In this forward-looking episode of the SCCM Podcast, Daniel F. McAuley, MD, explores how the clinical and research communities are rethinking acute respiratory distress syndrome (ARDS), shifting from a one-size-fits-all model to a focus on identifying and targeting modifiable traits. Building on his Thought Leader Session at the 2024 Critical Care Congress, Dr. McAuley unpacks the major thematic shift toward precision medicine in critical care.
Instead of treating ARDS as a single, homogenous condition, researchers are increasingly identifying biologically distinct subgroups—especially hyper- and hypoinflammatory phenotypes—that may respond differently to therapies. These insights are fueling a new generation of trials that aim to prospectively apply this knowledge to treatment strategies.
Central to this evolution is the Precision medicine Adaptive platform Network Trial in Hypoaemic acutE respiratory failure (PANTHER), of which Dr. McAuley is a team member. PANTHER is a Bayesian adaptive platform randomized clinical trial studying novel interventions to improve outcomes for patients with acute hypoxemic respiratory failure. Designed to be adaptive and biomarker informed, PANTHER will test therapies such as simvastatin and baricitinib, based on real-time phenotyping of patients with ARDS.
Throughout the episode, Dr. McAuley reflects on how advances in machine learning and biomarker identification are making precision treatment more feasible. He discusses the importance of maintaining evidence-based supportive care, such as lung-protective ventilation and prone positioning, while integrating new targeted therapies. Discover the latest investigations into potential therapeutic agents—including mesenchymal stromal cells, statins, and extracorporeal carbon dioxide removal—as Dr. McAuley aims to translate early findings into tangible improvements in patient outcomes.
This episode offers critical insights into the changing landscape of ARDS research and patient care, as Dr. McAuley articulates a hopeful vision for the future—one in which targeted, individualized treatments can improve outcomes for patients with one of critical care’s most challenging conditions.
Dr. McAuley is a consultant and professor in intensive care medicine in the regional intensive care unit at the Royal Victoria Hospital and Queen’s University of Belfast. He is program director for the Medical Research Council/National Institute for Health and Care Research (MRC/NIHR) Efficacy and Mechanism Evaluation Program and scientific director for programs in NIHR.
Access Dr. McAuley’s Congress Thought Leader Session, ARDS: From Treating a Syndrome to Identifying Modifiable Traits here.
Transcript:
Dr. Lin: Hello, and welcome to the 2024 Congress edition of the Society of Critical Care Medicine podcast. I'm Dr. Luwig Lin, your host. Today I'm joined by Dr. Danny F. McCauley to discuss his William C. Shoemaker Honorary Lecture for this year about the evolution of ARDS from treating a syndrome to identifying modifiable traits. Dr. McCauley is a consultant and professor in intensive care medicine at the Regional Intensive Care Unit at the Royal Victoria Hospital and Queen's University of Belfast. He is the scientific director for several of the UK research programs. I am so pleased and honored to be able to speak with Dr. McCauley about these topics. Before we start, Dr. McCauley, do you have any disclosures to report?
Dr. McAuley: So I've been involved in developing a point of care assay with a company called Randox, which part of the work that is ongoing, but they're not involved in any of the conduct of the trial. That would be the main disclosure.
Dr. Lin: Okay. Well, thank you for that. And let's get started because there obviously is so much to cover about ARDS.
It is one of the fundamental issues that we tackle in the intensive care unit every day, no matter where we are. Maybe I could just start out very broadly and have you give us your thoughts about how the paradigm in ARDS management has evolved over the years that you've been in this business.
Dr. McAuley: Thanks for the opportunity to have this discussion. So I guess the syndrome of ARDS, we're all very familiar with the patient with severe acute hypoxic respiratory failure with associated infiltrates, not due to cardiac dysfunction and that standardization of the syndrome has been incredibly important to facilitate supportive care and we have made significant improvements in supportive care with the finding from the seminal ARDS net paper of lung protective ventilation and more recently the role of prone positioning and possibly the additional benefits of high PEEP in patients with more severe disease and the use of ECMO. And I think that has been the major advantage of that syndromic definition.
