Mapping working example

Manually mapping hundreds of fields and values can be extremely laborious and prone to error.

Mapping suggestion algorithms offer machine assisted hints to manually selected mappings.

The algorithms work by starting with a context provided by the application, and via a series of iterations, narrow down the options to a small number. The algorithms are further tuned for specific patterns found in some source fields and values, and perhaps some authoring conventions when the target concepts were created.

In some cases, confidence levels are high enough to assume a single match (equivalent class axiom) and in this case it would be expected that a user validated a mapping once matched

Table and field hints

Take the following working example

{"Mapping": {
      "Source": {
        "Provider": "Barts",
        "System": "CernerMillenium",
        "Context": {
          "id": 1,
          "Table": "AdmittedPatientCare",
          "Field": "PatientClassificationCode",
          "Value": 1}}}}}

The context provided consists of the fact that it is Barts hospital, Cerner Millenum system and that a file has been provided that is loaded int a Table "admitted patient care" for the purposes of further analysis. This is documented on the NHS Data Dictionary web site. A field with a value of 1 needs to be mapped.

At the start, the mapping knowledge base has no knowledge of the nature of the table and thus a user would be presented with the data model.

The likely entity in the data model would be an Encounter and on further examination a number of specialised encounter types are available for selection. In particular a hospital in patient stay.

On further examination of the model it appears that the hospital in patient stay has a sub-component for a hospital admission entry

Its beginning to look like 2 entities are involved in the model. From the NHS data dictionary specification it appears that the field "Patient classification code" is part of the admission of the patient, which is a sub encounter of the in patient encounter.

So far, no information has been provided to the application to help context. However, a picture is building up.

Examining the admission encounter model further reveals a property of "admission classification of patient", which has a value range that includes 'ordinary admission' which has a comment suggesting that this matches to CDS code 1 for that field.

A user then selects this.