Their analysis raise a few questions however.
Firstly, the most basic type of semantic interoperability is that which enables one computer to correctly interpret a character string sent from another computer. "John Smith" for example, interpreted as the name
of a relative rather than the patient himself, or 02-19-1959 as a date of birth rather than a date of death. This is a modest but valuable type of semantic interoperability. It has been available for many decades in computer science, for example through semantic modelling in relational database management systems. As Dolin and Alschuler note, health informatics has yet to provide a truly reliable method for the semantic interoperability of structured data. They cite the example of the differential diagnosis of pneumonia. In fact, as William Hogan has shown, HL7 can struggle to reliably model even basic information such as marital status (http://bit.ly/fSRN70 ) and gender (http://bit.ly/iitzPL ). Ambiguity and complexity are hard-wired into the construction of HL7; these are not desirable features for a system intended to exchange meaningful information transparently.
Secondly, the meaning of a piece of data might not be self-contained but may depend on other data items. For example, a data point ("prescribed penicillin") could be interpreted as vital, or unnecessary, or harmful, depending on other data about the patient. Measures of semantic interoperability need to take into account the potential interdependency of data items in a record. Dolin and Alschuler discuss context sensitivity, but as far as I am aware there appears to be no recognition in any current HIT standards that the meaning of a piece of information may depend on the meaning of other pieces of information.
Thirdly, it is not clear how much of the meaning in a record is locked in unstructured data. This matters because even if all the structured data in a record could be exchanged accurately, a portion of the meaning of a record cannot be exchanged, at least not in machine oriented form. It could be argued that it is becoming possible to parse and auto code free text. But this is technically problematic. Doing so also highlights the largest challenge of all, which is that codification is always a translation, and translation risks losing, changing or adding to the semantics in the original text of a medical record.
In conclusion, Dolin and Alschuler's paper is very welcome, but it does not sufficiently challenge the assumptions about meaning made by messaging and coding standards for EMRs. These need to be examined, and a more sophisticated model of meaning within the EMR developed, before a framework for evaluation semantic interoperability between EMRs can be established.
Related Post:
Widget by [ Iptek-4u ]