But I think that syndromic definition has also been contributory in the limitations around the development of pharmacological therapies for ARDS and even though we might have a common final pathway that looks like ARDS, probably underneath that there are multiple different mechanistic pathways that are activated in different settings and I think the next step in the paradigm really is to move from a very broad syndromic definition but to move towards a more personalized approach, particularly for pharmacological therapies and I think the work by Karl and Kalfi, which is now 10 years old, which really defined that within this overall cohort of patients, whenever you looked at the data, you were able to identify subpopulations or subphenotypes that were present that had different outcomes and those subphenotypes appear to respond differently to both supportive therapies but also interestingly pharmacological therapies and Carolyn did a really elegant reanalysis of a study that we did looking at simvastatin in ARDS, for example, and found that in the two phenotypes that were identified, a hypoinflammatory and a hyperinflammatory, while there was no difference in the hypoinflammatory in terms of response to simvastatin, there was a very significant difference in response to simvastatin in patients who were in the hyperinflammatory phenotype.
Now that has to be taken with a pinch of salt because it was an unplanned secondary analysis but I think that, along with other data, there's some emerging data that there might be differences in hyper and hypoinflammatory phenotypes with steroids, for example, in the setting of COVID. So I think this emerging idea that there are subphenotypes within the overall population that respond differentially to pharmacological therapy suggests that there may be treatable traits or modifiable traits that exist within the overall population to different drugs and I think that's really the next step in the sort of investigation and management of ARDS.
Dr. Lin: Thank you. I think that's a really nice way to establish a broad structure for us in terms of shaping this conversation. So my next question for you is, for you as the bedside clinician, with all the knowledge that you have, let's talk about how your approach would be for, for example, a patient that has hyperinflammatory versus somebody with a hypoinflammatory.
Dr. McAuley: So it's a really good question. Ultimately, at the bedside, you can't tell. So whenever you look at risk factors, for example, sepsis and other causes don't differentiate clearly.
There's more sepsis than hyperinflammatory but not everybody who is hyperinflammatory has sepsis. You can't tell from the severity of the disease. So if you've got a very low PF, that doesn't necessarily mean that you're hyper or hypo.
So at the bedside, it's actually quite difficult. Now, the ability to differentiate these two phenotypes was based on a really complex statistical clustering method called latent class analysis. And while it's possible to do that retrospectively, it's not possible to do that easily at the bedside.
So what Kyle and Kathy and colleagues, Prateek Singha and others were able to do was say, here's this really complicated statistical modelling, but you know what, if we can look at several key factors, and that's a combination of serum bicarbonate and two or three biomarkers. And typically the biomarkers that are most predictive are soluble TNFR1 and IL6. And with those three parameters, you can say with pretty good confidence at the bedside that this patient is hyper or hypo.
The problem is to date, we haven't been able to measure those biomarkers at the bedside, but that's the, I guess, the exciting next step. And there are now point of care devices that can measure IL6 and soluble TNFR1 at the bedside. And I mentioned the Randox device as an example.
There are others I'm sure that will come. So what that means now we've moved from needing to do complex statistics that aren't possible at the bedside to measuring several parameters at the bedside that will allow us then to say this patient is hyper inflammatory and this patient is hypo inflammatory. I guess we need just to be aware of the risk that comes with that without knowing they're associated with benefits prospectively in clinical trials whenever we're testing new therapies.
So I think while it is possible to now phenotype at the bedside, the next step isn't to start giving treatments randomly. The next step is to randomize treatments in the setting of clinical trials. And I think that is probably where we're likely to make significant progress.
There is another way to phenotype at the bedside using a clinical classifier model. And that essentially uses artificial intelligence to take a limited number of parameters. And that's almost as good as the biomarker approach.
The problem is the learning sets for those clinical classifiers are based on a certain data set. And over time, if treatments or severity of illness change, those learning data sets may not be relevant to current practice. So while it's an alternative to the biomarker approach, my general opinion is that I don't think the clinical classifier will better than the biomarker approach.
But it might be an option where the biomarkers aren't available. So there are the two methods that are available, a biomarker approach and a clinical classifier approach using machine learning. But I guess, again, just to emphasize the caution, we shouldn't take the next step to randomly starting to treatments.
We must do clinical trials.
Dr. Lin: Right. That is so interesting. And I have so many follow-up questions I wanted to ask you.
And one of my questions is, it sounds like right now we shouldn't jump the gun and start thinking that we need to be doing things differently. I'd like to get to that part in terms of what we should be doing for all of our patients in terms of what you think the latest is in terms of beneficial treatment for everyone. But I think my immediate follow-up question for you is about the machine learning.
AI is obviously a very sexy topic, but it's also potentially beneficial to a lot of these treatment algorithms that we're trying to implement. And as you pointed out, the data sets, though, could be changing. And I was just thinking about this whole idea of a brave new world.
