Talk:International Classification of Health Interventions

Comparison with SNOMED-CT is not at all informative, or accurate

I am going to delete the section on SNOMED-CT unless someone wants to rewrite it in a neutral POV, citing sources, and using quantitative comparisons when appropriate. Some of the statements seem like they, at one point, may have made sense in another language and were automatically translated w/ a very old-fashioned method that lacks domain knowledge and proper use of idioms. E.g. " ~2025-39912-09 (talk) 21:59, 10 December 2025 (UTC)

Support removal for weight and verifiability reasons. Little pob (talk) 19:47, 11 December 2025 (UTC)

Suggestion to add a section on ICHI's architecture and AI/LLM applications

Hello editors,

I would like to propose a small addition to the article to highlight a unique aspect of ICHI's design that has become highly relevant with the rise of Large Language Models (LLMs).

Full disclosure: I am the developer of an experimental project in this area, so I have a conflict of interest and will not edit the article directly. I am posting here to ask for the community's review and opinion.

The current article mentions ICHI's structure but does not fully explore the implications of its post-coordination architecture. Unlike traditional, lookup-based classifications, ICHI functions more like a grammar, with a defined vocabulary (the axis elements) and syntax (the coding rules). This compositional nature makes it uniquely suited for a new generation of AI tools.

LLMs excel at understanding and applying such rule-based, language-like systems. The ability to construct complex codes from a finite set of building blocks aligns perfectly with the generative capabilities of these models.

I believe it would be valuable for readers to understand this forward-looking aspect of ICHI's design. A new, brief section could be added to discuss this.

As a practical example of this concept, I have developed CodAs (an acronym for Coding Assistant), an experimental LLM-based tool. It is a non-commercial research project accessible via "ichi.codes" that attempts to generate post-coordinated ICHI codes from free-text descriptions by applying the ICHI Reference Guide rules. It serves as a proof of concept for the synergy between ICHI's structure and modern AI.

I leave it to the community to decide if this topic is notable enough for inclusion. I believe it highlights a key strength of ICHI and its relevance for the future of health information technology.

Thank you for your consideration.

Best regards,

Moveo ergo sum (talk) 14:58, 16 March 2026 (UTC)

The existence of such a tool isn't enough. Wikipedia needs reliable sources that ICHI is being used in such a way. Also, automated coding is probably better discussed at clinical coder than it is each and every medical classification.

While stand-alone tools are a good backup, my understanding is automated coding will more likely be built into EPR packages to comb through the clinical entries – either in real time or at discharge. And, given many EPR systems are used internationally, I imagine they'll be built for SNOMED CT. Doing so makes them classification agnostic as long as there are "maps" for that region's reporting classifications (examples, ICD-10 and OPCS-4 in the UK, ICD-10-CM and ICD-10-PCS in the USA).

Anecdotally, not least because I can only speak to the UK, automated clinical coding at scale has been 5-years away for at least the last 15-years. Modern LLMs suggest it could finally be ready – right as the NHS is in a funding crisis. Little pob (talk) 12:15, 17 March 2026 (UTC)

Many thanks for the thoughtful reply, Little pob. Point taken on all counts – especially the need for reliable sources and the suggestion that automated coding fits better at clinical coder than at each individual classification. The "5 years away for 15 years" observation gave me a good laugh.

I'll work on getting the project into a citable publication first and come back when there is something solid to reference. Cheers, Moveo ergo sum (talk) 19:20, 17 March 2026 (UTC)