Thank you for having me here today. I'm Cody Gathers. I'm one of the third year IC fellows at CHOP. I completed this study as a co-first author with Dr. Jessica Barreto, who is a cardiology fellow at Boston Children's Hospital. I'm excited to talk to you about this study today. I have no financial disclosures to report. So by the end of this talk, we should all be able to talk about how we assess multidimensional levels of social determinants of health and how they may influence cardiac arrest treatments and outcomes. We'll classify these factors and their demographics into various neighborhood phenotypes or classifications, and we'll then use those classifications to assess the association of those phenotypes with cardiac arrest treatments and outcomes. As a brief background, we know that racial and socioeconomic disparities exist in cardiac arrests, and they may be mediated in part by pre-hospital and in-hospital treatments. Dr. Haskell has demonstrated that there are racial disparities in patients who achieve ROSC after in-hospital cardiac arrest, and Dr. Maryam Naeem at CHOP actually has shown that there are racial disparities in survival with a favorable neurologic outcome for patients without a hospital cardiac arrest, and that disparity may in part be mediated by who receives Bison or CPR after cardiac arrest. However, few have actually examined sort of granular level data or social determinants of health that may influence health outcomes in cardiac arrest. As such, we aim to actually identify neighborhood level social determinants of health phenotypes and examine their associations with treatments and outcomes. These phenotypes are sort of classifications that are fit through a latent class analysis and best fit probabilities form these subgroups, and we'll talk about that in a little bit. So we did a secondary analysis of personalizing outcomes after child cardiac arrest study. It's an observational multi-center study comprised of 14 centers. It collects serum biomarker data, neuroimaging, and EEG with one-year outcomes. We assess some of the demographic variables in that study, and that study includes all children less than 18 years old who are admitted to an ICU either with an in-hospital or out-of-hospital cardiac arrest. So POCA collects a lot of demographic variables. There were over 300 variables to analyze in this study. Over 300 US census tract variables, which are all derived from the patient zip code. Some of these variables obviously include race, ethnicity, education level, but some of them are more granular as well, such as how many parks are in your neighborhood, distance to those parks, other factors that may be associated with where you live, and trying to tease out whether or not that affects health outcomes. And within our outcomes, we looked at pre-hospital treatments and treatments that you got within the hospital as well, including bolus epinephrine, defibrillations, neuromonitoring such as EEG and MRI, and ECMO as well. Our outcome as in conjunction with POCA as well was survival with a Vineland adaptive behavioral skill score less than 70 or death by one year. And a VAB score less than 70 is indicative of significant disability. So we performed a latent class analysis to identify three neighborhood social determinants of health phenotypes. Remember, these are sort of classifications based off of your many exposure variables that we're getting from our database. We then conducted univariate logistic regression to analyze the relationship between these phenotypes with our treatments and our outcomes as well. So I just want to take a step back as a brief aside and talk about latent class analysis. This was something that was new to me whenever I was talking with everyone about this study. Latent class analysis is a methodology that uses a large number of covariates to sort of estimate best fit probabilities in the model. And from those covariates, those best fit probabilities sort of classify patients into groups. These groups, just for the simplicity of this study, we call neighborhoods. So through the latent class analysis, we identified three separate neighborhoods. And just so you know, latent class analysis has been used in pediatric ARDS as well to identify different phenotypes of ARDS based off of covariates. So we were hoping to use this as a model as a way to classify exposures. So when you look at some of our results, we have three separate neighborhood phenotypes that we assess. And we define the phenotypes based off of the most defining characteristic within each group. So we'll call this sort of neighborhood one or the high unemployment neighborhood. It also had the lowest education level and more white residents over 65 years old. And then the high poverty neighborhood had more black residents, more single mothers, and less access to vehicles and parks as well. And then the higher education neighborhood had more racial diversity, younger children, higher education, and more parks. There's a variety of factors that also stand out, but these are the highlights. So when we look at our treatments, I wanna orient you to this graph pretty quickly. So the first line is our survival with significant disability or death by one year. And you see that actually in the higher education neighborhood, there's a higher incidence of disability or death in the study. And then I group them into resuscitative treatments. When you look at ECMO, you see a significant difference in the higher education neighborhoods as well. And then in neuromonitoring, there's a higher incidence of CT scans in the high unemployment neighborhood. It's important to note that these differences remain significant even after adjusting for illness severity or PIM-3 scores. So when you really think about it, there are patients who have access to more resources who are getting more aggressive treatment despite adjustments for severity of illness. With regards to the outcome measure of survival with the BAB score, it's important to take that with a grain of salt. We haven't really fully done a multivariable logistic regression on a lot of this yet, and this is still very preliminary data. In addition to that, we had over 120 patients in the study who were either withdrew or were lost to follow-up. So I think that distorts some of the data a bit. So in conclusion, we identified three distinct neighborhood social determinative health phenotypes. Neighborhoods with higher unemployment rates had more utilization of brain imaging. Higher education levels had higher ECMO or death by one year, but that is to be taken with a little bit of grain of salt for now. We're going to get some final data in the next couple of months. And then importantly, there were no differences in treatments, resuscitative treatments, even after adjusting for illness severity. The study is not without limitations, but it does present a novel approach to causal inference, especially when we analyze the effects of structural racism in health disparities. It's a smaller sample size. It's mainly academic centers. We had 120 patients lost to follow-up, and it does not account for mortality actually prior to ICU admission. There are many implications for this. I think we're starting to move beyond the barriers of sort of our traditional measures of causal inference and racial disparities. And instead of looking at the association of race with an outcome, we're trying to measure more granular data and see what are the factors associated with one's race in society at the neighborhood level that may impact your health outcomes. It's important to note that it's actually, it's racism, not one's race that leads to disparate health outcomes. I want to acknowledge the entire authorship team, especially Erica Fink and Tony Fabio at Pittsburgh, who have been the senior mentors on the study and the entire POCA collaboration network. Thank you. I'm happy to answer any questions.

social determinants of health

cardiac arrest treatments

neighborhood-level social determinants

latent class analysis

structural racism