This briefing note seeks feedback on how to implement a new modelling approach and description format for 91302008 |Sepsis (disorder)| and its subtypes, while aiming to minimise disruption to users. These updates align content with the Sepsis-3 definition.
| Date created | February 6 2026 |
|---|---|
| Action | Feedback requested |
| Status | For comment |
| Disposition | Open |
| Feedback by | March 6 2026 |
BN Revision of Sepsis Modelling and Descriptions 20260602.pdf (285.4 KB)
Thanks.
Sepsis presents particular challenges to SNOMED CT and to the users of SNOMED CT. I do not believe that this proposal addresses the majority of challenges. Indeed, in the ‘next steps’ we are asked to address just one question: to decide between ‘option 1’ or ‘option 2’ - that is, whether SNOMED International should follow, or not, its own editorial guidelines on major and minor concept changes. Given so many other challenges I am curious as to why this question is given primacy.
Instead I would rather raise further points regarding the relationship between SNOMED CT, sepsis and sepsis variants.
1. The concept 91302008 |Sepsis| itself.
91302008 |Sepsis| has been present since SNOMED CT’s first release in 2002 (and no doubt in antecedent terminologies), roughly coinciding with the publication of Sepsis 2. At that time (and indeed until 2014) it had an FSN of ‘Systemic infection’ (which it still carries as an active synonym). It continued to be classified as a kind of ‘infectious disease’ until 2020.
We might, therefore, reasonably ask questions regarding the stability of 91302008’s semantics. Is it reasonable, at this stage, to impose upon a suspiciously ‘infectious’ concept (according to an active synonym and 20 years of modelling) a 2016 authority definition which seems to exclude infection?
The ancestry and descendants/extension of 91302008 |Sepsis| have been very changeable over the years. Is 91302008 the right code to take responsibility now? Do we really think stability will be achieved this time? This site prevents me from posting spreadsheets illustrating what has happened over time, but the ancestry animations posted here might help indicate the extent and nature of change.
2. 2016, Sepsis 3 and the perils of post-coordination.
In late 2016 the UK attempted to provide national guidance on how to use SNOMED CT to record sepsis and variants consistent with the newly-published Sepsis 3 document. There was insufficient time to undertake a reorganisation of international content so we worked with what was there. A lot of the guidance published related to immediate and pre-hospital management specified in a series of national guidance documents and toolkits developed in tandem with Sepsis 3’s release (reliant on phrases such as ‘Red Flag Sepsis’ and ‘Suspected sepsis’ - we have codes for these in our extension).
We did, however, also try to cover various patterns of in-hospital diagnostic coding using SNOMED CT, and took the modelling of Sepsis content at that time as a ‘pattern’ to inform our guidance. Guidance was provided for three compositional patterns, summarised in this table:
The changes proposed in the briefing note will modify the underlying reference data such that any expressions created in the last 10 years in line with the middle ‘Sepsis caused by infecting organism pattern will no longer be trivially detectable as equivalent with their pre-coordinated reference data counterparts.
I campaigned during the development of the SNOMED CT post-coordination guide to make clear to users the significant risks posed to post-coordination by model changes. I didn’t have much impact, and the result was this clause in the document:
“Maintenance processes need to be established to ensure that the expressions used in the system continue to work properly with the updated version of SNOMED CT.”
The passive nature of the language leaves it unclear where the responsibility lies for ‘establishing the maintenance processes’, so we may be on our own dealing with the impact of any sepsis changes.
However my preference would be for SNOMED International to consider the impact of changes designed to ‘improve the terminology’ on what users might actually be doing and factor this into their change ambitions. The inclusion of UK primary care usage data in this briefing note barely scratches the surface of ‘what people are doing’ (and IMHO I would question why they are included here).
Removing a modelling pattern that has been a constant feature of SNOMED CT for nearly 25 years is likely to have significant impact on the usability of compositional expressions created against that pattern.
