Skip to content
Oxygenation Targets and Outcomes after Critical Illness episode artwork

May 31, 2025 49 min

Oxygenation Targets and Outcomes after Critical Illness

from Maryland CC Project

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

Oxygenation Targets and Outcomes after Critical Illness

0:00 49:41
of matches

Transcript Auto-generated, may contain errors

Thank you, everybody, for joining us today for our Critical Care Ground Rounds. It's my pleasure to introduce Dr. Matthew Mart here, coming from Vanderbilt. He's an assistant professor there in the Division of Allergy, Pulmonary, and Critical Care Medicine. He did a lot of his training down in Kentucky. He got his MD there and then stayed there for internal medicine and ended up doing his fellowship at Vanderbilt. He's now stayed on his faculty there. During that time, he also participated in T32 and completed his Master of Science and Clinical Investigation degree. His research focused on critical illness, brain dysfunction, and survivorship.

And he's here to speak a little bit more about oxygenation targets and outcomes in critical illness. And so we're excited to have you here and hear what you have to teach. Great. Thank you, Caleb. And thank you for the opportunity to come speak about something that's an exciting topic to me and something that I think is worthwhile for us to discuss as a broader ICU community. So I don't have any particular disclosures that are relevant to this talk. I will acknowledge that I do have some funding. One, through the Department of Veterans Affairs, I have a career development award. And then I also have a small, I'm a co-investigator on a COVID-19 rehabilitation study funded by the NIH.

So the objectives for today's talk are to give some background regarding studies on oxygen targets in the ICU, including reviewing a few of the major trials in the last five years, particularly as it pertains to a broad general ICU population. Following that review, I'll transition to discussing some of the rationale behind studying the impact of oxygen targets on long-term outcomes, including key components of what we describe as the post-intensive care syndrome, including cognitive impairment. And then I'll move on to discuss the findings of our co-pilot study. And I'll finish up time permitting with some thoughts about where the future of this research may lead. So in the ICU, as we all know, providing supplemental oxygen is fundamental to what we do as intensivists, whether it's through invasive mechanical ventilation or other less invasive means, such as high-flow nasal cannula, non-invasive positive pressure ventilation, et cetera. And if you think about your own day-to-day workflow in the ICU, titrating oxygen is among the most common interventions that you probably provide on a day-to-day basis. We may not think of it as an intervention on a day-to-day basis, but it is truly an intervention, much like giving IV fluids.

Despite being such a common intervention, however, there are widely varying recommendations for oxygen targets across the international guidelines. For example, the NIH through the ARDSnet has traditionally recommended a goal saturation of 88 to 92%, whereas the most recent British Thoracic Society guidelines recommend oxygen targets as high as 96 to 100%. The Thoracic Society of Australia and New Zealand instead offer an intermediate range of 92 to 96%. And so with these differing guidelines and differing recommendations, I think it really behooves us as intensivists and as inpatient clinical providers to think about what is our goal and what are the risks of providing supplemental oxygen. And I'll argue that there are potential risks in both under-providing and over-providing oxygen therapy and targeting lower versus higher oxygen saturation targets. So I would offer that we should always think about what that is, what the right dose may be when we're providing supplemental oxygen. By targeting higher oxygen targets, we could be potentially avoiding...

lower arterial oxygen content, we could be avoiding inadvertent episodes of hypoxemia and potentially even the risk of ischemia. But by targeting lower oxygen targets, we could potentially be avoiding reperfusion injury or direct lung toxicity, or perhaps avoiding oxidative stress that can be induced by hyperoxia in the brain, in the skeletal muscle, and other organ systems. Or perhaps somewhere in the middle is best, avoiding both the risk of hypoxia and hyperoxia. And interestingly enough, intermediate oxygen targets have probably received less study than higher and lower targets, as you'll see when I go through some of the original data. So what do the data tell us about oxygen targets in a general ICU population? Well, observational data across multiple studies has suggested actually that outcomes may follow a U-shaped curve with regards to outcomes, with both low and high targets potentially leading towards outcomes. This paper is a nice demonstration of this by Vandenboom and colleagues, which was published in Chess in 2020, and it reflects this U-shaped curve that I mentioned. They conducted a retrospective analysis of two different electronic medical record databases, the EICU Collaborative Research Database.

and the Medical Information Mart for Intensive Care 3 database or the MIMIC database that many of you may be familiar with. And they performed regression analyses to analyze the association between the median SpO2 and hospital mortality. And they adjusted for things that we would expect to be associated with poor outcomes, age, BMI, SOFA score, along with duration of oxygen therapy. And they pre-specified some subgroup analyses for ICU type, main diagnosis, and comorbidities. They analyzed over 35,000 patients between both of these EMR cohorts, and they found that mortality was actually lowest in the intermediate range. And this finding was consistent in both cohorts, both database cohorts, and across all the subgroup analyses, again, suggesting that perhaps an intermediate target balances the benefits and risk of both lower and higher oxygen saturation targets.

