overflow between Dr. Fowles and I talk, so hopefully this will kind of be additive to you. I have one disclosure that's not relevant to this talk. And so just starting out with immunophenotyping and really what is it and what are we talking about. There's a couple different like lingo in the literature out there right now, so the phenotype versus endotypes. I just wanted to get that out of the way. So if you think about the disease itself, so sepsis, there's different clinical phenotypes. We see different types of patients. There's different demographics, different triggers, different ideologies for their sepsis to kind of create all different phenotypes within the disease. Within those phenotypes even further, there's an endotype, so the molecular mechanisms, the differential heterogeneity within the mechanisms that happen within even each phenotype. And so you'll see in the literature both immunophenotyping and endotyping for sepsis. And so why is it important? So we know that heterogeneity is real. We have all sorts of patient factors, environment, microbial factors, and so treatment for one patient may not be appropriate. In fact, maybe the two patients may need complete opposite treatments, right? And so really understanding and understanding the immunophenotypes for each of these patients is really important. This is just an example of one of our prospective observational graphs that I put up here. Just looking at HLADR, we talked about, it's been shown in the literature, low HLADR is associated with decreased functional outcomes. If you look at HLADR inflammatory monocytes, just in that first day one to two, and looking at the difference between those who had rapid recovery and those who go on to develop chronic critical illness, you say, oh this is really nice, you've got some clustering, early decreased HLADR is associated with chronic critical illness. But you also have a ton of patients that come in with decreased HLADR that end up doing fine, right? So how do we understand this? How do we put everything together? This is really important. So I think that the trouble with the trials and the issues that we have right now in the immunophenotyping world is multifactorial. So one of them is that we tend to not consider all the physiologic pathways, we tend to just focus on the immune response. We're ignoring the endotheliopathy or ignoring the secondary organ failure that's happening concomitantly. So, you know, all these studies kind of try to narrow down what their topic of choice is. Do we look at all the pathways, all of the mechanisms, every cell type has mechanisms that are disrupted in sepsis. Do you really have to look at all the data at once or can you really only focus on the one? These studies tend to be microbe ignorant, the ones that are specific to immunophenotype, and we know that the immune system has different responses to different gram, positive, negative, and also different toxins. We really don't know the best study endpoints and particularly the timing of therapy. So we know that cytokine storm happens acutely, right? Do we intervene then? Do we intervene on day three or day five when that cytokine storm is over? Do we re-aphenotype them on that later time point? We don't know. And we certainly have changes in immune function over time. You know, there's new inflammatory insults that happen throughout hospitalization that nobody's really looking at in their studies. So the patients that come in their septic event, whatever their cause may be, resuscitate and they get a little bit better. Some people get worse. Some people kind of just smolder along. Then you get the pneumonia on day seven. They get re-intubated and they're still intubated. Then they get on CRT. So all these other inflammatory insults nobody's really paying attention to when we're looking at these outcomes. And so and certainly we only, most of these studies to be fair, only use blood and urine and really really don't know the immune function at the tissue level. Does the level and what the immune cells are doing in the blood and in the urine and the BAL maybe are equivalent to what's happening at the actual tissue level at the site of infection and then at the secondary organ injury sites. And then most of these studies ignore the nutrition, the energy utilization, and the profound catabolism that's happening concomitantly that really can profoundly affect immune cell function, mitochondrial function, and function, etc. And so what do we know so far? So we've had a bunch of studies looking at immunophenotyping, trying to sort these out as Dr. Files was alluding to. I think clinically we agree that there's different phenotypes of patients. Are the biomarker studies agreeing with this though? And so I kind of, I'm flipping the script a little bit. So I was asked to use innovative ways to immunophenotype in sepsis. And I think my argument for here today, and hopefully I'll convince you, is I think we've actually entered into a phase in terms of the techniques. We have novel techniques now that we've entered a new phase. The single cell, the multiomics, the big data, the AI, machine learning. I think we've entered into that in terms of technical skills of learning how to immunophenotype. Here's a list here. But my argument is I think is I'll kind of go at the end of this is I think we're looking at it we need to really pay attention to certain questions if we do this in order to do this the right way. And so there's a bunch of studies I'm going to go through in terms of the big studies that are out there right now on immunophenotyping. Dr. Files had alluded to a couple of them. The MARS study was one of the first bigger trials in 2017 looking at 522. Again, a gene expression within the first 24 hours of ICU admission. You get MARS 1, 2, 3, and 4 categories. Great. This is wonderful. We have four different types. Maybe we can see