SCCM Resource Library-SCCM-Weil Research and Discovery Grant Recipient Presentations

Video Transcription

So, I just want to thank everybody for this award. It's been an absolutely fantastic opportunity,  so thank you to SCCM and thanks for inviting me here today to present and for everyone who's listening to this presentation. I should disclose that I had multiple collaborators  for this work, lots of mentors, lots of co-investigators, study coordinators and site PIs, so thank  you to everyone for helping. This is not a one-person job by any means. And I also want to mention that using the aims from the SCCM award as well as some of the work that we  started, we were able to write another grant for the AHRQ and that was awarded to Dr. Pickering,  so we've built an acute care learning lab, reducing threats of diagnostic fidelity and critical illness. So, that could not have happened without this fundamental and sort  of groundwork with SCCM. So, diagnostic error and delay, it's wrong, missed or delayed diagnoses,  a missed opportunity to make a diagnosis and a failure really to establish an accurate explanation of the patient's health problem or communicate it to the patient or in the  electronic medical record. And that's the National Academy of Medicine definition. Diagnostic error and delay is prevalent and impactful and occurs about 20% of patient  encounters, but it's very poorly understood, particularly in critical illness or acute  illness and much more likely to cause harm than in other settings. So, we had two aims. Initially,  we wanted to prospectively validate an EHR tool to identify diagnostic error and delay in hospitalised patients who had an RRT activation. And our second aim, we used mixed  methods research to identify some of the factors contributing to diagnostic error and delay.  So, our first aim, this is a prospective single centre study. We did this in Rochester. We looked  at all the rapid response team activations that occurred in the month of June. We did a blinded EHR review by two trained physician researchers, and they used a standard operation  procedure to identify, yes, there was a diagnostic error delay, no, or they were uncertain.  This pie chart shows what were the causes of the RRT activation. We can see about 30% secondary to  respiratory failure, hypotension and altered mental status, also common causes for an RRT  activation. This shows the baseline characteristics of our patients. Most of them were in their mid  60s, slightly fewer women than men. And we can see that about 50% then transferred to the ICU  after their RRT. And both reviewers identified within this cohort about a 15% rate of diagnostic  error and delay. Just to dig into those results a little bit more, we had about 285 RRT activations.  Some had to be excluded for various reasons that are listed there. And on initial review,  we had a Kappa score. That's what we used to identify agreement of 0.23, which is quite low.  We then refined the standard operation procedure, did a secondary review and improved our Kappa to  0.65. However, it still remained relatively low. So really, just to skip to the conclusion on this,  it's very challenging to identify diagnostic error and delay in the chart. And this is because a lot of things  are not actually documented in the chart. We don't know what type of communication is happening between  the different teams, between people that are doing perhaps interventions or investigations  and the primary team, between team members on the primary team. So it's hard to actually identify this.  It's also extremely time consuming. And we can see that in about a quarter of cases,  we could not find consensus whether an error or delay had occurred. We presented this work in 2019. Dr. Pinovich presented this work at the Diagnostic Error in Medicine Conference,  and Dr. Sulinani then published it in Applied Clinical Informatics. We've tried to do some other work looking at this approach, including time to diagnostic accuracy,  more time to diagnostic certainty, examining sepsis and pulmonary embolism.  It's remained very, very challenging. So we have put that aside for now and we are exploring other options.  I now want to talk more about AIM2, which is a mixed method approaches to try and understand the contributors of  diagnostic error and delay. Chart review, as we mentioned, is very, very challenging.  So we tried to examine clinicians' perceptions of whether diagnostic error and delay had occurred in rapid response team activations  and unplanned ICU admissions. This is a prospective multi-centre study. We conducted it at two tertiary care hospitals. So that was Jacksonville, Florida and Rochester,  Mayo, and also at Mankatia, which is a community hospital in Minnesota.  We basically tried to survey multi-professional clinicians involved in the care of the patient during those encounters.  This is the flow of the study. So the primary investigators of the research are fellows who identify when RRTs and unplanned ICU admissions occur.  We then communicate with our research coordinators who reached out to bedside clinicians and either emailed them a survey that could be filled in with red cap or paper surveys,  or just interviewed them briefly to gather that data. Essentially, the questions were,  was any new diagnosis made after your encounter? Was there any diagnostic error present before your encounter? Was there a failure to communicate, accurate diagnosis or plan?  