Equally, whilst distancing ‘sepsis’ from any named initiating infective agent may have intellectual appeal, it may also have unintended consequences for pre-coordinated data retrieval. Queries of a ‘<< Infectious disorder’ form will largely have missed any ‘Infectious disorder causing sepsis’ content for the last couple of years. The proposed changes will guarantee this outcome. Unless such queries include dot notation query augmentation, property chaining for DL-based approaches, or users instinctively know to add a supplementary ‘<< sepsis’ clause, a proportion of in-scope cases may well be missed. I’m not sure how many users are prepared for this.
Of course, the argument can be made that an ‘Infectious disorder causing sepsis’ ceases to be an infectious disorder at all (it’s now become a ‘sepsis’), but how universal is this viewpoint? Until 2023 ‘bacterial sepsis’ was a kind of ‘infectious disorder’ - but no longer is - this intellectually compelling in the light of the Sepsis 3 definition, but perhaps practically problematic? [Meanwhile, elsewhere, we have 186893003 |Rupture of syphilitic cerebral aneurysm (disorder)| still being returned as a kind of ‘infectious disease’].
Would it not be preferable to factor in a balance between data retrieval false positives/false negatives and logical alignment with an international Sepsis definition? It is appealing to align with the Sepsis 3 definition, but are we really making SNOMED CT ‘better’ for the management of patients with sepsis by doing so? The briefing note states that “…Users will experience improved classification and more clinically accurate terminology…” - I’m not so sure they will.
3. Meningococcal disease.
I’d like to highlight SNOMED CT’s handling of ‘meningococcal sepsis’ in recent years - I think it is relevant to a broader consideration of the challenges and changes in the area of sepsis.
a. 4089001 |Meningococcemia (finding)| has been in SNOMED CT since 2002.
Until 2021 it carried a synonym of ‘meningococcal septicaemia’, and until 2023 it was classified as a kind of Sepsis, but since then has been relegated/reclassified as a blood finding/bacteraemia.
It is almost certainly still used in records to describe patients with severe life-threatening sepsis caused by meningococcus (and so should be trivially retrieved by queries for sepsis or infection), but in recent releases of SNOMED CT would be missed by both query patterns.
b. 127311000119106 |Sepsis due to infection caused by Neisseria meningitidis|
Somewhat overlapping these changes, 127311000119106 has appeared. It is now semantically remote from 4089001 but they appear to have served similar functions for much of this time. In its middle years it had an FSN of ‘Sepsis caused by meningococcus (disorder)’, with its modelling including a ‘causative agent’ relationship. Both the terming and modelling were changed in late 2025 in ways that seem to mirror those we are now being asked to approve.
The wilful ‘loss’ of 4089001 as a kind of sepsis/infection code and the premature remodelling/reterming of 127311000119106 are both concerning and may point to deeper challenges.
4. Code usage figures.
Returning to this point, I am concerned as to why UK primary care usage figures are included in this international briefing note? If they are truly being used to inform editorial decisions then the reasoning behind such decisions needs to be made explicit in editorial documentation. I understand the temptation to think usage data may help, but it is far from clear how this should happen.
If changes need to be made to a reference terminology on editorial grounds then they need to be made, whether a code has been used once or used a million times. If they are made in a way that satisfies Cimino’s concept permanence and graceful evolution desiderata then SNOMED CT users should be able to absorb the changes in a non-disruptive way.
Overall I would urge SI to think more about the part it wants SNOMED CT to play in the management of patients with sepsis. To me the changes proposed are at best incomplete, and at worst problematic. I appreciate that the status quo conflicts with the 2016 Sepsis 3 definitions, but simply changing the data in line with this may have analytic and classification consequences that will hamper SNOMED CT’s contribution to care.
Kind regards Ed
Hi Ed,
Thank you for your detailed and thoughtful response.
