Okay, good afternoon, everyone, good to see you guys. And so I have the pleasure of going last, which means I know what's been said, we get to not duplicate ideas. In fact, I'm gonna try today in talking about sepsis subtypes challenge some of what's been shared, conceptually, so I'm gonna show very little of our own data, which I think some of you've read before, and we're gonna talk just more philosophically about this whole issue. So these are some disclosures, some roles that I have, I don't think they relate to this current chat. So because I was tasked with sepsis, I get to say one slide about why sepsis is important. That will be reviewed for everyone, I hope. Discuss a little bit about heterogeneity, some underpinnings of mechanism, and how, as we've discussed so far, how this can end up having meaning for patients. So there are caveats, of course, that many of you in the room, in fact, our colleagues here on stage have thought about this for 10 or 15 years. There are molecular approaches, lots that's been shared and lots more that we don't have time to go into, and it's also an interesting topic for our funders. The NIGMS Working Group on Research Priorities described heterogeneity as item number one on their very important document from a few years ago, and of course, if you added up millions of dollars from NHLBI or NIGMS to heterogeneity, that'd be a big pot of money. And so there's a pressure on us to deliver, right, and to deliver for the patients. I'm gonna skip over Christine's beautiful paper in Lancet because we know that sepsis is important. There's 50 million cases a year, and that's why I think the heterogeneity of sepsis is a priority for all of us to consider. The outcomes here have been unremittingly poor, and I would say that one out of five who don't survive is poor, and we should be able to do better, and then as discussed, these precision approaches seem like a plausible path. When we think about sepsis just as in ARDS, we can show pictures of the presumed biology under the surface. Imagine the picture of the alveolus in the New England for ARDS from Michael and Lorraine. Here's one from Harrison's about sort of the inflammatory cascade after an inciting pathogen here in sepsis, but this has been also documented by others many, many decades ago. So this is genome-wide transcriptomic profiling from Calvano looking at endotoxin infusion, and basically the point here is that there is a lot of things going on amongst patients who are ill than those are healthy. So I think conceptually we're all there that heterogeneity is a problem. We've talked about how it may underlie neutral trials, but the question is what are we doing with all of these subtypes? And I don't mean to be cynical, but I'm gonna challenge this notion. So I tend to read a little bit too much about taxonomy, and so the purpose of taxonomy is shown here, to describe biodiversity, to provide names when we communicate about one whale or one dolphin, and to sort of document the richness of our planet, and it's described in the phrase on the right, and I'll read it. The scientific discipline that explores, discovers, interprets, represents names, and organizes organic beings. This is actually kind of similar to what we're talking about for our patients and our diseases here. So in the context of sepsis, just find, replace sepsis into this, and we're seeing an effort to describe the diversity in sepsis, provide names so that we can communicate, and again, look at the phrase on the right. That should probably resonate with what we're doing. We're essentially being taxonomist. So I was like, well, maybe this has like a name, septonomy, septotax, taxosep, none of them really were, like, they're all kind of silly, but anyway, you get it, right? We're trying to create a taxonomy of these acute illnesses that have patients with inflammation and patients that don't. Okay, so the issue of what is meaningful also came up earlier, and to state it explicitly, I think clinicians may view something as meaningful as a subtype. If it has a treatment response, perhaps even the patient would as well. They want to survive and may not care about all of our machinations about who's reactive or uninflammatory, but rather, do I need drug A versus drug B? Our scientists, colleagues that do fundamental work may be interested in understanding sort of the biology of one group and another in order to lead to new therapies, and sort of folks that just like labeling things may want to be seeking the truth. Is the circle I'm putting around this group of patients real, or is it just an arbitrary derivation of an equation? So naive to those goals, though, we have charged ahead. Kim DeMurl, one of our fellows, did this appendix published review in a viewpoint last year that showed now there are more than 100 classes of sepsis proposed in the literature, and they're all different, and I'm not going to go through those, but they're different because of the data sources, because of the purpose of the investigators, and as we touched on, their generalizability. So go on this journey with me. Here is how we have began. Thinking about sepsis 30 years ago, like this Rothko, there are patients with or without the disease. Now you'll notice this isn't discrete, right? These are blended together. We don't really know who's septic or not, but this is where we were, and I argue that at this moment, we're more sort of in the Mondrian phase, where we've used clinical data, some biological data we've talked about to create these discrete buckets, as shown here in this painting. This makes life easy when we're communicating at conferences, we're applying labels to new data sets, but I'm not actually sure we even know what to do with the red versus the blue patient right now, as Professor Tinnah had talked before. Now, there's others that are moving into this world. So Kandinsky, a Russian painter, sort of one of the fathers of abstract art, and when I see our machine learning outputs, I think about this, right? We