And would early involvement of a specialist have helped establishing a diagnosis? These are our results. We can see we had about 1,800 RRTs and about 1,000 unplanned ICU admissions.  There were cases where we were not able to get a survey response from some of the clinicians,  but we did get 522 completed surveys, which is an approximately about a 50% response rate from someone on the team.  In some cases, we did have a couple of responses from the same encounter. And these are just some of our high level findings. If we compare community and academic hospitals,  we see that, as you might expect, academic hospitals have a higher rate of diagnostic error and delay perceived by clinicians.  This may be because patients are more complex. The current score there, we can see at the top, is higher in academic hospitals and the community hospitals.  And there's a higher perception a new diagnosis was made. There was problems with communicating and there was failure to activate subspecialty teams.  We think this is not just maybe related to more complex patients in academic hospitals,  but potentially because people are working in a smaller environment and they feel more loyal to their colleagues  and things like that and they're less likely to perceive that their colleagues have maybe made an error or delay.  Thinking about attendings perceptions of diagnostic error and delay, we'd like to think that attendings are diagnostic experts compared to, for example,  residents or other people on those teams that are not diagnosticians. And again, if we perceive an error has occurred, we see there's a failure to communicate.  We also see there's a failure, perceived failure to activate subspecialty involvement. So those are just some of our findings from the survey.  But again, we're left that there was quite a high perception of diagnostic error in these patients who are having an RRT activation or an unplanned ICU admission.  That manuscript is under review with Journal of Patient Safety. And we've also been doing some work looking at the correlation between our survey responses and our chart review.  And that work is going to be presented at this conference in a poster by Dr. Pinovich.  I'm going to talk more about some other mixed methods approaches that we've used. I just want to come back to diagnostic error and delay.  Often this has been perceived to be related to cognitive errors, cognitive bias on behalf of clinicians, people making mistakes, not asking the right questions.  But as medicine becomes more complex, of course, clinicians are working in environments where they have to interact with a lot of other people and systems and tools and tasks.  So, in fact, these can impact the ability to make an accurate and timely diagnosis. So the SAFER-DX framework and the SEPS-2 framework, which I'm going to show on the next slide,  both integrate the work of clinicians within that sociotechnical work system. And obviously, to capture the complexity of this, we need to do qualitative research.  We can't just kind of do quantifying methodology.  So this is Holden's work system, and we can see here we've got persons, which could be a clinician, and they need to interact with tools and technology.  So that's going to be things like the EMR tasks, that's ordering tests, that's making sure investigations are done and making sure they're being followed up.  There may be organisational issues that are impacting the ability to make a diagnosis. For example, how many people are scheduled to work that day?  What your workload's like? Can that test be done? Or is it, you know, are there no CT scanning times available that day? There's also the internal environment.  So all of these things impact a clinician's ability to make a diagnosis. So our qualitative study, this is what I spent most of my time doing over the last couple of years.  We did it across four sites. That's a pretty large qualitative study. We did it in two tertiary academic hospitals and two community hospitals,  and we did 11 focus groups with 64 participants in total. And we tried to keep our focus groups role and site specific.  We developed our moderator guide for our questions based on Holden's framework and the SaferDx framework. We recorded all our focus groups, transcribed and de-identified them,  and then we coded them in triplicate. So myself and two qualitative researchers went through those and we coded weekly to consensus.  We entered all the data into NVivo and did sort of further analysis and developed themes and sub-themes. So this is the framework that we've built.  As you can see, it's building on the side of the diagnostic journey. So to make an accurate and timely diagnosis, we need to be able to gather information from the charts,  from patients, from families and that kind of thing. We then need to be able to interpret and synthesize the information. So see how that's changed potentially.  And then we need to be able to make decisions and obviously communicate that to patients, families, but also in the chart and to our colleagues.  If we're thinking about what organizational factors are impacting our ability to do this, we can think of things, as we've mentioned before, the EMR. How easy is the EMR to use?  Can we detect trends, changes that are happening to lab results? How are lab results communicated back to us?  