There were many strands in this workshop, and my reporting of it is inevitably yet another. That notwithstanding, here goes
'Coding' is not the best term: perhaps 'use of a controlled vocabulary' would be better, as coding was (as Read 3.1 rightly assumes) essentially a means of saving storage space, which is increasingly irrelevant these days, and is not a fundamental requirement of a representational system.
All agreed that the medical record, especially in primary care, is expressively rich, and that context is vital to its successful interpretation. Freedom of expression is a fundamental necessity for health care professionals, and the record can hold data that expresses several points of view - the GP's, the nurse's, the patient's, the carers. Its function inside the consultation was to enable:
Outside the consultation, it should provide data that could be aggregated in a meaningful way for use in-house and externally for:
It should also provide material for unambiguous communication between the author of the information and him- or herself at a later date, and other health care professionals, sometimes in other organisations (e.g. on referral and discharge). It was also a vital source of information for use in medicolegal disputes.
At the same time, there is an increasing need for software to be able to interpret at least some types of data in the record and to be able to reason with it, especially where software is expected to provide decision support of various sorts. Until text analysers are a lot more sophisticated and practical than they are now, there is a fundamental tension between the requirements to keep a comprehensive record, and one that non-human agencies could reason with. As dual recording is neither practical (nor desirable on a large scale, because of possible inconsistencies), 'trade offs' are the name of the game - what should we attempt to hold in a structured way, and what not?
Using a controlled vocabulary brings its own problems. Forcing users into recording something that is not (quite) what they wanted to say is not acceptable. Much, if not most, of the items in general practice records are not diagnoses, but events, actions (past, ongoing and future), signs, symptoms, and other, softer information, such as interpretations. So where a controlled vocabulary is used, it must be comprehensive, for example capable of representing negation, uncertainty and imprecision. On the other hand, it must also cater for specialist information provided by hospitals and others. It will be necessary to define at least some of the terms in a controlled vocabulary, and for users to have an explicit policy for using it, to ensure consistent use and interpretation. A controlled vocabulary should also be capable of expressing the relationships between concepts - e.g. 'I think this is the cause of that', 'This action is intended to manage that problem' - its context.
There is another tension between the ease of data entry, and that of storing it. Although structured data is needed in the record, it is not a natural form in which to conduct dialogue, and even to record interpretations of that dialogue (on the other hand, text entry via a keyboard is also pretty unintuitive for some, maybe many). The GP consultation is brief and difficult enough, without the intrusion of a pernickety terminology server when one attempts to enter data. In some areas, where the vocabulary is constrained and the sequence in which types of information are recorded is predictable, structured data entry is helpful and mirrors what occurs now - form-filling is the obvious example, entering prescriptions and lab results is another. In other phases of the consultation, for instance, those recording the patient's history, objectives and problems, and making observations in order to form a clinical view of the problem, the picture is much larger, and prediction at the top level of data entry is not possible (although it may be possible lower down, hence the idea of data entry 'templates'). Use of a controlled vocabulary should ideally be transparent to the user, and part of the responsibility for achieving this must lie with computer system and application interface designers.
As a general point, the design of the record display was critical to its satisfactory use, for example, the need for various views of it, summary displays, and so on: this applied whether data was structured or not, although it was easier automatically to allocate the data to a view if it was structured. GPs were good at spotting patterns, given a good view of the material containing the pattern(s). Bad record display could give rise to false interpretations.
As can be seen from the above, a controlled vocabulary was thought of as the bottom level of an overall design that also included the patient record. It was necessary for the design to be underpinned by an information model. Such a model would represent the processes leading to the record, and so the purpose for which the data was originally collected.