Thank you for this article that explores the topic of coding in medical records.1 I think there are deeper conceptual problems with coding than we realise.
First, coding is categorical, not probabilistic or Bayesian. Our structure of thought as GPs is Bayesian — we are trying to work out what is most likely given the medical and social background, and the current symptoms. Our conclusions are often tentative, and diagnosing and confidently naming skin rashes is an obvious example here. Especially as they can easily alter over time, and we then may need to revise our thinking.
For coding to become easier could I suggest that we could have symptom codes, examination finding codes, and diagnosis codes? For diagnosis codes we should be able to indicate (estimate) our degree of confidence in our diagnosis. It would be high when we have histology and scan results, or very clear symptoms or signs (for example, a clear shingles rash). It would be proportionately less when we are seeing a non-specific but likely viral rash.
Also, could codes be properly time bound? For example, an episode of anxiety coded many years ago is not evidence of a current ongoing problem. However, an insurance company or occupational review might pick up the code and assume it’s a current issue. The code can become the reality, rather than a reflection of the reality it was summarising.
Coding is not only very important and very useful, but it’s also potentially awkward and misleading. It’s hard to revise if the presentation evolves over time. Could we have a more subtle and less categorical system of coding? One that goes with the grain of the structure of medical thinking, rather than imposing categories over fluidity?
- © British Journal of General Practice 2024