We are one international population. And a lot of these hyperinflammatory versus hyperinflammatory responses are probably based on one's genotype. And genotypes, thankfully, are mixing.
We are becoming a very different population. So how accurate are these large data sets, especially in light of the rapidly changing demographics?
Dr. McAuley: So that's a really good question. I guess in terms of the machine learning, learning data sets, and then its applicability to prospective patients, but also the diversity of prospective patients, I don't think we know. And I guess that's the biggest risk about applying a learning data set to a population that may not be the same as a learning data set.
So I think that's a really important risk that we need to understand. I think it's also relevant that your genotype and your diversity is probably also likely to drive your biological responses as well. So that criticism might equally be leveled at the biomarker approach as well.
But it's really interesting. Again, Carolyn Calfee has led the way in this. And she identified in the ARDSnet studies that these phenotypes existed.
And that was a very different population that we see in the UK. And yet, genuinely, much to my surprise, whenever we repeated the biomarker approach in the simvastatin study that I mentioned that was a UK-based study, we found the same thing. And Carolyn has gone on now to replicate those findings in over 10 or 12 different populations, trials, observational cohorts, pediatric cohorts.
And consistently, we see the same thing. So I'm much less concerned on the impact of diversity in genomic responses on the biomarker approach based on the data. But I don't think we have that data yet for the machine learning clinical classifier approach.
Dr. Lin: Thank you. I totally see your point. And it's always good to have data rather than why guessing away assumptions, because that's what large data and AI does, is make assumptions.
And thank you for really making us think about that. So then my next follow-up question is, not knowing right now which camp a patient falls in, what are some of the things that we should be doing for every ARDS patient that rolls in?
Dr. McAuley: So I think, and again, much of this work has been informed by the ARDS net and other clinical trials from across the world. I think the gold standard must be lung protective ventilation, aiming for 6 mls per kilogram predictive body weight, and plateau pressures less than 30. It's clearly not as black and white as that.
And whenever the comparator was 12 mls, is there a better tidal volume between 6 and 12? We don't know. The other question is, should we go lower than 6 mls?
And I think we have a bit of evidence. Another study that we did was called the REST trial, where we looked at conventional lung protective ventilation of 6 mls per kilogram versus targeting lower tidal volumes aiming for 3 mls per kilogram facilitated by extracorporeal CO2 removal. And at least in that study, we didn't see any additional benefit by further reduction in tidal volumes.
Now, dose of extracorporeal CO2 that we achieved in that study was relatively modest. So the question may be that if we'd have got the tidal volume down further facilitated by extracorporeal support, there might have been a difference. But at the minute, I think the gold standard should be 6 mls per kilogram.
I think then the next standard of care should be prone positioning in patients who have moderately severe ARDS. And that should be implemented after a period of optimization. PF ratio remains less than 150.
Then we should be giving people ventilation in the prone position for up to about 16 hours daily until they improve. And then after that, the data becomes a bit less convincing. Higher peak is probably important in those with more severe ARDS.
And there's probably sufficient evidence to support ECMO in the very severe cohort. In terms of pharmacological therapies, I have a slide in my talk, which basically is a blank slide. And I don't think we have any convincing pharmacological therapies.
Some recent guidelines have suggested neuromuscular blockade should be considered. I'm not quite convinced. And whenever you see different guidelines concluding different things, that probably means that we don't know.
So I think neuromuscular blockade, not routinely, but perhaps in those patients who have asynchrony, it might be worth a trial. And then the other age old therapy that we still don't know the answer for is steroids. And again, there is, you know, more emerging data to suggest steroids may have a role, but it's by no means convincing.
And there are two really large dexamethasone trials that are now started, one led by the Canadian group looking at dexamethasone and another one led by Manish Shankarhari and myself in the UK, looking at dexamethasone 20 milligrams. And I think those two studies will hopefully give us the answer in terms of steroids, but we're not there yet. So that's, I think, the standard of care currently.
Dr. Lin: Beautiful. Thank you so much for that. Let me ask you yet another follow-up question, I'm full of them today, about what you had just elaborated on.
So the lower tidal volumes, I think initial ARDS net results, the hypothesis was that people had improved outcomes because of decreased tissue damage. And that was linked to, surprise, surprise, IL-6 levels. Do you think there is a difference in that secondary damage in terms of the two groups that you were looking at, the hyper versus hypoinflammatory?