But what about clinical trial data? I think observational studies are vital to the discovery process. They're certainly part of what I do in my own research. But to truly understand causation, we really need clinical trials. And interestingly, at least earlier clinical trials, which here I'm defining as far back as 1976 and as most recent as 2017, these earlier trials may have suggested that conservative oxygen therapy may be more effective in improving outcomes. This is a meta-analysis called IOTA meta-analysis.

which looked at 25 randomized controlled trials of oxygenation targets and included up to 16,000 patients, all who had a variety of diagnoses, including sepsis, critical illness, stroke, trauma, myocardial infarction, emergency surgery, and cardiac arrest. And they made a comparison between what they defined as a liberal oxygen target, which for this particular meta-analysis they defined as 94% to 99%, and a conservative target, which again was 93% to 98%. Of course, they're collating all of these different studies who have slightly different methodologies, so you can see the wide range of how they describe conservative versus liberal oxygen targets. And what they found was that compared to a conservative oxygen strategy, a liberal oxygen strategy was actually associated with an increase in hospital and 30-day mortality, as well as mortality at the longest follow-up. However, other outcomes, such as the modified Rankin score for disability, were similar between the two groups. And the authors note that the mortality findings that they see were similar across some groups of diagnoses. And you can see from this Kaplan-Meier survival analysis plot that there was a decent separation around a little before day 30, and that continues to their longest follow-up in these studies.

Now, I think this was an important advance in our understanding of oxygen therapies and oxygen targets up until this point, but there were certain limitations with this meta-analysis. Patration targets differed across studies. They typically had different approaches to target to obtaining oxygen targets. Either they were aiming for a fixed dose of the FiO2 or they were targeting oxygen saturation targets, which makes it a little hard to compare studies. They had variable follow-up. They had a combination of primary admitting diagnoses across multiple small trials. And so the majority of studies in this meta-analysis actually had fewer than 300 patients. And it was heavily weighted by the oxygen ICU trial, which I'm not going to go into a lot of detail here. But to say only that, that trial had multiple problems with methodology, including some skewed baseline characteristics.

that favored the conservative therapy arm. And so given these limitations, including the heterogeneity of study populations and the meta-analysis, it was pretty clear that we needed additional adequately powered studies to generalize these findings to a broader ICU population. So what I'm going to do is go over what I think are some of the most important trials that informed our own work. And of course, I'm not going to hit on every single oxygen target trial in the last 10 years, but I think these are going to be the ones that are demonstrative of the mixed data that exist in this area. So since 2020, there's been several randomized trials looking at oxygen targets in different populations, including broad general ICU populations, as well as more specific disease processes such as ARDS. And two of these studies were actually both published in March 2020 in the New England Journal of Medicine. And the first one I'm going to discuss is the LOCO2 trial.

So this trial was a multi-center parallel arm randomized trial of conservative oxygen therapy, which they defined as a PaO2 of 55 to 70, with an SpO2 maintained between 88 to 92 percent, versus liberal oxygen therapy, which they defined as a PaO2 target of 90 to 100, with an SpO2 target greater than 96 percent. This was in critically ill adults with ARDS per the Berlin criteria who were intubated and receiving mechanical ventilation, and they had to have been mechanically ventilated less than 12 hours prior to enrollment to be eligible. I will note that, interestingly, they did exclude patients who had had a cardiac arrest, who had a traumatic brain injury, and they also excluded those patients who required long-term oxygen therapy prior to their acute illness. And their primary outcome in the study was death from any cause at 28 days. Then they had additional secondary outcomes that included ICU and 90-day mortality.

So they enrolled 205 patients, and they actually had four who were withdrawn, leaving 201 patients for the final analysis. And for the primary outcome, they found no statistical difference in 28-day mortality between the conservative oxygen group, which had a mortality of 34%, and the liberal oxygen therapy group, which had a mortality of 26.5%. Very notably, following one of the pre-planned interim analyses for this trial, safety data showed that in the conservative oxygen arm, five patients had experienced mesenteric ischemia, with none in the liberal oxygen group arm. Given this, the Data Safety Monitoring Board decided to stop the trial due to the increased potential risk of adverse events and likely futility.

