Thanks so much, Kate. Thanks to the organizers for inviting me to talk. My given title today is actually Physiologic Subphenotypes in ARDS, Can They Make a Difference? So it's gonna be a little bit of a slightly different focus to start off with. These are my disclosures, which are really not relevant to what I'm gonna cover today. So just like there are many different ways to slice an onion, there are lots of different ways to phenotype ARDS. And I think we've spent probably the last five plus decades in ARDS talking about clinically obvious phenotypes. Those may be ARDS risk factors, direct versus indirect lung injury, mild, moderate, or severe ARDS, diffuse versus focal, and of course, in the last three years, COVID-19 versus typical ARDS. Now, this is not an all-inclusive list, but what we're talking about with phenotypes, I think is just a way to split this heterogeneous syndrome into groups that may be slightly more homogenous. I really think, why should we care about phenotypes? Why are we here today? I think we're here because we're looking for groups that respond differently to treatment. And so I'm gonna talk about physiologic phenotypes today since that's what I was asked to cover, but really with this lens of thinking about can we identify groups that respond differently to therapies? What do we mean by physiologic subphenotypes? Well, I'm gonna focus primarily on respiratory physiology because I think there's some really interesting data there. We'll talk about severity, PF ratio, lung compliance and driving pressure, and then data-driven algorithms combining multiple parameters. I'm gonna talk just a little bit at the end about non-pulmonary physiology, though Dr. Wilson, who follows me, is really gonna talk more about phenotypes that are focused outside of the pulmonary physiology space, and again, with this lens towards treatment-responsive phenotypes. So let's focus on these respiratory physiology phenotypes. And I'm gonna start with just severity, and I really just have one slide on this. I don't really know if this even falls into the rubric of physiology-defined phenotypes. We could quibble with whether this is a phenotype since it's really just using one parameter to split patients into different groups, but this is obviously the most clinically implemented and probably most evidence-based way that we have to split ARDS into subgroups. And I think we have really strong evidence base for using a PF of less than 150, for example, to focus our proning strategies, different PF cutoffs for using ECMO. So this is perhaps the simplest physiologic phenotype. What about lung compliance? We talked a lot about this during COVID, and we'll get to that, but I wanna start with some evidence that actually largely predates COVID. And this is a really interesting paper written by Ewan Gallagher, Marcelo Amato, and colleagues in which they did a meta-analysis of five different clinical trials that focused on high versus low tidal volume, and they asked if there was any evidence that lung compliance could modify the treatment effect of tidal volume. And what they found here is that as elastance, if you see on the left-hand slide, as elastance goes up, which means, right, compliance is going down, the impact of tidal volume on driving pressure became more pronounced. The blue here is the higher tidal volume strategy, and the red is the lower tidal volume strategy. And that makes sense, right? If you have a stiffer lung, then you're probably going to see bigger changes in driving pressure with higher tidal volumes. What they found when they looked at the risk of death and how it varied according to elastance was really quite interesting, and that was it was really at these levels of high elastance or low lung compliance where they saw the biggest difference between the low and the high tidal volume strategy, suggesting that it's really those patients who have low lung compliance that benefit the most from a lung protective strategy. Now, this is secondary analysis of an RCT. I would not recommend that you use this to treat your patients, but certainly thought-provoking that there may be differences in physiologic phenotypes that should at least prompt us to think about how we're ventilating our patients. Similarly, Dr. Gallagher and colleagues have done some really interesting work on ECOR and lung compliance. These are figures from a paper in the Blue Journal in 2017 in which they did some really interesting physiologic modeling to predict the change in driving pressure with changes in compliance, as you can see here on the left-hand side, and alveolar dead space fraction on the right-hand side. And what they found was basically that as compliance goes down, you would have a higher predicted change in your driving pressure with institution of ECOR, and this is presumably because we think that at least the benefit of ECOR may derive from decreasing lung injury, lowering tidal volume, and driving pressure. And the opposite was observed with dead space fraction. So as dead space went up, the predicted change in driving pressure with ECOR initiation became more noticeable. They went on to sort of validate this physiologic theory in a secondary analysis of the SuperNOVA trial, which was published in Intensive Care Medicine in 2019. This was 95 patients with early moderate to severe ARDS treated with ECOR, and what they found was these figures on the bottom, the y-axis is change in driving pressure, okay? And of course, what you're looking for here, ideally, we think, is a bigger change in driving pressure. And as the dead space fraction went up, as the dead space fraction went up, in fact, the driving pressure went down, as the ventilatory ratio went up, the driving pressure went down, and as compliance went up, then the estimated change in