You’ve captured well the central tension we are grappling with: the intellectual appeal of closer alignment with contemporary clinical definitions versus the practical realities of data capture and retrieval. This is not a straightforward issue. As the global clinical community increasingly adopts the Sepsis-3 paradigm, we must consider whether SNOMED CT should retain legacy modelling primarily for continuity, or evolve to remain aligned with current medical science. The intent behind the proposed update was to explore how such alignment might help ensure these concepts remain clinically relevant; however, we do not want to underestimate the downstream consequences of change.
The topics raised in the briefing note originated from a Member request, and given the scale of the proposed inactivations, we felt it essential to publicise the work early and actively seek feedback on potential impact before progressing further.
Your observations regarding the historical evolution of 91302008 |Sepsis (disorder)| are entirely fair. Stability is fundamental in a reference terminology, and the extent of semantic and classificatory change this concept has undergone over time strongly suggests that inactivation and replacement may be the most appropriate path forward. Moving from a Sepsis-2 to a Sepsis-3 definition is not merely refinement; it represents a substantive shift in meaning. Providing new concept IDs aligned with the Sepsis-3 definition preserves the meaning of the original concepts as they were understood at the time the data was recorded, rather than forcing users to guess which definition was intended. This applies not only to 91302008 |Sepsis (disorder)| itself, but also to the existing “Sepsis caused by X organism” concepts, whose meanings are effectively changing under the newer definition. For the proposed inactivations, the inactivation reason would be “Outdated” and the historical association would be ‘replaced by’ to link to the new active concepts.
I also understand the concern regarding long-standing concepts such as 4089001 |Meningococcemia (disorder)| and how deeply embedded they are in historical use. This sits within the broader reality that the terminology must evolve as clinical definitions evolve. That said, I agree that simply re-terming or remodelling concepts such as 127311000119106 without inactivation and replacement risks introducing ambiguity. Where the meaning is substantively changing, making that change explicit through inactivation and replacement is likely the clearer approach.
Your points regarding retrieval impact and post-coordination are well taken. Some mitigation may be achievable through modelling decisions and implementation guidance, but we cannot fully control how the terminology has been used historically or will be used in future. What we can do is ensure that any evolution of the terminology is conceptually clear, internally consistent, and transparent, while providing the best possible guidance to support users through the transition. That said, data migration is never painless, no matter the approach taken.
Finally, thank you for your feedback on the framing of the “next steps” and the presentation of usage data in the briefing note. It is very helpful to understand how this was received, and we will take that into account in future communications.
Thank you again for engaging so constructively in this discussion. Your feedback is invaluable as we continue to weigh both the editorial and practical implications of potential changes in this area.
Kind regards, Sarah
Thanks.
A few comments in response:
I remain very concerned that the SI desire to “…adopt the Sepsis-3 paradigm…” will, far from giving users “…improved classification and more clinically accurate terminology…” result in a situation that actually hampers SNOMED CT’s contribution to care.
The application of an external authority definition to an already released notion (whilst somehow preserving older data) cannot be painlessly achieved with the limited mechanisms available to SNOMED CT. The highly predictable “inactivation as ‘outdated’ and replacement with” path proposed solves very little. Paradoxically, given 91302008 |Sepsis| has proved so enduring (surviving several authority definition changes, an FSN change and a lot of classification moves) I’d be inclined to leave it active but use other mechanisms to point out how troublesome its semantics are to tie down.
Many SNOMED CT concepts acquire novel meaning in use (the ‘Humpty Dumpty principle’). If I want to record ‘sepsis’ SNOMED already provides me with a code - why would I need a new one? It’s up to me, communicating partners and collaborating research partners now and through time to demonstrate what I meant by ‘sepsis’ each time I used the code. This is far more likely to be determined by guidelines and business rules than by some feature I can find in SNOMED CT itself.
Furthermore, consider the concept 136611000119100 |History of sepsis (situation)| (and/or its information model ‘past history’ analogue). A shiny new ‘sepsis 3’ sepsis code would conflict with the intended semantics of ‘past history’ for any effective time prior to 2016, so what does the new code actually ‘mean’ if not just ‘sepsis’?
