Excellent. Thanks so much. Well, I think Manu has done an excellent job summarizing both sides of the debate, so hopefully it'll be easy to continue to sway you why we shouldn't be just using these in all patients. I've got no financial conflicts. I've put some funding in rolls, but really these are just my views that I'm sharing today. So I think perhaps the reason I was invited to give this talk was from this counterpoint that I published with some colleagues a couple years back where we argued the con side to steroids in ARDS, and we advocated for essentially this framework where you think about what is the likelihood that my patient may benefit from steroids based on their clinical characteristics, and then on the flip side, what are the risks of harm for this individual patient based on their characteristics, based on their comorbidities, based on their sort of acute presentation and competing medical issues, and based on that, trying to bucket people into yes, give steroids, no, don't give steroids, or the sort of really difficult cases in the middle where you're going to have to sort of figure out on a case-by-case basis. And so I'm going to walk through some of the thinking that we did as we worked on this perspective. I'm going to talk about this concept of heterogeneity of treatment effect, also sometimes called heterogeneity of net benefit. I'm going to talk a little bit about guideline recommendations, trying not to go over too much of the same stuff that Manu already did. And then I have three different patient examples about how in sort of just bedside rounds we use this framework to decide whether to proceed with using steroids in patients in the ICU. So first I'll start out with what I learned in medical school, probably what many of you may have learned in medical school, and that is that the results of a trial are generalizable to all patients in that trial and to all future similar patients. So this textbook of evidence-based medicine said, yes, if the patient would have been enrolled, i.e. met the inclusion, didn't have the exclusion criteria, then there is little question that the results are applicable. And then this one on the right says, we suggest that you accept the treatment's overall efficacy as the best starting point for estimating efficacy in your individual patient. And that one gets closer to the truth. And that is that even though there is an average treatment effect from a clinical trial, the individual benefit experienced by patients in that trial can be highly variable. And so the reality that I think has been increasingly realized over the past couple decades is that even within a single trial, the effect of that treatment is often highly variable across patients, particularly in critical care where we have very syndromic sort of definitions and enrollment. So heterogeneity of treatment effect or net benefit is defined as non-random, explainable variability in the direction and magnitude of treatment effects for individuals within a population. And this can happen really for two key reasons. First is differences in the baseline risk of the outcome. So imagine we have a miracle drug for ARDS that reduces mortality by 50%. If your patient is really sick and has a 60% probability of mortality, that's like a 30% absolute risk reduction if you give them the miracle drug. On the flip side, suppose you have a patient who's sort of barely meets criteria. They're already kind of improving. They're much less ill. Maybe their predicted mortality is 10%. The miracle drug will only give them a 5% absolute risk reduction. And then maybe this miracle drug comes with these kind of random catastrophic side effects that happen at random to 5% of people who take it. Suddenly this is no longer a beneficial treatment. So baseline risk of the outcome is really important a lot of times in understanding the heterogeneity of benefit for the patient. Basically at a certain point, if you're not sick enough to experience the outcome, you're not going to benefit from a treatment to protect against that outcome. And then the second issue is differences in pathophysiology that confer a different relative risk reduction. So that gets to differences in underlying mechanism of disease or sort of etiology of ARDS. So classically heterogeneity of treatment effect was really sort of initially described from this study. This was published in New England Journal of Medicine in 1991. This was a clinical trial looking at carotid endarterectomy surgery. And you can see the conclusion here. They say it's highly beneficial in patients who have had recent hemispheric or retinal TIAs or non-disabling strokes and have high-grade ipsilateral stenosis. So overwhelming positive trial. And then this same trial was reanalyzed in this paper in The Lancet a few years later where they took the baseline characteristics of these patients to then predict the likelihood of having a stroke. And they grouped the patients, again, based on baseline sort of pretreatment characteristics of low risk, medium risk, and high risk, and then analyzed that trial based on these subgroups. And so for the patients who were highest risk, this was overwhelmingly beneficial. Number needed to treat of seven patients. But when you look at the lowest risk group, which was 38% of patients who had been enrolled into that clinical trial, there was actually a net harm. And that's because a certain number of patients just end up having a complication of that carotid endarterectomy surgery itself. And so there was a number needed to harm of 71 patients. So variability within this sort of clinical trial population such that the overall average treatment effect