have groups that are overlapping, circles that are nudged next to one another, and it's kind of chaotic, in my opinion. They're often derived, as we said, retrospectively after a trial has been run. They're not yet available at the bedside, and I have no idea how many of these circles is enough. As a clinician, it'd be much easier to go back to the Rothko, but as researchers, we keep doing this stuff. Okay, and here's an example. My own paper did this kind of thing three years ago, where we went through many of the steps described, and as has done Tom Vanderbilt in his lab, and many, many others, we keep making more circles. And so then you ask yourself, well, do we really need the circles at all? So this is a Srot, right? Pointillism, the individual, I tend to really enjoy that painting, and when you zoom in really close, they're all individual little points, right? Like our patients, and so we face a lot of challenges, I think, at the moment left off from our prior speakers in trying to find meaning, and some of the challenges are shown here. One is that there is just a word we use, phenotype anarchy. Sort of, we stole this from a term taxonomic anarchy that was published in Nature a few years ago. There are a lot of similarities, again, between the challenges of our taxonomy colleagues and the things that we're trying to do up here. We also love to be discreet, and I'll talk about that in a moment, and then consensus is another challenge. So here's Instagram in reality, right? Here's the Mondrian, or Mondrian, I don't know how to say the name properly, but we think everything is separated, one versus the other. I label this person hyper-inflammatory, I label this person hypo-inflammatory, I'm very certain that they're in those groups. But the reality, of course, is that it's quite blurry, and that we don't truly know anything about the margins or those patients. It gets back to some work we were fooling around with at the time of sepsis three, where there is this so-called zone of rarity in species between birds or mammals, and there sits the platypus right in the middle. It's some of both, and the reality is that our patients are often some of both. They are not black or white. When we tried to search for discrete things in the Seneca work years ago, what you do see, however, is this plot on the right. This is the probability of membership in the four Seneca classes, and of course it's not a straight line. There's a color distribution moving left to right of those with a low probability versus a high probability, even if you were assigned to the group. Okay, so that brings up the notion of the edges. If these are all living in some circle in the Kandinsky painting, I'm kind of interested clinically on the patient on the margin, and when we use sort of quantitative methods to find those who are kinda not really a member of one group or not really a member of the other, we find by following them over time, those are the patients more likely to transition to a different phenotype, and then even when we look at treatment response, to have a different response to therapy than those in the quote center of the phenotype, subtype, subphenotype, whichever word. Okay, so the point here is that the edge is not the same as the middle, and is the circle really enough? Okay, then we can look at this. So these are more Kandinsky paintings. These are small circles. That's the name of this art, and it sort of resonated with me that amongst these very well-recognized labeling groups, they all kind of look different, and I'm not sure what to do with that. We're not alone. Of course, taxonomists struggle with this issue of overlapping classification systems that don't always agree, or maybe they agree in unknown ways, and there are mathematical solutions to this problem, but we also might reflect on what our oncology colleagues do when they evaluate women with their BRCA, their mamma print, their hormone receptor levels. They're using multiple strategies to assign patients to a treatment regimen, not just one circle or another. There's data to show this. Lonica Van Vought in Tom Vanderpoel's lab has been working diligently to map a variety of classification strategies across each other, and this alluvial plot, which shows proportions of Seneca subtypes on the left mapped to proportions of Mars subtypes on the right, you see that there's quite a bit of disagreement, patients in one that get in an unexpected group in another. All right, so you might ask yourself, well, do we need groups at all? And there are statisticians who are pushing the envelope in what are so-called individual treatment effects. Instead of differential treatment effects by group, they're using mathematical models to create distributions of how a patient would be if we knew what their counterfactual patient looked like, and then we know from a trial what their treatment response was. And so, although I'm not an expert in these methods, they're coming up with strategies to say if you're on one end of these distributions of the so-called ITE, great, get the treatment or not. But if you're in the middle, perhaps that's the group we need to randomize. All right, so I'm almost done. Sort of last slide is that sometimes I wonder if chasing these classes, taking this set of classes and putting it in another data set is really accomplishing things for our patients. And this is my own self-reflection. Whether we're building a sort of a taxonomy of sepsis or ARDS that has relevance and has consensus, I think that's an effort that many folks in the room are working towards. The sort of desire to be discreet is I think limiting our progress, as well as, as mentioned before, the clear lack of trials that implement these subtypes at the moment of randomization. So, with that, I hope I wasn't too pessimistic about this area. I'm still hopeful and we look forward to your questions. Thanks.

taxonomy of sepsis

defining subtypes

clinical and biological data

individual treatment effects

heterogeneity of sepsis