I also remain concerned that putting even one additional step of retrieval distance between ‘infections’ and ‘sepsis’ is of questionable value. It is perhaps worth considering a couple of quotes from this 2022 Evolution of the Concept of Sepsis paper:
“…Sepsis is an infection associated with some degree of organ dysfunction—put very simplistically, sepsis is a ‘bad infection’…”
“…Sepsis is, and has always been, an infection associated with some degree of organ dysfunction…”
…bearing in mind that these words are written by Jean-Louis Vincent - one of the authors of the Sepsis-3 Consensus paper.
I personally feel it would be better for a number of reasons to handle ‘infections with sepsis’ as kinds of ‘infections’. SNOMED CT’s organising principles mean it is already rich with exceptions. Deep in the analytics guide we are told that “…assertions…in SNOMED CT [may be] logically sound but may be counterintuitive to clinicians…”. This phenomenon is commonplace.
Treating sepsis codes as kinds of infections may be ‘wrong’ through the strict lens of Sepsis-3, but we have to decide between SNOMED’s default behaviour for a << 40733004 |Infectious disease| query being:
I’m sure you can guess I favour the first option.
Ed
Thanks for your comments, Ed.
I agree that 91302008 |Sepsis (disorder)| could remain active. It was not included in the original briefing note as a concept proposed for inactivation and replacement.
To review the current status of 40733004 |Infectious disease (disorder)|. This concept does not currently subsume the following concepts:
91302008 |Sepsis (disorder)| (since 2020)
10001005 |Bacterial sepsis (disorder)| (since 2022)
770349000 |Sepsis caused by virus (disorder)| (since 2022)
700054008 |Sepsis due to fungus (disorder)| (since 2020)
To confirm the scope of the briefing note and the broader discussion required: the proposed naming and modelling changes correspond to the inactivation and replacement of 80 concepts and are intended to improve classification under 91302008 |Sepsis (disorder)|. These changes would not introduce wider hierarchical adjustments—for example, they would not involve placing 91302008 |Sepsis (disorder)| back under 40733004 |Infectious disease (disorder)|.
You noted in your earlier post that “I do not believe that this proposal addresses the majority of challenges.” For clarity—both for myself and for others contributing feedback—I wanted to restate that the current proposal is limited in scope and does not attempt to resolve the broader hierarchical questions you have raised.
Regarding the phrasing “infections with sepsis”: if this implies that infection and sepsis are occurring concurrently, this would suggest modelling such concepts as subtypes of infection. For the phrasing “sepsis due to infection”, this is currently captured by ‘Due to’ modelling.
Thank you for sharing the November 2022 paper (https://pmc.ncbi.nlm.nih.gov/articles/PMC9686931/). As you mentioned, the author played a central role in shaping the modern understanding of sepsis, including authorship on the Sepsis-3 consensus paper. I also reviewed two subsequent publications:
December 2022 (The Lancet eBioMedicine). The author states, “Infection is the underlying cause of sepsis and responsible for the initiation and perpetuation of immune dysregulation,” and further notes that infection cannot always be microbiologically confirmed. He emphasises that antibiotics should be administered in all cases, even when infection cannot be definitively proven, which occurs in more than 30% of sepsis patients. https://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(22)00500-X/fulltext
March 2023, Jean-Louis Vincent goes on to write an article titled “Sepsis and infection: Two words that should not be confused”. In this article he states “The pathophysiological mechanism underlying sepsis is best described as a dysregulated host response to infection.” https://pmc.ncbi.nlm.nih.gov/articles/PMC10033658/
It would be helpful to hear others’ views on the two options you outlined in your last comment, particularly as they frame the question of whether sepsis concepts should be changed to become subtypes of 40733004 |Infectious disease (disorder)|.