that clinically as well. But when you look at the description, and certainly there's a list of the different genes that are up and down, you know, key, innate, and adaptive is one. Pattern recognition cytokines is two. Three is quote-unquote a low-risk cohort. And four is similar to tube and interferon. I don't know what that means. Are those different? Are those not different? There's so much overlap. And then on a follow-up study in 2022, a validation study, actually looking at the validation of this particular four different endotypes, really looked at both univariate and multivariate. Didn't really find that that endotype class is actually changing your outcomes. And so actually looking at those. Seymour's group, here's the alpha, beta, gamma, delta phenotypes. And they did a ton of impressive, impressive machine learning analyses on these. And this is what they came up with. Alpha has maybe a little bit here, a little bit there, some a little bit of other. The beta has a lot of other, maybe some renal. The gamma has a whole bunch of heme, hepatic, inflammatory. And then the gamma has, are these different? I'm not sure. The biomarkers are telling us this. The machine learning is beautiful, beautifully done. But are these making sense clinically? There's two other additional studies, 2018 and 2022, have done some other classification of the genetic datasets. Inflammopathic, adaptive, coagulopathic. That sounds interesting. That may be interesting clinically. But there's a ton of overlap in these patients as well. Here's another study in 2022, the most recent one, kind of looking at the differences and have different kind of categories. Are these really different? Are these clinically relevant? So one thing I want to take a step aside. So that's what we know so far. Those are the studies out there about immunophenotyping. One of the things I really think is really important that I want to take a side step for, at least, is looking at the heterogeneity and the variety of patients that come in, and particularly the ICU versus non-ICU patients, right? We've done all these things. We can take care of critical patients based on all the things that have been researched in this forum. But what's happening to the survivors? We know that those patients that stay in the ICU for greater than 14 days, which is a fairly significant proportion, about 20 percent of all ICUs. And so those patients, we know, half of them are dead at six months. They've got significant impairments up to the year after discharge. So we're taking care of these patients really well in the acute phase, but they're kind of going on and smoldering and smoldering and smoldering the subpopulation. And so the, you know, sepsis mortality, we've reduced it in half, but we've tripled the number of patients going to rehab. And so one of the things I think is really, really important that we tend to kind of ignore is that those, when we're looking at these sepsis outcomes, we say their initial diagnosis. We focus on an acute phase, but again, 20 percent developed that chronic critical illness. And so what does that look like? They've had their multiple organ dysfunction. They've had their renal failure, their liver shock, their shock liver, their acute lung injury. Those are still tissues that need to heal after that initial septic event that we're not paying attention to in these immunophenotyping studies. And so you get persistent activation in these particular patients, and they remain immunosuppressed for months and months. And this requires extreme amounts of energy, and these, the catabolic aspects of these is pretty impressive. And so this persistent inflammation, immunosuppression, catabolism syndrome isn't even talked about in some of these immunophenotyping studies, right? So again, most of these studies are focusing on the first 24 hours of presentation in that genomic storm. I think you really need to consider everything after that initial event to truly understand this disease process moving forward. There's a lot of difference between patients who have rapid recovery who get better and those who remain in the ICU for days. All these biomarker studies are showing there's a lot of overlap though. There's patients that have low HLADR that do fine, and those that have low HLADR and go on to have chronic critical illness. How do we differentiate between those? I think you can't just evaluate that first 24 hours, 48 hours, 72 hours even, ignore the rest, and then look at 30-day outcomes. And so again, what I'd argue is I think from a technical standpoint, from a biomarker standpoint, from the nerdy stuff that I do in my lab, I think we're in a new phase of the innovation and the techniques that we can use to immunophenotype patients. The omics, the single cell analyses, all this machine learning. I think the innovation is the way we approach the heterogeneous disease process and not the techniques we used. So the innovative factors, I hope the people in this room can answer a lot of these questions for me. I've asked a bunch of people and have come across silent answers, but this is, I'm going to kind of be a little bit provocative here and what's the innovation that I think is really important that we need to do moving forward. So here's just a series of questions. I don't know the answers to, but hopefully people in the group will at least start thinking about it, and I kind of implore anybody that's doing prospective observational trials, particularly looking at the immune responses, particularly endotyping these patients, consider all these factors. These are not exhaustive. I think these are the main ones that come to my mind every time I'm doing this. So one of the things I've asked every single smarter person that is much older than doing this longer than me, what are the appropriate control groups for these patients? Is it young, healthy patients? Is it