And I'll note that this sample size of 201 patients is notably lower than the proposed sample size of 850 patients, which they felt would allow them to detect a 9% mortality difference with 90% power. And I think we could have a lot of discussion about whether a 9% mortality difference in a critical care study is actually an achievable goal with what we know about most critically ill patients, but that it is important to note the study was terminated early. And so the risk of an effect overestimation here is potentially high. However, if we look at the Kaplan-Meier curve out to 90 days, even though the study was terminated early, there is an early separation in those curves between the two study arms. And so what I would say is that these data are at least provocative to suggest maybe there's potential harm, at least in patients with ARDS with lower auction targets. And this, of course, was in contrast to the meta-analysis I just showed you of earlier studies that showed a potential benefit with conservative auction targets. So I would say at this point in our scientific knowledge, we have conflicting data.

But also published at the same time as the LOCO2 trial was the ICU ROX trial, which was a larger study with a more heterogeneous patient population, different oxygenation targets, and a different primary outcome. So ICU ROX was a multi-center RCT that randomized 965 critically ill adults in 21 ICUs in Australia and New Zealand who were mechanically ventilated, and they were randomized to either see conservative oxygen therapy or what they described as usual oxygen therapy. The way they operationalized this was by setting a lower limit for both groups of 90%. So both the conservative and the usual care arm said that the lowest you can go on your SpO2 is 90%. But in the conservative oxygen arm, if the SpO2 reached 97% or higher, the FiO2 on the ventilator was actively titrated down until the SpO2 was less than 97% or the FiO2 reached 21%. There was no titration, however, in the usual care arm for SpO2 values of 97% or higher.

This time for this study, the primary outcome was ventilator-free days from the time of randomization until day 28. So a little bit different outcome than what was seen in the LOCO2 trial. And so for this study, the median age was around 58 years. And about a third of these patients actually were admitted after surgery. And 40% had what they described as acute brain disease, meaning which they defined as having traumatic brain injury, ischemic encephalopathy, a CNS infection, stroke, subarachnoid hemorrhage, or another CNS pathology. And what they found was that the number of ventilator-free days did not differ significantly between the conservative oxygen arm and the usual oxygen arm, with a median duration of around 21.

0.3 days and 22.1 days respectively, with an absolute difference of negative 0.3 days and a non-significant p-value. So with this study, you see a slightly different population, you see a slightly different outcome, and you don't see any difference between these two groups. But usual care and conservative care in this particular study were interesting and not 100% comparable to other works such as in LOCO2. I also say that if you look at the Kaplan-Meier curves for survival in the ICU-ROX trial, it makes it pretty clear there was no difference between the two. One notable finding, though, in ICU-ROX is that the patients with ischemic hypoxic encephalopathy were found to have a potential benefit from conservative oxygen strategy. Among those patients, those with conservative oxygen therapy, 43% had died by day 180, whereas 59% had died in the usual care arm.

So in contrast to the LOCO2 study, there was no difference between oxygen targets in this larger trial, but there was a possible signal for a differential effect in those patients who had suffered from some type of hypoxic encephalopathy. And so the data at this point, guiding most intensivists on oxygenation targets, I would argue remain mixed, at least in terms of how to manage a general broad ICU population and how to apply this to every patient that walks into your intensive care unit. So this was followed by the HOT ICU trial, which was published again in the New England Journal in 2021. This was a multicenter RCT that randomized almost 3,000 patients across 35 ICUs in Denmark, Switzerland, Finland, the Netherlands, Norway, the United Kingdom, and Iceland. And the population here were patients with acute hypoxic respiratory failure, which they defined as receiving at least 10 liters of oxygen in an open system or an FiO2 of 50% in a closed system to receive either a targeted therapy towards a PaO2 of 60 in the conservative arm or the lower arm versus a PaO2 of 90 in the higher target arm. I will note that all of these patients actually also had to have an arterial arm placed, which depending on your own institution may or may not reflect common practice in your ICU. The primary outcome here was death by 90 days. And interestingly enough, most of these patients were a little older. They were about 70%. Their median age was 70 with approximately 65% being male. The majority were admitted for medical critical illness with pneumonia being the most common admitting diagnosis at 57%.

And about 60% of these patients received invasive mechanical ventilation in both arms, and about 12% to 13% received non-invasive ventilation. And by 90 days, approximately 42% of patients had died in both arms, and that left an adjusted risk ratio of 1.02 with a non-skipping p-value and a 95% confidence interval that crossed one. And you can see that the probability of survival here is not very different in our Kaplan-Meier curve that was published in this journal. Now, there's certainly some limitations to the study as well. The median PaO2 in the lower target group was actually about 70, which I think demonstrates the fact that it's a little harder to titrate a PaO2 of 60 than we think it is, depending on the patient population that you're studying. And the median SpO2 compared between the groups was 93 versus 96%, which may not be significantly different. And again, the one thing we don't see in this study is we don't see an intermediate target. So the earlier observational data that I showed you is that perhaps an intermediate target would actually be a better target for all outcomes in our general ICU population.