driving pressure, or the observed change in driving pressure, rather, went up. So this suggested, okay, maybe these patients are really more likely to benefit from ECOR that have higher dead space in VR and lower compliance. Well, what happened in an actual trial of ECOR? This was the REST trial, led by Danny McCauley and colleagues, out of Queens University Belfast, published in JAMA in 2021. This was a trial of lower tidal volume ventilation combined with ECOR versus conventional tidal volume ventilation for patients with early acute hypoxemic respiratory failure, PF of less than 150. The trial was stopped early for futility. You can see the survival curve here on the right-hand side. No difference in mortality between ECOR and standard of care. And interestingly, there was no difference in the effect based on ARDS, severity, plateau pressure, driving pressure, or PaCO2. And you can see here, this figure on the right at the bottom is just meant to show that there was actually separation of these two treatment arms on the basis of driving pressure. So why weren't these theoretical benefits of ECOR borne out more in the patients with different lung compliance? Well, unclear, maybe because it wasn't all ARDS, maybe because there were off-target effects or risks of ECOR, including higher risk of CNS bleeding. And in fact, they didn't show us data based on subgroups based on compliance or ventilatory ratio. But it's just to say that while there's this theoretical benefit with going to much lower tidal volumes for these patients with low compliance, so far we don't have strong data to suggest that that is the case. What about ECMO? Might this be differently effective for patients with different pulmonary physiology? Well, this is a really cool study done by Martin Erner and Eddie Fan and colleagues published in the BMJ last year in which they did what's called a trial emulation using observational study data from 7,000 plus patients across five continents. And what the idea is here is that you use observational data to try to emulate a pragmatic RCT for therapies like ECMO, which it would be very hard to randomize 7,000 patients to. And they did a sophisticated analytic approach that I'm not going to try to explain here, but to try to essentially adjust for confounders to try to approximate the treatment effect of treating patients with ECMO versus not. And what their primary analytic question was was if we applied ECMO for all patients who had a PF of less than 80 versus conventional mechanical ventilation without ECMO, what do we estimate the treatment effect of that would be? And what they found was actually for that primary question that there was an estimated risk difference of minus 7% in hospital mortality and a similar improvement in the rate of discharge alive, again comparing ECMO initiation for a PF of less than 80. They also simulated starting ECMO with different thresholds of driving pressure and found that essentially their simulation suggested that if we initiated ECMO when the driving pressure is greater than 15, that that would confer the greatest mortality benefit. Now again, this is not RCT data, so I don't think we should rush out and start implementing this right away, but it's the idea here that there's meaningful heterogeneity within the physiology of our patients with ARDS that may have an impact on how different treatments benefit our patients. All right, now I alluded to this earlier, what about compliance phenotypes in COVID? So this was like all the talk in spring 2020, started I think in part by this article by Luciano Gattinoni and colleagues in critical care as well as by an article in the Blue Journal followed by numerous, numerous publications on the respiratory physiology of COVID. Was it in fact different? Is this the same ARDS and are there distinct phenotypes, physiologic phenotypes within COVID? And these are just two of the papers that were published to try to answer that question, which I think that the answer was essentially, yes, there is meaningful heterogeneity in COVID in terms of pulmonary physiology, just like there's meaningful heterogeneity in regular ARDS, typical ARDS, right? So this on the left-hand side shows the variance or the variability rather in P to F ratio, plateau pressure, compliance, and PEEP within patients with COVID. And you can see that, yes, there's a lot of heterogeneity, but it's not like it's clearly following into two groups like had been proposed. Similar findings here from Giacomo Grasselli and colleagues showing this variability. And I think here that part of the problem was this idea that because we see heterogeneity in something, we see there's variance in compliance, that that means that the treatment that's offered to patients should be different. And I think that that is not necessarily correct, right? We have large RCTs in patients with heterogeneous pulmonary physiology suggesting, for example, that lung protective ventilation is strongly beneficial. And so I think when we don't have evidence that there is this heterogeneity of treatment effects, we really need to hew to that data. So is there really data to suggest that when we look at the respiratory data from patients with COVID-19 that there are distinct subgroups? Lou Abbas and colleagues set out to answer this question doing a data-driven analysis of respiratory variables only in patients with COVID. And you can see those variables here on the left-hand side of the screen. What they did was to analyze data at baseline from these patients using a technique called latent class analysis that seeks data-driven subgroups in heterogeneous populations and found actually no evidence that just looking at that baseline data there were distinct subsets of patients. However, when they analyzed both data from baseline and