demographically matched, non-hospitalized patients, right? Septic patients averages around 50. They have comorbidities. Is that your appropriate control? Do you look at patients who are admitted for infection that don't have sepsis? What's the appropriate control here? Nobody's really defined it. Probably all of them, all of the above would be my guess. You have to compare to all these groups. What are the best endpoints and outcomes to consider? So, you know, we've talked about this ad nauseum in this forum. Do we really look at 30-day mortality? Is that even reasonable to look at in this patient population? Do you look at length of stay, length of stay in the ICU? Do you look at time to resolution of their sepsis or inflammation? What are the outcomes that were going to be the most important, the primary outcomes that are going to help us determine what are these immune phenotypes and how do we treat them? Do we evaluate the extreme phenotypes? Right now, we tend to look at the average and the median population and what does the average median person do? How do they respond? But what about the patients that survive and we think they're going to die? That's a pretty unique population. The reverse is true too. Death in patients that we expect to live. What happens in those patients? Why don't we understand them? Do we need to separate them from the average and median? Do we start with the clinical phenotypes and work backwards? Right now, the phenotypic data that's out there in the biomarker studies are starting with the biomarker, starting with the machine learning algorithms, but they're not matching up quite well to the clinical phenotypes. We know the clinical phenotypes. We see them. Two patients come in on rocket fuel. They have completely divergent care despite what we do. What is good and what's bad? You know, so we do all these cytokine trials. They have high levels. They have low levels. We don't even know what quote-unquote high and low is, right? So is high TNF good? Some of us say yes. Some of us say no. IL-7, interferon gamma, these may be some of the reasons why we're not getting so much signal through all these trials that we've been doing. And then what is the threshold between infection and sepsis? I've been knocking my head and asked, again, many, many older people about this for many years now. We know that you get systemic cytokine release in a little lady that comes in with a urinary tract infection or pneumonia. What's that threshold that turns them into sepsis? Nobody really knows, right? And so when we're looking at that immunophenotyping, we really need to understand that as well. What do we do with all this single cell data? Does it matter that we have individual cellular responses when every single cellular system in the body is going haywire and septic and septic shock? You know, are gene profiles alone good enough, or is multi-omics too much? Again, so one of the things that we're facing is looking at this prospectively. There's so much noise in these patients. There's so much noise. And I don't know, this is for people smarter than me and know the machine learning, but do we look at all of the signals? Do we need to look at all the signals? Or is it too much? Are we never going to get, or sorry, do we look at all of the noise? Every single pathway involved, is that what's going to give us the information? Or is it too much? Is it creating too much noise and we'll never find a signal? I'm not sure. Well, knowing these endotypes and phenotypes help with timing of intervention. Again, you know, do you endotype them on day one? Do you endotype them on day three? Do you endotype them on day seven? What's the best timing to do these particular trials? Does it matter what happens in that first 24 hours when they're, everybody's storming? Or does it matter what they're doing after that initial resuscitation? I don't know. The other big thing that comes up is, do we, do we stimulate cells ex vivo to determine their function? So in kind of a related question, does immune dysfunction studied ex vivo actually reflect the inability for your immune cells to function in vivo? Right, so there's a bunch of studies looking at, looking at neutrophil function, say, taking them ex vivo. Do you stimulate them with LPS or do you not stimulate with them with LPS? Do you do both? Which one's the right answer? I don't know. We don't have the answers to these questions yet. Are there completely different immune phenotypes and chronic critical illness or is it just the acute phenotypes that predicted? I don't know. And then what happens if you change endotypes during your hospitalization? Like I said, does that change the whole algorithm again? Do you restart over? I don't know. And then most specifically, are these phenotypes specific to sepsis or do they apply to other critical illnesses? Is this a phenomenon that just happens in infectious field or in the damp associated the ARDS non-infectious kind of sterile inflammation? I don't know. And so I think for the next steps and kind of final thoughts, again, focusing on that acute phase only, obviously it's really important, but it's a little bit short-sighted. We need to consider all those multitude events after the initial septic event improves. I think we still agree that we'll need to target multiple pathways or multiple upstream pathways and we need to leverage that multi-omics and machine learning. It's just a matter of how much and how little. And then determining individualized endotypes are likely our next best option towards guiding the optimal therapies, but we've got a lot of work to do and a lot of things to consider. And so hopefully some people will answer some of these questions for me maybe at the end in the question session. But with that, I think we're going to be ending.

immunophenotyping

endotyping

sepsis

machine learning

individualized treatment