And so that led to work done here at Vanderbilt on a study called a pragmatic investigation of optimal oxygen targets or the pilot trial. And I'll spend a little more time talking about this trial because it sets the framework for the copilot long-term outcome study that I'll describe here shortly. So this was a single center, pragmatic, cluster randomized, cluster crossover trial that randomized all mechanically ventilated patients admitted to the medical ICU at Vanderbilt or to the ED within a planned admission to the MICU. And the intervention was targeting either a lower oxygen target, defined as a goal of 90%, versus an intermediate target, which was 94%, versus a higher oxygen target at 98%. These were critically ill adults. No kids were involved in this. The primary outcome was days alive and free of mechanical ventilation through day 28.

And I want to point out this particular study graphic because this was a pragmatic cluster randomized trial, which if you don't have as much experience with it, may be a little bit confusing. The study was conducted with a waiver of consent, meaning every patient, unless they met an exclusion, for example, they were pregnant or a prisoner, was randomized to receive the study intervention as soon as they hit the door. So if a patient came into the emergency room, was intubated, and then admitted to the medical ICU, at time zero, the time of intubation, they were randomized to the study because they were intubated in a participating unit. And for this trial, a single oxygen target was assigned for two months at a time, which is called cluster-level allocation. And every two months, the SpO2 target was switched in a randomly generated sequence, which is the cluster-level crossover. And the allocation sequence was concealed and blinded to investigators at the start of each two-month study period.

The FO2 was titrated to the target SpO2 beginning within 15 minutes of the initiation of mechanical ventilation, and it stopped when mechanical ventilation was discontinued or the patient was transferred out of a participating ICU, i.e. the medical ICU. All other aspects of mechanical ventilation, including analgesia, sedation, and timing of extubation, were determined by treating clinicians. So I think that's worthwhile to understand the overall study design here. Here's a table highlighting the auction targets in the pilot study more clearly. The trial intervention itself targeted SpO2 values rather than PaO2 values, but PaO2 values could be targeted if the SpO2 values were felt to be inaccurate. And as I mentioned, all other aspects of invasive mechanical ventilation, such as tidal volume, PEEP, use of rescue therapies for hypoxemia, were all determined by treating clinicians and institution protocols.

Clinicians could choose to modify the assigned protocol for certain specific reasons. Some of those included pneumothorax, the presence of carbon monoxide poisoning, bleomycin toxicity, et cetera. And we wanted to give clinicians the ability to titrate or change the oxygen target if they felt it was clinically unsafe to participate in this. Because again, all patients who came through the ICU were randomized to receive this during the time the study was running. So these are the baseline study characteristics. The patients had a median age somewhere around 58. About half the patients were female. And you can see the acute illnesses here, about 15, anywhere from 11 to 15 percent of patients had an acute cardiac arrest or they had sepsis or septic shock. About a third had that. And the median non-respiratory SOFA score was about five, which, again, that just removes the respiratory component because we're controlling that in the trial.

This is the primary outcome for the study. And so you can see here that the probability of survival at the top is similar between all three groups. And then a kind of secondary outcome here for breathing without mechanical ventilation, basically how many patients were extubated and not requiring additional mechanical ventilatory support was similar between all three groups. Now, an important question we wanted to ask as part of the study is, were there certain subgroups that may benefit differentially from lower versus intermediate versus higher oxygen targets? As you saw in the prior studies that showed, there may be potential groups that benefit from different targets. And we wanted to pre-specify this analysis up front. And we did this. And so in the cardiac, we had looked at cardiac arrest, acute myocardial infarction, ARDS, sepsis, receipt of vasopressor, and wanted to know, did outcomes differ at all between those pre-specified subgroups? And they did not cross any of these, as you can see from the overall plots here.

So what do we take away from this? Well, I would say, you know, thus far the trial data are mixed. But in general, I'd say it doesn't appear that oxygenation targets impact short-term mortality or expedite liberation from mechanical ventilation, at least in a general ICU population. But it's unclear if there are subgroups that may benefit from one strategy versus another. For example, are there different outcomes in patients with ischemic encephalopathy versus ARDS versus other subgroups perhaps we've not identified?

But more in line with what I study on a day-to-day basis, how do these oxygenation targets, this common intervention that we provide every day at the bedside, how do they potentially impact longer-term outcomes among patients who survive their critical illness? Again, what are our goals when we're providing this particular bedside intervention? So many of you know this, but about three quarters of our patients will survive their ICU hospitalization if we look at national data. But up to half the survivors will suffer from one or more impairments in cognitive, physical, or mental health function. And despite multiple disease processes leading to critical illness, we see similar impairments across multiple domains when we follow these patients up long term. And this constellation of symptoms and syndromes is collectively known as the post-intensive care syndrome, or PICS.