is not the same effect experienced by the individual patients. And so some of you may be thinking, well, yeah, but that's a surgery. You know, are we really going to sort of see the same type of thing for something like steroids that doesn't per se come with those same catastrophic complications that could happen with a surgery? But there is evidence that heterogeneity of treatment effect is present for steroids, at least in sepsis. So this is a study, I think a really elegant study, that was published a couple of years back that took data from four different clinical trials of steroids and sepsis. And then it did this statistical analysis where they asked, what would have happened under these various treatment approaches? What would have happened if we treated all patients with steroids, no patients with steroids? Only those patients who are sickest, so kind of a risk-based one, getting at what I was talking about, that a greater baseline risk, that your relative risk reduction confers a greater absolute. So they used an approach of, you know, what if we just gave it to the highest risk patients? And then finally, they used an approach where they used a machine learning model that tried to estimate what was the likely individual treatment effect for these patients. And then they picked different thresholds based on the estimated or predicted individual treatment effect. And so what they found in their analysis is these individualized approaches, both the risk-based and the direct estimation of predicted individual treatment effect, were better than the treat all or the treat none approach. And this super learner model, this machine learning ensemble model that tried to estimate individual treatment effect, performed best, had the sort of maximal overall benefit when 20% of the patients were treated. And so again, this sort of shows you how, for something where we have a conditional recommendation, that at least in this analysis, it suggested that it was really a minority of the patients that sort of bore out the benefit of that treatment that is yielding an overall average benefit to the population. So how do you bring that to bedside? So this was a question we were asked in this accompanying editorial to please address for the reader. And so we sort of grappled with this. And the super learner model is really complex, right? So this is not something that you can sort of do in your head on rounds. This is going to require some kind of app or integration into the electronic health record. And so I think that as we go forward and have more sophisticated models that try to understand heterogeneity of treatment effect, we really need to think about how do you integrate these into your clinical workflow? How do we present this data to clinicians in such a way that is interpretable and credible, such that it then actually leads the clinician to want to act on these recommendations? And ultimately, we need to have, I think, prospective trials that test the benefit of adding these types of tools into our clinical workflow. That's where we need to go in the next years. But in the meantime, I think the study really highlights some of the challenges and nuance of taking clinical trials and the average treatment effects that are generated from clinical trials and trying to apply that to individual patients at the bedside who have huge variability in terms of their acute presentations and also their sort of chronic conditions and sort of overall clinical picture. So is there likely heterogeneity of treatment effect in ARDS? Absolutely. Huge differences in the etiologies of ARDS, big differences in terms of risk of the outcome depending on the severity of ARDS, and then also differences based on patient characteristics in terms of risks of treatment-related harms from steroids. This is a perspective that was a review study that was written a couple years ago that talked about this concept of precision medicine and heterogeneity of treatment effect in ARDS. And they summarized that really there is evidence of heterogeneity for most interventions in ARDS. The exception here really was lung protective ventilation, which seems to be beneficial sort of consistently across the board. Maybe the sort of degree of benefit differs, but consistently beneficial. Whereas for a lot of these other interventions, there was evidence pretty consistently of heterogeneity by the type of lung injury and the severity of lung injury. This study did not, however, look at steroids, was not one of the specific interventions that was examined. So what do recent guidelines say? So Manu previously shared this guideline that was just published a few weeks ago. I thought this figure was excellent because it summarizes their recommendations very nicely. For steroids, there is this conditional recommendation across the board for severity. And then in terms of implications of a weak conditional, we suggest recommendation. They say that different choices will be appropriate for different patients. And then it talks about sort of consideration of preferences and values. I think preferences and values are of course hugely important. I think the other thing that is really going to be important here is like we've talked about the clinical characteristics of the patient at bedside. In terms of the evidence summary, in terms of benefits, probably reduces mortality again for the average treatment effect across patients studied. May reduce duration of ventilation. And then in terms of harms, increases risk of hyperglycemia, may increase risk of GI hemorrhage, uncertain for neuromuscular weakness, and then infectious complications could not be systematically