Dr. McAuley: So that's a really good question. So the postulated mechanism is this idea of mechanotransduction. So if you endurously ventilate the lung, that causes systemic inflammation that then causes high cytokine levels, including IL-6, which then causes organ dysfunction in the kidney and the liver.
And there's some really nice data to support that. There certainly is a corollary with that in the hyperinflammatory setting where the levels of cytokines are higher and we see more organ dysfunction in the hyperinflammatory population. So it's an interesting hypothesis.
I guess the other thing, and I think Carolyn would admit this, the terms, or would agree with this, not so much admit, the terms hyperinflammatory and hypoinflammatory are fundamentally flawed. It's probably hyperinflammatory and inflamed, but in a different way. The idea that the hypoinflammatory is not inflamed is not really valid and they definitely have high inflammatory indices.
So there may be different mechanisms of inflammation present in the two different phenotypes that drive a different pattern of organ dysfunction. And then I think the other thing that's really interesting in the two groups is the hypoinflammatory probably has substantially more burden of comorbidities. So it may be that the mortality in the hypoinflammatory is driven more by the comorbidity rather than the actual inflammatory process.
And the hyperinflammatory is predominantly driven by the inflammatory process per se. And what that means is that that group might be more susceptible to having an attributable fraction of the mortality, which could be modifiable by drug therapy. It may be that the mortality in the hypo is just fixed because of the comorbidities that are modifiable, whereas the hyperinflammatory is more likely to be modifiable.
Dr. Lin: Yes, this is rapidly becoming more and more complicated. And it sounds like there's a lot for us to learn. But I think that is probably your original point, which is, you know, personalizing this and really appreciating the diversity involved in this patient population is what we really need to learn about next.
Dr. McAuley: I think that's right. And I think the idea of our approach to critical care has been very much syndrome-based. In my mind, I think the future will be syndrome agnostic.
And I probably shouldn't say that as someone who's spent a career investigating ARDS. But I think what will happen is you'll come into intensive care, you'll be on a ventilator, you'll be on pressers, and you'll be phenotyped independently of a label of a specific syndrome of ARDS and sepsis. But if you have hyperinflammatory, you'll be randomized or maybe even receive specific therapies.
If you're hypoinflammatory, you'll receive something else. So I think that absolutely the future will be syndrome agnostic, I would predict.
Dr. Lin: Hmm. Well, that is the inherent intention in this, in that I think mortality has decreased a lot in the last couple of decades in things like sepsis and ARDS from algorithms in standardizing care. But what you're proposing is now to go sort of beyond that.
And maybe what we'll have to do is do a combination of the two, with maybe the help of things like AI to help sort of alert us to the possibilities.
Dr. McAuley: So I think that's really important. I think it would be remiss if it was not to emphasize the importance of attention to detail and standardization of care. And I think we're not good at that.
I mean, if you look at the data on compliance with lung protective ventilation, we know it is suboptimal, both in the UK and in the US. And there's some health economic data that's a bit historical now, but shows that if you simply employed a person to make sure that patients were compliant with lung protective ventilation, that would have a much greater health economic benefit than most of the other things that we are proposing. And I think, you know, as humans looking after all the stuff that goes on in the intensive care, it's hard to keep a handle on everything.
And I think that's where, you know, something like artificial intelligence will be able to deal with the monotony of getting everything that needs to be delivered as standardized care right, to allow us maybe to think about the potential new therapies that might incrementally reduce mortality further. But it's a really important combination.
Dr. Lin: I like that. Thank you. Let's pivot a little bit and talk about some of your future efforts in research into this area, because obviously, that's your niche is doing the research.
I know that there's a study ongoing right now called PANTHER. Can you tell us more about that?
Dr. McAuley: Yeah, I would love to. So PANTHER is really one of the first studies to actually try and randomize patients based on inflammatory phenotypes that Carolyn has described. And it's great because it's a really fantastic international collaboration.
We've got investigators from 10 countries across the world, including the US, Ireland, Canada, the UK, Australia, and many others. And that in itself is great because it guarantees the diversity that we need. But also whenever we were developing the trial and we spoke to patients, they felt it was really important that we had this international representation because that meant that we were able to find useful therapies.
It was more likely they would be adopted into practice internationally. I think there is a historical challenge that if we find something in the UK, it's not implemented in the US. And if we find something in the US, it's not implemented in the UK.
So I think it's useful for implementation, hearing what the patients think. So as well as that, the program is really heavily inclusive of early career researchers, which we think is also important. And essentially what PANTHER is, it's a platform trial.