In work done here at Vanderbilt that really helped advance our understanding of the degree of impairments that patients can have after a critical illness, Prateek Pondarapandi, Wesley, and others conducted the Brain ICU Prospective Cohort Study. This was a study that enrolled 821 patients admitted to the ICU, multiple ICUs here at Vanderbilt, for respiratory failure or shock. And it followed these patients longitudinally out to 12 months. And numerous outcomes were assessed, but the primary outcome was actually a measure of cognition, which we assessed here using this instrument called the repeatable battery for the assessment of neuropsychological status, which is a comprehensive assessment of cognition. A normal score here is 100, with a standard deviation of 15, which is age-adjusted. And as you can see, across all these age ranges, global scores were low, even though this study actively excluded patients with pre-existing cognitive impairment, dementia, or other acute brain pathologies, and the models were adjusted for a surrogate marker of cognition prior to ICU admission.

Cognition scores here were similar to those seen in mild cognitive impairment and in TBI populations, with about 25% of participants at 12 months having a severity of cognitive impairment comparable to that of Alzheimer's disease. And so I think this study was really fundamental to our understanding of post-intensive care syndrome and the deficits that our patients experience that impact both their lives and their recovery. One of the other important findings from this study was that even after adjusting for multiple markers of severity of illness, one of the strongest independent predictors of cognitive impairment after the ICU was actually the duration of delirium. And this finding has been replicated now in multiple studies, and it's one of the key motivations for proactively screening identifying delirium in the ICU, as well as avoiding no delirium precipitants, such as benzodiazepines, in most patients. Now, that's a little bit of a simplification of the findings around delirium and long-term cognitive impairment, because really the pathophysiology of both long-term cognitive impairment and delirium is really complex and multifactorial.

That being said, one of the more prominent potential pathophysiologies leading to both delirium and long-term cognitive impairment is that of impaired brain energy metabolism. And there's been this longstanding theory called the cerebral metabolic insufficiency hypothesis that proposes that delirium actually may be precipitated due to failure to meet the energy requirement of the brain, which is really substantial relative to its size. And I think this figure highlights that point nicely. The brain only makes up about 2% of total body weight, yet it receives about 15% of the total cardiac output, and it accounts for 20% of total body oxygen consumption at rest, along with 25% of glucose consumption. As you can see, these are pretty substantial energetic needs, and it's not too hard to imagine that in profoundly deranged physiology of critical illness, that alterations in oxygen delivery to the brain may be the match that lights the tender box for delirium in our patients, again, outside of even the context of a direct brain injury.

And there is at least some clinical evidence that these derangements in oxygenation can contribute to acute brain dysfunction in the form of delirium and long-term cognitive impairment. This is a paper from a colleague and friend, Tim Girard, who's at Pittsburgh, where he identified clinical phenotypes of delirium in the original brain ICU cohort that I mentioned, including delirium driven by hypoxia. And what he found was that longer durations of hypoxic delirium, along with septic delirium and unclassified delirium, were predictive of worse cognitive function at 12 months. Whereas, interestingly, the duration of metabolic delirium or delirium brought on by multiple metabolic derangements were not predictive of cognition at 12 months. I think that's really striking and perhaps suggests that something like oxygenation is something that could be modified to improve potential outcomes, at least in the short term.

Additionally, outside of the context of delirium, the brain itself is really sensitive to hypoxemia-related injury, especially as it ages. Neurons themselves are susceptible to injury, even with brief decreases in oxygen delivery, and inadvertent hypoxemia occurs not that infrequently during mechanical ventilation, especially if we're targeting lower oxygen targets. Additionally, the brain may also be susceptible to really high levels of systemic oxygen therapy, which can generate oxidative stress and free radicals. So combining the available data with potential pathophysiologic mechanisms related to oxygen and energy metabolism, along with the fact that recommendations regarding oxygen therapy and oxygen targets vary widely, and that inadvertent hypoxemia and hyperoxy are both common in the ICU, we felt like this was a really great opportunity to try to rigorously study this and understand the impact of a modifiable clinical practice like oxygenation targets on long-term outcomes like cognitive function in our ICU survivors.

