct to these Drug Concepts. But that is not typical.

Adverse effect of medication. If possible, these are mapped to a direct equivalent. However, that usually does not exist, and then they are treated like the above history of medical treatments. For example, both ICD10CM T36.8X5A concept_id 45551127 “Adverse effect of other systemic antibiotics, initial encounter” and T36.8X5D “Adverse effect of other systemic antibiotics, subsequent encounter” map to 437191 “Antibacterial drug adverse reaction”, while ICD10CM T36.8X5S concept_id 45560654 “Adverse effect of other systemic antibiotics, sequela” maps to 4207283 “History of drug therapy”.

Underdosing of medication. This information is handled similarly to the adverse effects, and the equivalent SNOMED concepts generally do not exist. For example, ICD10CM T36.0X6A concept_id 45565479 “Underdosing of penicillins, initial encounter” and T36.0X6D concept_id 45565480 “Underdosing of penicillins, subsequent encounter” go to 40488434 “Medication dose too low”, while T36.0X6S concept_id 45565481 “Underdosing of penicillins, sequela” points at 4207283 “History of drug therapy”.

Status of organ absence or transplant/prostetic presence. Absent organs are due to a procedure having removed them (unless they are inborn conditions, which are mapped as such). Therefore, they are mapped to 4215685 “Past history of procedure” and the appropriate procedure that removed the organ through a “Maps to value” link. For example, ICD10CM Z94.0 concept_id 35225404 “Kidney transplant status” maps to 4215685 “Past history of procedure” and 4322471 “Transplant of kidney”. Z95.5 concept_id 35225418 “Presence of coronary angioplasty implant and graft” is pointed to 4215685 “Past history of procedure” and 4184832 “Coronary angioplasty”.

Pre-coordinated Concepts listing two or more semantic components through AND or OR. These concepts are treated with the following order of precedence:

  1. To an equivalent combination concept that is also hierarchically well connected
  2. To both components separately
  3. To the most common ancestor

Examples for these 3 possibilites are:

  1. ICD10 A01 concept_id 45576225 “Typhoid and paratyphoid fevers” has a single “Maps to” relationship to 4022808 “Human typhoid AND/OR paratyphoid fever”.
  2. IC10CM F12.22 concept_id 45591098 “Cannabis dependence with intoxication” has two “Maps to” relationships to 4052690 “Cannabis intoxication” and 440387 “Cannabis dependence”.
  3. ICD10 L02.0 concept_id 45596354 “Cutaneous abscess, furuncle and carbuncle of face” has a single “Maps to” relationship_id to 400082007 Disorder of skin of head.

Maternal care. Many conditions require attention not because of a condition of a pregnant woman but of the fetus. However, all such conditions are being mapped to the mother anyway. For example, ICD10 O35.6 concept_id 45567927 “Maternal care for (suspected) damage to fetus by radiation” has two “Maps to” relationships to 199553006 “Fetus with radiation damage” and 289908002 “pregnancy”. Both conditions are recorded with the mother.

Need for immunization. These Concepts are mapped to an Observation indicating such immunity gap. A second mapping with the relationship_id “Maps to value” is then directed to the Condition (represented as a SNOMED Concept) the immunization is inocculating against. Note that it is not mapped to the vaccine itself (which would be represented as a RxNorm Concept). For example, ICD10 Z23 concept_id 45556822 “Need for immunization against single bacterial diseases” maps to 170536002 “Vaccination required” and maps to value 87628006 “Bacterial infectious disease”.

Conditions indicating abnormal levels of a test. These are split into the Measurement and result Concepts. For example, ICD10 R77.1 concept_id 45553745 “Abnormality of globulin” has a “Maps to” relationship to the Measurement 4353510 “Globulin measurement” and a “Maps to value” relationship to 4135493 “Abnormal”.

Conditions resulting from the use of Devices. Adverse or other outcomes of the exposure to a device are mapped to 473023007 “Complication associated with device and with another “Maps to value” relationship to the device itself, if possible. For example, ICD10 Y73.0 concept_id 45590664 “Gastroenterology and urology devices associated with adverse incidents, diagnostic and monitoring devices” has a “Maps to value” link to 45767866 “Gastro-urological device”.

documentation/vocabulary/mapping.txt · Last modified: 2018/07/09 11:55 by dimshitc1