Are there protocols and procedures that help or hinder us in that process? What's the infrastructure and capacity of where we're working? Are people overworked?  Can they do those tests, that kind of thing? If we're thinking about interactional factors, this is things like handoffs, how our communication is with our team,  how we communicate with people who need to do investigations our patients need. Clinician factors are the things that we probably know about, implicit bias, cognitive bias,  what type of training background experience a clinician has had. And I'm not going to touch on patient factors today. So these are some of our quotes.  We go dumpster diving to dig out data of our EMR. It's a systems problem. I think the machine's not so well oiled here in the office, it's just too big.  I was like a roadrunner last night. So people are just overworked and they're not able to look at the details about patients  to make an accurate diagnosis or they may miss something. These are problems of handover. I've gotten in trouble where the patient turned out to be sicker than expected.  We've missed hip fractures and stuff like that because people are sent for shortness of breath and nobody mentions they've fallen.  I think we all know the handoff is a high risk of loss of information. And then individual clinician, these may be things like we're influenced by the case that  happened recently or had a big impact. And then this problem of anchoring that I think we all recognize.  We can't shift gears when we get new information and we hang on to some type of a diagnosis. So this is part of an initial exploratory sequential mixed methods approach.  We think the error might be cognitive, but when we talk about diagnostic delays, these are often related to systems and interpersonal and interactional factors.  We've used a survey then that we've distributed in Rochester using these factors. And we did want to distribute it to other sites.  But obviously, with Covid surges, we weren't able to do that everywhere. That manuscript has been accepted in Journal of Patient Safety.  And I also presented this work at the AMIA conference, particularly focusing on the electronic medical record and diagnostic error and delay.  We're using our understanding of the socio-technical framework, that's Holden's framework, to work with the virus registry now to try and understand high performance and low  performance hospitals response to Covid-19. Because again, we recognize it's not just individual clinicians that they're working in  complex systems, which affect their ability to do their job. I just want to thank everybody at SCCM for this award, which has been a great opportunity.  And obviously, I want to thank all my investigators, site PIs, study coordination mentors, everybody who's helped with data analysis and writing manuscripts. Thank you very much.  Good afternoon, and thank you for listening to this digital presentation of the 2020 Discovery Award.  My name is Brian Briscoe, and this is a presentation on the multi-omic profiling in pediatric acute respiratory distress syndrome. So I have no relevant disclosures.  So I'm going to start with the conclusion. So the take home points of what I'd like you to learn today is that the nasal transcriptome can  identify three different subtypes of pediatric ARDF. There's subgroup A, which is the homeostatic group, group B, which is the hyperinflammatory group,  B, which is a hypoinflammatory and prosciliary function group. Group B, which is the inflammatory group over time, will transition to subgroup C and then A,  and that the nasal epithelium is a good surrogate for the transcriptome of the distal conducting airways.  So the transcriptome and the methylome of both the bronchial and nasal epithelium will correlate more closely by subject than by collection site.  The nasal methylome will identify two stable groups that do not change with lung injury status,  and that these genes may represent epigenetic risk loci for the development or protection from ARDS.  Then lastly, the serum immune proteome is highly correlated with the presence or absence of infection, but it doesn't really change with lung injury status.  So we'll compare and contrast that nasal transcriptome later. So the first question to address is why do we think that we can use the nasal epithelium  to understand what's occurring in the distal lung? So the reason for this is that the nasal epithelium is very similar to the epithelial structure of the distal conducting airways.  So in this review by Tata and Hogan, you can see that in the distal lung, which is what we typically think of as what the lung microstructure is composed of,  you have alveolar capillary network, and then you also have a host of epithelial and mesenchymal cells.  So you have alveolar type II cells and alveolar type I cells, which is where the surfactant is made and gas exchange occurs,  and you also have mesenchymal cells, which comprise the structure of the lung, and then you have a host of immune cells. So you have basophils, eosinophils, alveolar macrophages,  and then you have IL-C2 cells and other sort of interstitial macrophages as well. But the other thing to notice is that once you move just beyond this gas exchange surface area  into these really small conducting airways, you have this really bland-looking epithelium, which is composed of principally of two epithelial cell types.  