So to answer this question, we developed and conducted the cognitive outcomes in the pilot trial or co-pilot. This was a follow-up study among survivors who were originally enrolled in the pilot trial and survived out to 12 months after their critical illness. And as you may recall that I mentioned, this was a study that the original pilot trial was one of a waiver of consent, meaning patients were not individually consented when they came into the ICU. So we had to follow these patients up and actually enroll them in the study as part of the long-term follow-up. So approximately one year after enrollment, patients were screened for eligibility criteria, which for the most part were the same as those in the original pilot study, in addition to having not died in the interim between that time. A letter was sent out to the patients explaining the study and stated that they would be contacted by phone. And the letter provided participants the opportunity to decline or opt out of study participation by mailing back a postcard or contacting the study coordinator on the provided information. At approximately 12 months, eligible patients who had not declined study participation were contacted. And if they agreed, they provided consent over the phone.

And after informed consent was obtained, trained study personnel who were blinded to their original trial group assignments measured a variety of long-term outcomes, with the primary outcome here being the telephone version of the MOCA cognition screening instrument. We also measured, in a smaller subset of patients, a more in-depth set of cognitive batteries, including different domains of cognition, such as executive function, along with measures of disability and quality of life. Our statistical analysis approach, I think, deserves a little bit of explanation. We use a proportional odds logistic regression model, which generates odds ratios, which I'll explain when we show the results. And the independent variable for our statistical approach was oxygenation target assignment. And we pre-specified our covariates in our model. We wanted to adjust for confounding due to race, age, sex, comorbidities that are listed here, years of education, the presence of cardiac arrest, and the baseline non-respiratory SOFA score, as well as time since enrollment, because we thought that may make a difference.

One of the difficulties in conducting long-term outcome studies in critically ill population is the high risk of a competing event. And in the statistical world, we think about this in a few different ways. But death is a competing event, right? You cannot have an outcome. If you die before, we can even measure the outcome. Or if you have such a profound disability or impairment that you can't participate in long-term follow-up. And so we did an additional analysis in this study where we defined death as a T-MOCA score of negative one, giving it the worst possible score.

And for the handful of patients who had such profound cognitive impairment that they couldn't participate in the study, we assigned them a T-Mobile score of zero, again, trying to compensate for the fact that there are competing risks in the long-term studies of critical illness survivors. And lastly, we also replicated some of the heterogeneity of treatment effect analyses to try to understand if there are subgroups of survivors, not just patients admitted to the ICU, but survivors who may differentially benefit from different oxygen targets. This is our consort diagram. So there were 2,541 patients who were originally enrolled in their original pilot study when they hit either the emergency room or the Vanderbilt MICU doors.

There were 808 patients in the lower SPO2 group, 859 in the intermediate group, and 874 in the higher SPO2 group. In each study arm, approximately the same number of participants died either in the hospital or between hospital discharge in 12-month follow-up, with mortality rates being the same between groups. And after accounting for mortality, study attrition, 501 patients were enrolled across the three arms in the study and completed the assessment battery at 12 months. These are our baseline characteristics by treatment group. They're very similar to the findings in the original pilot trial.

And these were primary results. And so I'm going to give some context for those of you that may not think about MOCA scores and cognition scores on a day-to-day basis, somewhat like I do. So the T-MOCA score is an assessment that scored from 0 to 21 with higher scores indicating greater cognition, much similar to the original MOCA battery. Scores of 19 or 18, depending on the source that you cite, indicate cognitive impairment. And so here, the presence of higher adjusted odds ratios to the right indicate greater odds of a higher T-MOCA score, meaning better cognition.

As you can see, the median T-MOCA scores were the same across all three arms at 17, which is not normal. And in making the compare-wise comparisons between groups, there was no difference in T-MOCA score among patients assigned to lower versus intermediate, intermediate versus higher, or lower versus higher oxygenation targets at 12-month follow-up. I will note that I think it's notable here that while the majority of patients had a normal cognition according to the T-MOCA score at follow-up, between 30% to 40% of participants did have evidence of cognitive impairment, which in my mind is sobering but consistent with prior work in this area. We also measured a variety of secondary cognitive outcomes. The scores here are reported as standardized Z-scores, which basically measure the number of standard deviations that a given data point is away from the mean, with negative scores indicating that a cognitive domain is worse than the standardized mean. The global cognition composite score that you see at the top is just an aggregate of all the Z-scores that you have for each of the individual components of this cognitive battery. As you can see, the scores are similar across groups. I will highlight that executive function, which is measured by the Haley test here,

here, is one of the domains most commonly affected by ICU survivors, which is executive function. And it was consistently low across groups, which is, again, consistent with prior data. And as I mentioned, we also want to understand if auction targets really varied between pre-specified subgroups to assess for this possibility of heterogeneity of treatment effect. And we pre-specified these differences in outcomes with regard to race. clinical diagnoses of cardiac arrest, COPD, heart failure, coronary heart disease, and end-stage renal disease. And we really were interested actually in the race question because of the concern about pulse oximeter is actually not working as well in individuals with darker skin tone. And so we wanted to make sure that we didn't see a difference, at least in the data that we had available. And again, we saw no difference across groups.

