Good morning. It's a pleasure to be here today to talk with you a little bit about some of the controversies that we encountered during the process of developing these new sepsis criteria. I have no relevant disclosures. I was part of the grant that was mentioned earlier by Talon and Nelson. So, we've already discussed a little bit by way of introduction around the task force and the development of these criteria, but I wanted to sort of re-emphasize a point, which is that this task force included 35 individuals from a very vast area of the world. So, we had people from 12 different countries, 6 different continents, and ultimately a vast, diverse set of experiences and expertise. So, it's no surprise then that controversy would be a big component of the development of these criteria. We've already discussed a little bit about some of this with regards to the international survey and the systematic review, which have helped to contextualize where some of these controversies might point us. And then even within our own group, it was known that we had to make some key decisions around certain elements of these controversies, and we developed a criteria, which is the Delphi process, to determine how these criteria could be ultimately made. I want to point out that the controversies were indeed fostered by the leadership, and I think this made the criteria stronger in the end. And our paper has a supplement, which includes a lot of the details around the controversies, and I'd encourage you to go look there. I'm just going to talk about a few points and give you a little bit of a taste of some of the issues that we dealt with. So, I think we can divide the controversies into sort of stage 1 and stage 2 controversies. So, the stage 1 controversies were predefined by the leadership and were part of the discussions we had in Salzburg back in 2019. So, of these, some key ones included, for example, the conceptual definition of sepsis, which was already touched on earlier. Also, among the topics of discussion was how to operationalize the criteria of infection, which has also been a little bit mentioned. How to define shock. Are we looking for a sensitive criteria or a specific criteria? What about sepsis phenotypes? Can we include different phenotypes, or are we going to have a single criteria? And then the age range. Are we going to capture neonates in our criteria or not? And then, of course, issues around the stages of sepsis. Early, screening, severe, etc. And then what is the ideal outcome to be used in the criteria development? Are we looking for mortality or mods, or what is the ideal outcome? And then there was the issue of high versus low resource settings. Are we looking to have separate criteria or the same criteria for both of these areas? In the next stage, which is really the stage over the past year during the data phase, we had to discuss issues around which organ criteria to use, which was outlined earlier by Tal and Nelson. How do we consider interventions? Are they also eligible to be used as criteria? Of course, one of the big issues was the issue around model parsimony. Are we thinking more about the criteria as a prediction tool or explainability? And there's a natural tension there. What about the types of sepsis and severity? This was readdressed during these last few months of model development. And then, again, the issues of high versus low resource was an important area that required quite a bit of discussion. Issues around remote organ dysfunction was already mentioned, and then the screening criteria as well. So the first criteria that I'll touch on is the issue of high and low income settings. And ultimately, our desire was to have a globally relevant criteria, and this was something that we had to think about carefully in the development process. So if we look at under 5 death as a proxy for sepsis, because unfortunately there just isn't a good data source to capture sepsis prevalence and mortality in the global south, we can see here that the primary area where sepsis occurs is in the global south. And if we contrast this with the resources and the distribution of resources using physician distribution as a proxy, again, the issue is that there is no good data on metrics for resources. We can see that the distribution of resources and the prevalence of sepsis mortality is not ideally suited to where the disease occurs. And this really informed a lot of our discussions about how do we make criteria that is applicable to all settings. So in terms of some of the specifics, so it was important to discuss the issue of the availability of variables. So are the variables included in the criteria going to be available in all settings or in the majority of settings? And then of course the issue around validation. Are we going to build criteria that are validated in the settings where they could be used? And of course, given the disparity in both mortality and resources, how do we operationalize these criteria in a multitude of settings? So in order to mitigate at least some of these, we sought from the get-go to bring in data from low-income settings, which was not an easy feat. And we did get some data, but one of the issues that we see is the issue of data poverty in low-income settings, and it was very hard to find high-quality data. As we've already heard, there was lots of missingness in these regions. It was a goal to develop criteria that were simple and could be easily implemented so they could be used in these settings, and we did have some qualitative work where we did focus group discussions and key informant interviews with end-users and other experts to better understand the pathway for implementation in these settings, and this ultimately helped guide some of the decision-making around these criteria. So the other issue that I want to bring up is the issue of parsimony, and this was also mentioned earlier a little bit by Tal and Nelson. There's this natural