You have club cells and you have ciliated cells, and then below that you have a smooth muscle layer.  And that's important because as you move up in the airways to the larger conducting airways and to the nasal epithelium, what you have is you have some of these same cell types.  So you have ciliated cells and you have club cells, but you begin to add in other cell types.  You have goblet cells and you have basal cells, which is sort of a stem cell compartment for these other cell types,  and then you have submucosal glands, which makes a lot of the mucus in the airway. And then up in the nasal epithelium, it's almost the same.  So you have more goblet cells and fewer ciliated cells and fewer club cells, but the cell types are all present, although the relative distributions are a bit changed.  And so in other adult lung diseases, people have used this to understand distal lung processes. So in a series of papers in the New England Journal by Silvestri and some others,  they found that if you do bronchial and then nasal brushings of patients with suspected lung cancer, so people who have lung nodules that may or may not be cancerous,  you can, with pretty high sensitivity and specificity, determine who has cancer and who doesn't  based on the transcriptomic profile of that epithelium that is nowhere near that lung cancer nodule. And this serves as the scientific basis for a commercial test,  which is now used to treat patients even in the absence of a biopsy-proven lung cancer.  Another group has shown that in COPD, if you do bronchial brushings and compare them to the transcriptome of nasal genes,  in both control and COPD subjects, the genes that are up or down regulated in the bronchial epithelium, the nasal transcriptome correlates with those.  So although you don't get the same array of genes that you do with a bronchial brushing,  the genes that you do get, they match the same expression pattern that you see in the distal lung.  So for this reason, we thought that perhaps we could do the same thing in pediatric ARDS. And so in this study, what we did is we took 25 pediatric ARDS and 23 control subjects.  We did nasal brushings and collected serum on days 1, 3, 7, and 14.  And if patients were undergoing a clinically indicated bronchoscopy, we did concurrent bronchial and nasal brushing.  So from these nasal brushings, we started out doing standard mRNA-seq. We found that we had only about 25% success doing that.  So then we employed a low-input technique, which gave us about 95% conversion of the specimens into usable data. We had 10 sets of nasal and tracheal specimens.  And for all these, we did transcriptomic analysis using extracted RNA.  From many of these ARDS subjects, we also did a methylomic analysis of the DNA using pretty standard methyl-seq pipelines.  And then we also did serum-immune proteome analysis using a SiOCD, which we'll talk about later in the presentation.  So what does this data show? So we'll start with the nasal transcriptomic data.  So when we analyzed this, we found that there were about either 5 or 6 principal components by scree plot.  And so when we did k-means clustering of these, they clustered pretty nicely into 3 different groups. So these 3 groups, we simply termed A, B, and C, red, green, and blue.  And you can see by the clustering plot and also by the principal component plot  that these 3 groups are distinct with A and C having some degree of overlap or being pretty close, but B being distinct from those other 2.  And so what's the differences here? So when we did gene set enrichment analysis on these 60 specimens,  what we found is that when you compare B to group A, there's a significant upregulation of a lot of hyperinflammatory and myeloid immune processes.  And then there's a downregulation of a lot of important epithelial processes related to ciliary function and intracellular transport.  When you compare C to A, it's actually hypoinflammatory. So you have downregulation in a lot of those processes.  And of course, when you compare C to B, these differences are intensified. And in C, you have upregulation of a lot of these ciliary-related processes.  And so when you look at the different cellular components, so what parts of the cells are being affected,  it's pretty much what you would expect based on that information I just told you.  There's upregulation of a lot of secretory inflammatory sort of cell processes such as secretory vesicles.  And then there is, in B, there's downregulation of less ciliary-related processes. And then in C compared to B, it's the inverse.  So how does this compare to the control? Because you said that A was homeostatic, and I didn't really tell you why I thought A was homeostatic.  The reason is that the majority of control specimens clustered with group A. And so we re-ran the entire analysis, including an additional 24 control subjects.  And what this shows you is that A clustered the pediatric area subjects that were classified as A clustered with about 60%.  And then most of the remaining subjects clustered with group C. When we re-ran the analysis, A, B, and C largely remained in their respective groups,  even though there was a little bit of crossover from pediatric ARDS group C into group A.  And so what this is telling us is that the distinction between A and C is there's a little bit of overlap between them,  but B remains distinct, and that none of the control people who don't have acute lung injury, they do not cluster in group A.  