Again, consistent with the message that I've given across this entire study, was that we also wanted to assess disability and quality of life, and there was no difference between these groups either. Here, lower CATS or functional activity questionnaire scores mean less disability, and higher EQ5E index scores mean greater quality of life, and there's no difference when you make pairwise comparisons between all three groups. And so what are the take-homes? So I would say the receipt of lower versus intermediate versus higher oxygenation targets was not associated with any differences in cognition or other patient-centered outcomes. We didn't find evidence of heterogeneity of treatment effect with regard to important subgroups. But I think methodologically, it's important to know that it is feasible to measure these long-term outcomes in a pragmatic critical care trial. And these pragmatic critical care trials, I think, are really valuable for us as a scientific discipline. And we need to continue to find ways to incorporate patient-centered outcomes and follow these longitudinally. And I think this study helps demonstrate that this is possible to do.

Now, like with any study, there are certain limitations. This is a single center study. We had purposely broad inclusion criteria, so we enrolled every single person coming into the medical ICU for the duration of the study. That being said, that lends itself to generalizability, but it doesn't help you if you think that a particular subgroup has a potential benefit with a lower target versus a higher target, for example. Because of the pragmatic nature of the study, we weren't able to characterize based on cognitive function, though I will say randomization in this instance actually helps reduce confounding, and hopefully the amount of pre-existing cognitive impairment was equal between groups.

We use the T-MOCA score, which is more of a cognitive screening tool rather than an in-depth neuropsychological battery. Again, because of the pragmatic nature of the study, this made more sense than trying to do a 45 minutes and hour long cognitive assessment over the phone. Study attrition is one of the inevitable things in long-term critical care outcome studies, and it's one of the things we think about a lot in trying to handle this statistically. I will say here that mortality was the same between groups, and actually, if anyone's interested in the data, the baseline characteristics of those patients who participated in the study versus those who were lost to follow-up or declined versus those who died were similar between all three groups. We think this is less of a limitation. And I'll point out that the absolute numbers in the heterogeneity of treatment effects subgroups is relatively small, and so you would really need an adequately powered trial enrolling participants with that diagnosis to more definitively determine the presence or absence of benefit in a certain subgroup.

So how does this compare to other data, long-term outcomes data in these oxygenation trials? I will say there's actually not much data out there beyond what we published. This is from the ICROC study where they followed participants out to 180 days and they did a telephone interview for cognitive status or TIC score in these participants. And these data actually match what we see here is that there's no difference, at least in these two groups. There's a category of severity of cognitive impairments a little different than what we use, but again, somewhat similar. There's more data to come. The HOT ICU has a protocol paper out indicating that they're going to have cognitive function data. And I'm going to anticipate that MegaRocks, which is going to enroll 40,000 patients in an oxygen trial, will probably have some long-term outcomes data as well. So what are the future directions? So I think there are potentially several. One important one is identifying responder subgroups and trying to characterize individualized treatment effects.

So one study that's come out recently is the HOT COVID trial. It's basically a mimic of the HOT ICU trial that I talked about earlier in my talk. This is a multi-center RCT, and it randomized 726 adults with acute hypoxic respiratory failure due to COVID-19 to receive oxygen therapy with a PAO2 of 60 versus a higher target of 90. The primary outcome here was days alive and without life support at 90 days. Interestingly, this study was terminated early due to slow enrollment. It started enrollment in August 2020, and it finished up in March 2023.

At 90 days, however, the median days alive and without life support was 80 in the lower oxygen target group and 72 days in higher oxygen target group, which was a significant difference. The absolute mortality rates were not different, however, and there was no difference in adverse events of the number of days a patient was alive and out of the hospital. And so while early trial termination limits our ability to really draw definitive conclusions, this study, again, may suggest that inciting illness like viral pneumonia may lead to a differential response to oxygen targets. Another potential avenue to understand whether there are certain patient populations that will respond to oxygen targets is to investigate the impact of individualized treatment effects. So randomized trials report the average effect of treatment on outcomes for an entire trial population. Variation in the magnitude of the direction of the treatment effect or heterogeneity of treatment effect is really not fully addressed by a subgroup analysis because each patient may have multiple characteristics rather than a single diagnosis that influences the intervention on the outcome, the impact of the intervention on the outcome.

But accounting for multiple patient characteristics in a traditional frequentist statistical analysis of heterogeneity treatment effect can be really hard to do. But with new approaches such as machine learning models, we may be able to overcome that limitation. So Kevin Buell at the University of Chicago and colleagues in conjunction with the Pragmatic Critical Care Group here at Vanderbilt tested this idea. They used a machine learning technique, and they derived predicted individualized treatment effects from the pilot cohort using clinical and demographic data. They then categorized them into whether there was a predicted benefit, whether with lower oxygen targets versus intermediate and higher oxygen targets, as shown in panel A.

