Thank you for staying. We're gonna talk about identification of molecular endotypes in two cohorts of pediatric ARDS, which in retrospect, possibly should have come before Dr. Balcarcel's talk, but we're all here now. Oh, snap, this is, there we go. These are my disclosures. So heterogeneity has plagued ARDS trials, like all critical illness syndrome trials. So like after 50 years, this is what we know we should do. Don't use 12 per kilo with plateau pressures up to 50, just don't do it, no matter how much you wanna do it or your boss tells you to do it, they should have retired and you should not do it. Also, maybe, if your PF ratio is less than 150 for half to one day, you should prone, in adults, probably, because an effect size of 50% is potentially implausible. But in 50 years, this is how far we've gotten. And endotyping has really taken off in the last, like Dan mentioned some of this stuff, and a lot of this was like, really like, we owe a lot to Dr. Calfee and re-analysis of the ARDS network trials, as well as like, some European cohorts and some European trials as well, which has generally shown a consistent hypo and hyper-inflammatory sub-phenotype. And the initial analyses, like as Dr. Flory mentioned earlier, like was a combination of clinical variables, routinely collected labs, routinely collected vital signs, albeit clinical trial level data, not continuous EHR data, but like whatever was recorded and the exposure misclassification that comes along with that. Whereas biomarkers tend to be fairly reliably measured because you know, you have an ELISA, you can measure them accurately, report them accurately. So the precision of that versus the imprecision of some of the clinical variables has led to these sub-phenotypes, which are the hyper-inflammatory being defined at higher levels of IL-6, IL-8, soluble TNFR-1, and lower levels of blood pressure, protein C and bicarbonate. So, this has been shown. Like in adults, this is fairly well established. This is like a review article, which was looking at like the hyper and hypo-inflammatory sub-phenotypes in multiple ARDS trials. And this tiny little mortality right in the back, like is pediatrics and like re-analyzing the restore data, which had a subset of patients which had blood collected and were able to perform sub-phenotyping. And indeed, like we found the exact same thing. These are two pediatric examples of sub-phenotyping that have made it into the literature. So, Matt Zinter and Anil Sapru's group like demonstrated that matrix metalloprotein profiles could also differentiate two subclasses of ARDS with differential mortality, according to their MMP profile. And on the right is the same sub-phenotyping strategy, which like the adults had done, performed in the restore cohort. So now, an important consideration here is that restore is not an ARDS trial. Restore was a respiratory failure trial. And the ARDS assignment was made on the backend. So, there's some questions as to like how generalizable this is. And in general, like, you know, if my initial spiel didn't convince you, we should probably replicate some of the stuff the way the adults have. And we can't always just be this like, you know, sad little corner in the back with just one representative pediatric study. So, what we aimed to do was to identify biomarker-based sub-phenotypes in two separate cohorts of pediatric ARDS. So, the cohorts were CHOP, which had 333 children with Berlin-defined ARDS, and the UCSF cohort, which was driven out of UCSF, but represents five institutions as a multicenter study, of 293 children with the old 1992 AECC-defined ARDS. And we separately, CHOP performed analysis, UCSF performed analysis, we removed collinear biomarkers within our own cohorts, and we performed the latent class analysis separately in both cohorts. And then we just compared clinical characteristics. And these are the two sub-phenotypes. CHOP's on the left, UCSF is on the right. And what you can see is that like UCSF one measured a lot more biomarkers. So, there's more stuff on the x-axis on the top graph. CHOP measured fewer, and there's fewer things on that axis. Nevertheless, hyper and hyper-inflammatory sub-phenotypes were readily identified with very, very comparable worst mortality in the hyper-inflammatory. In both groups, a two-class model was chosen as the best fit. All the same players are implicated. Okay, because the UCSF cohort measured many more biomarkers, we actually find some novel players, which are probably collinear with some of the other players which have been implicated before. But MMP is implicated, and 2.2 and TNFR1 were the major drivers in this particular sub-phenotype, in this particular latent class analysis. Factor VII, alpha-2-macroglobulin, and TCSP, which I will admit to you readily, I've forgotten what it stood for, made up the underrepresented in the hyper-inflammatory sub-phenotype. So, the adult algorithms could predict the sub-phenotypes that we found with reasonably good ROC curves of areas under the receiver operating characteristic curve of 0.93 and above, and reasonably good accuracy of well over 80%. More intriguingly, in the CHOP cohort, we actually had reasonably good medication data, and it was a large enough study that we were able to perform adjusted and unadjusted analysis to test for heterogeneity or treatment effect of non-protocolized steroid use, particularly specifically methylprednisolone use, in which we found both an adjusted and unadjusted analysis that in the hypo-inflammatory phenotype, there may have been a protective effect on mortality, whereas in the hyper-inflammatory, it may have been harmful. Because it was observational data, and a lot of people yelled at us about this being observational data, and why are you trying to do what the trials do when you're not a trial dude, then we also performed a Provence score analysis in which we repeated the analysis after inverse probability of treatment weighting on the Provence score and found qualitatively the same results, where in the whole cohort, there's no effect of methylprednisolone on mortality, but in the hypo, it's protective, in the hyper, it's harmful, at least the point estimates, although the harmful point estimate touches one, and the interaction p-value is significant. So overall, we confirmed hypo and hyper-inflammatory in two separate pediatric cohorts, which show clear overlap with the adults, and with Dr. Dahmer and Dr. Flory's study. Hyper-inflammatory ARDS has elevated biomarkers of innate immunity and endotheliopathy, and this has some potential for prognostic enrichment, definitely. I don't think any of us question that the hyper-inflammatory does worse. I don't think any of us are that surprised by it either, but we're not gonna question that. And potentially predictive. The differential efficacy of steroid, according to this, is I can see that there's a lot of limitations with that being an observational cohort, and certainly I welcome any questions questioning that afterward, but the potential for predictive enrichment I think is proven here. And with that, I'll take questions. And these are my collaborators. And that's my daughter, who clearly hates what she's doing right now. And I love that she hates it so much that I put this picture up every chance I can.

molecular subtypes

pediatric acute respiratory distress syndrome

biomarkers

hyper-inflammatory subtype

hypo-inflammatory subtype