tension that we see between simplicity and explainability, and we saw this with the four versus eight organ criteria, and we also see this with the issue between low- and high-income country criteria. Ultimately, we recognize that we used a predictive approach, and this is not always going to provide the most explainability in a criteria, but I think in the end it produced a criteria that's cross-cutting across different geographies from which the data was captured, and I think this allows for a more feasible adoption across various global settings, and hopefully more collaborative research across these settings as well. So I'll pass it off to Mark now. Actually, one more slide. I wanted to give a taste of the Delphi process. So for a lot of the controversies, we had voting, and that was mentioned earlier, and this is an example of the voting that was done for the four versus eight organ criteria. So you can see here that on the upper row, which is the eight organ system, it was generally acceptable by the task force to have that, but it was more preferable to have the simpler one, and I think this then led to the adoption of a unified criteria for high and low settings because it was already as simple as it was going to get, and then we had very high agreement for the high versus low resource criteria being the same. So that's the taste of my piece, and I'll pass it off to Mark. Thanks very much. So some of us who were invited to join the task force took the view that our role, without machine learning or huge database expertise, that our role was to ask the difficult questions, and I consider myself one of those people because when we first started, I think I said at one of the calls, I'm not sure sepsis is a thing. I should have probably been kicked off the call. I think it's a thousand different things, and so I've got two examples of a controversy, one that was prior to setup and one sort of in face of the results, and so they're both sort of heterogeneity issues, and this is the first one, early versus late as a cause of heterogeneity, and this is just an excuse for me to show one of our old papers from nearly 10 years ago, but this has a key of this slide. It's a survival curve for children in blue and with no comorbidity and a diagnosis of sepsis referred to our transport team, and in red for children with a previous comorbidity, bad enough for them to be in hospital in the last year, and the point is that in the first 48 hours or so, they die at the same rate, but then something different happens and their cumulative mortality separates dramatically, and so to try and predict across these two situations is really problematic. There's a real signal to noise issue if you're looking at late mortality. In other words, there are multiple ways you can die in the context of infection, and this is one of the possible limitations of sepsis 3. There's a circular argument that risk of death is the definition, so if you died, you had sepsis, and you can go round and round that little loop, and I would argue that that first group had sepsis as a clearly causal link, but for their sepsis, in legal speak, they would not have died. The second group, those in red who are dying after 48 hours, maybe the infection is a marker of their general debility, and they're dying of their acute myeloid leukemia, which renders them vulnerable to the infection, so could we really expect something to predict across those two sets of criteria? We can characterize that as early versus late. The team put me in my place about this because actually when they did a sensitivity analysis of the data, cutting it by early and late, the performance is almost identical, so I can be quiet about that. The next one still stands, though, and this is the issue of intervention as criteria. Some of us are less is more intensivists, some of us are more is more intensivists, perhaps depending on your billing system. I'm very jet lagged. The British attitude is different. In these criteria, there are at least three points which relate to the level of intervention, and so that's kind of complex because imagine I'm on call. This is supposed to represent me at short notice, and we're offering relatively little support to this patient, but if the same child had been admitted under my fellow, it's quite possible they would be less tolerant of the respiratory dysfunction and maybe of that hemoglobin and maybe that adenovasopressin also. The physiology would be the same, but the physician has altered the score by virtue of their threshold for intervention, and this then links into this point about remote sepsis because sedating for ventilating would be an organ failure outside of the original site of infection, and it's super appealing, this idea of remote sepsis, because it sort of captures something that I think we all innately believe, that something systemic is core to our view of what sepsis is, but it's very difficult to pick out if we all have different thresholds for intervening, and we're using intervening as the characteristic for defining. So that remains a problem, but despite that, we've got a score that outperforms anything that's existed up until this point, and we should recognize that. So our conclusion then is that I think these criteria are tempered by the white heat of the discussion we had at some of these meetings, and they were pretty meaty. They persist. I think it's fair to say that controversy was encouraged in an attempt to temper this. I think the final criteria capture some of the true variability, whether that's pathophysiological or geographic or therapeutic, and we can finish with Francis Bacon, if a man is content to start with certainties, he can, he ends with doubt, that way around, and if he starts with doubt, he will finish with certainties. So I think that should reflect what's happened here. Thank you.

sepsis criteria

international task force

Delphi process

global healthcare

low-income settings