They then took that model, the machine learning model, and they applied it to a different data set to try to validate it. So they derived these potential individualized treatment effects in the pilot cohort, and they moved it actually to the ICU ROX trial to try to validate this. And indeed, what they found was that the machine learning-derived individual treatment effects were actually predictive of differential mortality between oxygen targets and identified individuals who may benefit from lower oxygen targets versus higher oxygen targets. And this, in my mind, represents really a novel way to get at the question that we all ask as clinicians. Will the patient in front of me respond to the therapy I want to provide based on the data that I have? And I also think there's also a natural extension of such work in identifying patients who may have better long-term outcomes with individualized treatments with different auction targets or other bedside interventions that you can imagine. So I think there's really, this is an exciting area of work that I think Matt Simler and colleagues are going to take forward. And there's real opportunity to build that into the post-intensive care syndrome space.

And so lastly, I'll just make a short pitch that I think there's also a lot of value in trying to understand what are the biological mechanisms that may link things like oxygenation targets or other modifiable practices with long-term outcomes in our critical illness population. To put it another way, can we translate the things that we do at the bedside as intensivist? back to our understanding of how long-term outcomes develop and what are the particular domains and see if there's a way to take the bedside to the bench and the bench back to the bedside and identify molecular or biological markers of predictors of recovery and things that we could potentially intervene upon in the post-intensive care unit and post-intensive care syndrome space. Right now, we don't have good interventions to improve recovery after critical illness, but we're desperately in need of those given the number of patients who suffer from PICs each year.

So I'll make a pitch that I think one of the things that could potentially be playing a role is mitochondrial function. I think my interest in energy metabolism plays into this. We're going to attempt to try to study this in a new cohort study that just started enrolling this month called the Amplify ICU study. So I will just make a pitch that there's more to come in that space and more data to be had. And so with that, I'd like to acknowledge the University of Maryland Critical Care Grand Round Series for inviting me. I really appreciate the opportunity to talk about some of the work that we do. Obviously, I'd like to thank the Vanderbilt Critical Illness Brain Dysfunction Survivorship Center, which I'm a part of. Dr. Wes Ealy is one of my primary mentors and one of my primary scientific advisors, along with Dr. Jen Hahn, who actually was one of the original folks who came up with this idea to study long-term outcomes after the pilot trial.

I'd also like to thank the Vanderbilt Pragmatic Critical Care Research Group, especially Dr. Matthew Simler, who is really, I think, an expert pragmatic critical care trialist and is somebody who I learn from every time that I get the opportunity to talk with him. Then, of course, the NIH, the Department of Veterans Affairs for their funding support for my research, and then for patients and their families, right, because we can't do this type of research and do what we do every day without their consent and without their willingness to participate in research. And with that, I never miss an opportunity to show off my kids, but I'd be happy to take any questions.

Beyond Hope Project with Jason Tharp Beyond Hope Project Welcome to 'Beyond Hope Project,' a beacon of inspiration in the podcast world. Join Jason Tharp as we embark on a transformative journey, exploring stories that awaken the will to start anew and find the strength to rise from setbacks. 'Beyond Hope Project' is about lighting the way to your starting line, not dictating your path. We delve into narratives that foster optimism and belief in a future where dreams become achievements. Each episode is a tapestry of positivity, enthusiasm, and comforting support, resonating with the essence of hope. This podcast goes beyond storytelling; it's a movement of positive change rooted in authenticity and transparency. We provide real tools for real journeys, infusing more hope and joy into the world. 'Beyond Hope Project' stands as a testament to the transformative power of hope and the limitless potential of the human spirit. Join us in transcending boundaries and breaking barriers. Here, we prove dai Veteran Chat Project Kyle Turner Our goal is to reach as many Veterans as we can. We noticed a lot of the podcast personalities tend to be unrelatable to the every day joe. Not taking anything away from anyone just hoping to give more people a place to feel connected to something. KZ and Kyle are Army vets who deployed together for OIF 07-09 and OND. Now close to 15 years later we are finding a way to open up about our issues, and find a way to build a community. Paramedic Philanthropy Project Melissa Ade Podcast by Melissa Ade Bigfoot Information Project Podcast Bigfoot Information Project Occasional conversations with leading researchers and investigators regarding the animal commonly referred to as bigfoot.

Frequently asked questions

How long is this episode of Maryland CC Project?

This episode is 49 minutes long.

When was this Maryland CC Project episode published?

This episode was published on May 31, 2025.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this Maryland CC Project episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.