And what that means theoretically is that it could go on forever. So we could be testing new therapies and adapting as we continue to find therapies that work to include those in standard care and continue to find new therapies. So as a result, that emphasises the need to have early career researchers embedded.
People like me and others will disappear, but we want to make sure that the trial continues and continues to find useful therapies. So what the essential plan is, is to recruit patients who fulfil the new global definition of ARDS. And one of the comments to that might be, well, that's even more heterogeneous than the previous definition, but that's okay because we're going to then phenotype those patients into hyperinflammatory and hypoinflammatory ARDS.
And then we're going to randomise patients to treatment compared to usual care. And the way the platform is set up is that we'll have two active interventions running compared to usual care at any one time. And the two therapies that we're going to start the trial with are simvastatin, and that's because of the secondary analysis that we have from the HARP2 trial.
And also perhaps because as anyone who knows me, I just love trying to prove that simvastatin works for something other than cholesterol reduction. So we're going to test simvastatin and baricitinib as the first two therapies. And in collaboration with our colleagues in Imperial College in London, we have put together a really strong methods team with an expertise in Bayesian trials and adaptive trials.
And what that means is that essentially you learn as the trial continues. So the classical trial compares group A versus group B with a predefined sample size. And the disadvantage of that is that you may get to your thousand patient sample size and you have a strong signal, but you don't actually hit a statistical trigger.
And there remains uncertainty as to whether or not the treatment is effective or not. Essentially with the Bayesian approach, it's very much based around probabilities. And there are a priori defined points at which there is a interim analysis conducted, and that's conducted in a blinded way with oversight from the Data Safety Monitoring Committee.
And there are predefined points where you say this probability defines benefit or this probability defines futility. And it's only whenever those points are reached that you actually stop the trial. So you stop the trial at a point where you know you have an answer rather than hoping that your fixed sample size gives you the answer.
And the primary outcome in that trial will be 20 a day organ failure free days. And that's quite a useful outcome, not only because it's important for patients. We know from talking to patients that they value that as an outcome, but also it translates, even though it's at 28 days, we know from RemapCap that it translates into 180 day outcomes as well.
So it's a good short term outcome for long term benefit as well. And the other advantage is it's statistically very powerful, so it increases the efficiency of the trial. So it's important as an outcome, but also beneficial for trial design and delivery.
So that's the brief summary of the trial.
Dr. Lin: Sounds really exciting. It also sounds like the future is now not just in terms of various technologies, but in terms of the way we actually design and conduct the research. That's pretty cool stuff.
So thank you for explaining that to us. I feel like we could go on talking forever about this topic. We eventually do need to draw this episode to a close.
So I wanted to make sure that I asked you, is there anything that we didn't cover that you feel like are essential to have our audience take away from this conversation today?
Dr. McAuley: No, I mean, I think you've summarized it really well in your discussion as well. I mean, I think that the core should be making sure we deliver excellent standard of care. But I think now with point of care, diagnostics, novel trial design, we're really at a cusp where we hopefully will find some new therapies for our patients with ARDS.
That'd be really great.
Dr. Lin: Well, I think I would like to thank you for your efforts in all of this. And it's also just incredibly intellectually stimulating and inspiring to hear you discuss all of it. And I really can't wait for us to have some follow-up conversations and to read about the results.
So thank you. This is going to conclude another episode of our Society for Critical Care Medicine podcast. If you liked it, definitely subscribe to our podcast series.
And thank you for listening. This is Dr. Ludwig Lin.
Announcer: Ludwig H. Lin, MD, is an intensivist and anesthesiologist at Sutter Hospitals in the Bay Area of Northern California and is a consulting professor at Stanford University School of Medicine, where he teaches a seminar on the psychosocial and economic ramifications of critical illness. The SCCM podcast is the copyrighted material of the Society of Critical Care Medicine, and all rights are reserved.
Find more episodes at sccm.org/podcast. This podcast is for educational purposes only. The material presented is intended to represent an approach, view, statement, or opinion of the presenter that may be helpful to others.
The views and opinions expressed herein are those of the presenters and do not necessarily reflect the opinions or views of SCCM. SCCM does not recommend or endorse any specific test, physician, product, procedure, opinion, or other information that may be mentioned. Join or renew your membership with SCCM, the only multi-professional society dedicated exclusively to the advancement of critical care.
Contact a customer service representative at 847-827-6888 or visit sccm.org/membership for more information.