Best wishes,
Sarah
The Norwegian NRC has reviewed the current briefing note and agree that the logical definition of sepsis ought to be updated to adhere to the current clinical definition of the disorder.
As per Sarah’s last comment, the re-modeling of these concepts won’t significantly change their placement under 40733004 |Infectious disease (disorder)|, and even considering a potential future analysis and further action in the direction discussed in the comment section, we believe the suggested changes will improve the consistency of the terminology in the intermittent period of time.
The significant change in modeling that has been suggested ought to be transparent in the way changes and/or inactivation of content is handled. In light of the changes between Sepsis2 and Sepsis3, and the excellent point Ed has made regarding code usage vs. the graceful evolution of a terminology, we would vote for option 1 - inactivating and replacing the current content.
We recognize that we have a relatively small, young and simple implementation compared to other member countries. This proposal therefore doesn’t affect any post-coordination or information models for sepsis in particular for our end users. We respect the implementation and maintenance burden other countries might experience as a result of the suggested changes, and will accept the outcome either way.
Best regards,
Hanne
Thanks.
I’m not going to make the same points again. If SI wish to disregard my concerns then so be it - I have tried. The changes proposed will impact beyond the limited scope suggested, and users (those who are aware they have happened) will have to know how to adjust accordingly (at each point of the recording process). Those who do not adjust will experience unexpected results.
However I will comment on the apparent distinction between the phrasings “infections with sepsis” and “sepsis due to infection”: I purposefully used that phrase in my earlier note to flush out the problem. Is it really the belief of SI that the intent of the “Sepsis due to infection…”, “Organism X Sepsis” and “Sepsis due to Organism X” content is solely to represent those cases of sepsis where the named infective cause has resolved and only the sepsis/dysregulated response element persists? Are SI saying that users who wish to capture the co-occurrent infectious element should record this separately? Such behavioural details are unlikely to be available to those creating records (even if you document them in a briefing note or release notes). What would be SI’s response if someone requested a matching set of “Infection X with sepsis” concepts? They cannot be rejected on the grounds of simple co-occurrence, but wouldn’t their inclusion also cause confusion for both recording and analysis?
To the references you present - the therapeutic imperative of the first one supports the value of considering ‘sepsis’ an infectious phenomenon, and the second goes on to includes these two quotes:
“…the word “sepsis” always indicates presence of an infection…”
“…Sepsis is the most severe form of infection…” which seem to concur with the statements from the ‘Evolution’ paper.
Thank you, Ed, for taking the time to provide further useful feedback on this topic, which will all be taken into consideration when planning the next steps for this work.
You raise a very good point regarding the distinction between “infection with sepsis” and “sepsis due to infection”. My previous response referred more generally to how SNOMED CT models “with” and “due to” concepts. Based on the current editorial guidance, the use of “with” is typically intended for situations where two or more conditions are strongly associated by means other than causality or a temporal relationship. In this case, “infection with sepsis” would not clearly represent the causal relationship between the two disorders, where the infection is understood to be the precipitating factor for the sepsis.
To clarify, the intent of concepts such as “Sepsis due to infection caused by organism X” is not to imply that the infection has resolved and only the dysregulated host response remains. Rather, these concepts are intended to represent sepsis where the infectious cause has been identified. The modelling reflects the current understanding that sepsis arises from a dysregulated host response to infection, while still acknowledging infection as the initiating factor. In this sense, the modelling does not suggest that the infectious process is absent, but instead reflects the way the current sepsis definition frames sepsis as a host response to infection rather than as an infection itself.
More broadly, the points you have raised regarding concept stability, post-coordination patterns, and query behaviour are recognised as important considerations. Feedback such as yours helps highlight and share these considerations for continued and future discussions.
Thank you again for engaging so thoughtfully in this discussion and for sharing your expertise and observations.
Kind regards, Sarah
Thank you to everyone who has commented on this topic. The feedback period has now closed. The comments will be used to develop a revised briefing note taking into consideration the comments made here.