evaluated. The next guideline that I'll highlight here, this just came out this month. This is the SECM update on guidelines for use of steroids in sepsis, ARDS, and pneumonia. I'm just going to focus on the ARDS recommendation where again, consistent with the ATS guideline, this is a conditional recommendation. We suggest administering steroids in patients with ARDS. And for them, for the implications of this week recommendation, they say different choices are likely to be appropriate for different patients. Therapy should be tailored to individual patient circumstances. So again, talking about trying to tailor to the patient based on clinical circumstances. They talk about some of the key reasons for uncertainty in the conditional nature of the recommendation. So a few positive trials had a large impact on this overall positive effect. So it wasn't seen consistently across all studies. There's limited longer term outcome data. And the data evaluated for this guideline, about half of the patients had community acquired pneumonia. So our evidence is much stronger for those patients and less strong for other ideologies of ARDS. And then absence of evidence about the potential harms of steroids, which are not always able to be analyzed across all of these studies. So getting back to what do we do in practice? And so I'm going to return to this framework here where we think on the one axis, how strong is the indication to use steroids based on the evidence that we have? And then how strong is that risk of complications? And trying to sort of work through these two things. And so I'm going to give a few different cases of patients. The first one is a 40-year-old. This patient presents with severe COVID-19 related ARDS, a very low PF ratio, no comorbidities. So this patient is pretty straightforward. This is where we have a lot of evidence that steroids are likely to be beneficial. This is a straightforward presentation in a patient without comorbidities. And we're not told that they have other sort of acute medical problems other than the ARDS. And so the management decision here was to give steroids. Patient two is a patient I had in my ICU recently. This was an 81-year-old gentleman. He had poorly controlled insulin-dependent diabetes. He'd been having a progressive cognitive decline over 6 to 12 months. He had a heart rate loss and was admitted to the hospital with failure to thrive after neighbors had called on a wellness check. And he was then brought into the hospital. When he was in the hospital, he then was drinking water and had a witnessed aspiration and became acutely hypoxic, had a rapid response called, ended up in our ICU. So this patient had aspiration pneumonitis. He was on heated high flow and had hyperactive delirium following this event. So likelihood of benefit for this gentleman, I think, is lower given that we know the etiology was this aspiration on water while he was taking his medications. And then I think risks of harms. He's got a number of reasons why he's probably not going to tolerate steroids well. He's had this progressive cognitive impairment. He's acutely delirious. He's got insulin-dependent diabetes. And he's an older patient who is at risk for poorly tolerating. And so we did not give steroids to this patient. And he weaned down, you know, off of oxygen to room air within about two days. The third patient is a much more challenging patient. I still don't know that we made the right choice, but I'll walk you through kind of our thinking. So this was a 60-year-old patient who had AML, who had received induction chemotherapy, and then had a rapid progression of hypoxemic respiratory failure, ended up in the intensive care unit with mechanical ventilation, 90% FiO2, bilateral infiltrates, and was just really critically ill. And so benefit, I think, was really hard to estimate here. On the one hand, he's got very severe ARDS. But at the same time, it's hard to know in a patient like this who's just had chemotherapy and his immune system is already relatively impacted by that. Risk of harm. I think the key thing here was further infection risk. We ultimately decided to give this patient steroids. Two days later, this patient weaned from about 90% down to 40% oxygen. His infectious workup all came back negative. He had a BAL, bronchoscopy with BAL that looked bloody. And so our theory then was that he had bled, had alveolar hemorrhage related to this critical thrombocytopenia. And now he's got new issues, delirium, hyperglycemia, positive fluid balance. So it's no longer oxygenation that's the key thing keeping him on the ventilator. It's the fact that it's all these other sort of issues and that he has delirium. We're not able to do a good spontaneous breathing trial. And so at this point, we say, no more steroids. And we stopped them. I think he maybe got two doses. And so I think for this one, again, it was really challenging to know what to do up front. But I think every single day, you want to reassess the treatments that you're giving. And are they still indicating? Does the estimated risk-benefit equation still make sense and reassess whether to continue with your treatment going forward? So in sum, I think both Manu and I have conveyed there's big variation in patient presentations. Steroids have poliotropic effects, both good and bad. And so I think it's important to have this day-by-day consideration and try to think about what are the anticipated benefits, what are the anticipated risks, and make a personal ized decision for the patient at bedside. I will close there. Thank you.

ARDS

steroids

personalized medicine

heterogeneity of treatment effect

clinical decision-making