trajectories over time, they actually did find two subphenotypes of patients. And this figure here shows the respiratory variables on the x-axis, the difference between the two phenotypes on the y-axis, and then the different colors of these lines are different days in the trajectories of these phenotypes. And what they found was actually these phenotypes were not characterized by differences in respiratory compliance. You can see that's here, and really the phenotypes had similar compliance. But what they found was that subphenotype two was characterized by higher mechanical power, higher minute ventilation, and higher ventilatory ratio over the first four days of mechanical ventilation. There was no difference in mortality. That phenotype did have fewer ventilator-free days and more thromboembolic events, and no evidence, albeit in observational data, of differential treatment response. So yes, there is meaningful heterogeneity within COVID-19, but that doesn't necessarily mean that we should be treating our patients differently on the basis of that physiologic heterogeneity. What about data-driven physiologic subtypes in typical ARDS? Well, an interesting paper published by David Cimiello and colleagues in 238 patients, where they did a latent class analysis of only respiratory, well, primarily respiratory variables in patients with non-COVID ARDS. And they identified two phenotypes which they termed recruitable and non-recruitable on the basis of CT findings after a standardized recruitment maneuver. This figure on the right-hand side shows the variables on the x-axis, the standardized mean value on the y-axis, and you can see that these phenotypes, the recruitable phenotype has differences in the proportion of lung tissue that was inflated after this recruitment maneuver. These phenotypes had prognostic value for ICU mortality. No differences in response to treatment were tested in this study, other than, of course, the recruitment maneuver, which I mentioned, which actually defined the subphenotypes. And I think this is a really interesting finding, but needs replication in additional cohorts before we can do more with this. All right, what about phenotypes based on non-pulmonary physiology? I'm just gonna touch on this briefly because Dr. Wilson, I think, is gonna talk quite a bit about this in her session. Our group and others have described the hyperinflammatory phenotype, which is defined in part by higher plasma levels of inflammatory biomarkers and consistently worse outcomes. This phenotype seems to respond differently to mechanical ventilation, conservative versus liberal fluid status, and pharmacotherapy with simvastatin. You may be thinking, I thought you were talking about physiologic phenotypes. Why are you getting into biology? Actually, these phenotypes are a hybrid of clinical, i.e., physiologic data and biological data. And if you look at the characteristics that define these phenotypes, again, here on the x-axis, I've just put red arrows by the clinical variables in this figure, and you can see that actually most of the variables that define these phenotypes are, in fact, clinical, either laboratory or based on physiology, like systolic blood pressure, plateau pressure, et cetera. These physiologic and clinical variables can actually be used to identify these phenotypes, even though they're not the only defining characteristics. This is work by Pratik Sinha and Matt Chirpek using a machine learning model called XGBoost, which found that just using these clinical variables, you can achieve a pretty good area under the curve for identifying these phenotypes in both clinical trial and observational data sets. So it's just to say, well, I'm focused mainly on these respiratory physiology. Of course, there are lots of other physiologic parameters in our patients besides just the lung. Where do I hope we'll go to with this? I mean, I really hope that we'll get to a point where we don't think of phenotypes as just physiologic, just biologic, et cetera, but we really start to think about phenotypes and incorporate all of the different parameters that we measure in our patients. This is a figure taken by a perspective piece led by David Maslow and John Marshall that was published in Nature Medicine last year. And I really, I like this image because I think it summarizes nicely the aspirational goal we have, right, which is that we can take all of these parameters that describe our patients, biological characteristics, clinical factors, physiology, imaging, genetics, multi-omic sequencing, and identify distinct states within our critically ill patients, hopefully be able to identify these patients in a pragmatic way at the bedside, and then really test the hypothesis that these groups respond differently to treatment. Because I think that's really why we're all here is we're trying to find groups that will respond better to the treatments for our patients than what we're giving them today. All right, so in conclusion, I'd say pulmonary physiology in ARDS is heterogeneous, strong evidence-based, supporting differences in treatment based on the PF of 150, suggestive evidence, but no RCT data supporting phenotypes based on lung compliance, and multivariable modeling of multiple respiratory physiologic variables, both in typical and in COVID-19 ARDS. But I think our ultimate goal here will be phenotypes that incorporate multidimensional data, including physiology, and respond differently to treatment. And I hope that our field can move towards more prospective studies with detailed physiologic phenotyping, particularly those that are embedded in RCTs, to understand how best to tailor treatment. Thanks so much, and I look forward to questions.

physiologic subphenotypes

ARDS

respiratory physiology

treatment response

lung compliance