I think it is important to lay out some principles for the capture, storage, display and communication of information derived from diagnostic tests – commonly labelled in Australia as pathology and radiology. In particular, I’m concerned about maximising the reuse of information, both for direct patient care, as well as for research. In order to achieve optimal reuse and system to system interoperability of diagnostic information, it is necessary to consider information independent of any particular message or message syntax. It is important to describe the information in ways that make it amenable to reuse, such as identifying common patterns in data, or by categorising different types of information – context vs. findings. Identification of patterns, their standardisation, and their subsequent engineering into the infrastructure of clinical and communication systems is vital to safe, efficient and effective information management.
There are classic modelling techniques used by software architects and information modellers to help in this task of constructing patterns. Some of these include:-
- classification/categorisation/abstraction/naming of concepts, e.g. anatomical pathology, pathogens, testing methodologies.
- part whole relationships, e.g. a white cell count can be composed of counts of neutrophils, lymphocytes, monocytes, eosinophils, and basophils.
- aggregation and grouping – blood pressure measurement is associated with patient state (sitting/lying/at rest), as well as location of cuff and time of measurement; diastolic needs to be grouped with systolic. Another example is the staging of cancer using the 3 components Tumour information + lymph Node information + Metastatic information to build a TNM grading of a specific individual’s cancer at a given time.
- specialisation and generalisation – e.g. one assay might be a specialisation of a more common assay type.
- contextual textual terms, phrases, qualifiers, clauses – e.g. “no evidence of …”, ” considerable likelihood of …”, “severe …”
- properties of measure and units of measure – e.g. mass concentration vs concentration by volume. micrograms per litre vs milligrams per decilitre.
- constraint mechanisms – e.g. Colon cancer staging values might be constrained to the Duke’s Classification scheme
Identifying, constructing and using these patterns to manage, communicate and process clinical information makes sense because many of these pattern types are ubiquitous in modern object-oriented programming languages that will underpin clinical information systems for the foreseeable future. Thus, use of these patterns provides a necessary and consistent foundation for decision support.
Whilst structured clinical terminologies like SNOMED CT have been broadly accepted as a foundation for decision support and semantic interoperability, their more important counterparts, clinical information patterns and structures have largely been ignored. It is mainly due to a small group of pioneers, particularly the openEHR Foundation and their innovative two-level modelling approach of layering clinical patterns (archetypes) on top of generic information structures, that we have made any progress at all towards standardising clinical information sufficient for shared electronic health records and decision support. We need much more progress before we realise many of the potential significant safety, efficiency or healthcare improvements touted through the use of information technology.
Although much of the above can be applied right across the healthcare information spectrum, the particular complexities of pathology and radiology data make the documenting and standardising of their clinical patterns all the more important. We need to start now.
However, when we finally do start the task, it won’t be easy. It will make the 7-10 year, multi-hundred-million dollar struggle to introduce Australia’s very simple individual and healthcare provider numbers seem like a trivial stroll in the park! Can anyone see a need for some governance?