And then while some of them might cluster with group C a little bit.  So perhaps you could see this a little bit more clearly when we look at these principal components in two dimensions.  And so what I have here is just the pediatric ARDS specimens with the shapes denoted by severity of lung injury and then the colors for groups A, B, and C.  And so what you can see is that there's a little bit of overlap between groups A and C right there.  If you just look at what happens to the specimens that start out in subgroup A over time, you can see that if a subject starts out in subgroup A, they stay in subgroup A.  This is in contrast to subjects that start in subgroup B. With resolution of their ARDS, they move into C and then A. And so since the time points of these are 1, 3, 7, and 14 days,  there are a few subjects that go straight from B to A and we can't know for sure whether they were in C in the interim. But the pattern seems to be going from B to A.  Subjects that start in subgroup C either stay in subgroup C or they move into subgroup A. So there's never an example of a subject that over time went from C or A into C.  It was all this pattern B to C to A. So originally we had done standard mRNA-seq.  And so we went back to those specimens to see if this was really true, if we could validate our findings.  And so what we found when we looked at standard mRNA-seq is that we had these same clusters A and B and C,  and that when you compare B to A, B had upregulation of a lot of these inflammatory processes. Comparing C to either one of these was difficult because there were only two subjects,  so most of that comparison was significant, but it followed the same general pattern. So what are the differences in these subjects?  So if you just assign subjects by whatever subgroup they were assigned to initially,  the only significant difference between A, B, and C is that more had a genetic syndrome or some sort of developmental delay, which was highly overrepresented in this group.  You know, probably two-thirds of our subjects had some genetic syndrome or developmental delay. What was most important is what was not significantly different.  So subjects' assignment to group A, B, and C didn't correlate with severity of illness, it didn't correlate with immunocompromised state,  and it really didn't correlate with the highest severity of lung injury that they experienced over the course of the study.  However, if you look at specimen-level data, so for any given specimen that was assigned to one of these subgroups, were there any correlations?  You were more likely to have more severe lung injury if that particular specimen was in subgroup B, even though over time you might change.  If you were initially group B, you were more likely to have more severe lung injury.  And there was a trend towards group C having a bit higher, being enriched a bit in direct lung injury.  So then we wanted to compare, does this nasal transcriptome match what's going on in the digital airways? So as I said, we had ten subjects that had bronchoscopy and nasal brushing.  Unfortunately, only three had quality specimens that we could do RNA analysis on.  But what we saw in these three sets of specimens is that when you compare the nasal and bronchial transcriptomes, they were more similar by subject than by site.  You can see that here with this Euclidean distance plot,  that ARDS subjects 1, 2, 1, and 8 were more closely correlated with each other than they were by site, so nasal or bronchial.  And then you see the same thing when you do k-means clustering, how they cluster by subject.  And then you can see it visually in two dimensions when you do the principal component plot.  The subjects are more closely correlated than the sites, which are sort of circles in the triangle.  So to wrap up this part of the talk, we found three subgroups by nasal transcriptomics and pediatric ARDS subjects,  the homeostatic, the inflammatory, and the hypoinflammatory slash increased ciliary function subgroup,  that over time subgroup B will transition to C and then A, and that control subjects tend to be A, which is consistent with this being a more homeostatic subgroup,  and that the nasal transcriptome is indicative of what's happening in the lower conducting airways. And so this story sounds kind of similar to what's been described in sepsis,  that in sepsis you have an initial hyperinflammatory state, which then resolves, and you have a compensatory anti-inflammatory response, which you actually have hypoinflammation,  which then normalizes and you restore homeostasis. So different tissue, different disease, similar story.  So next we're going to talk about epigenetics. So epigenetics is the regulation of gene transcription at the level of altered DNA confirmation.  And so this usually is classified as either DNA methylation or histone acetylation,  and we looked at DNA methylation because this was the most accessible given the specimens that we had.  And so what DNA methylation is, is when you have cytosine residues in the gene, there's complementary guanidine.  And so when they cluster in C-G combinations, that's a C-P, and there are clusters of those together, that's a C-P-G island.  And so these C-P-G islands occur around areas of transcription start sites. And so you'll have a few dozen C-P-Gs that cluster together into this island,  and typically they're either coordinately demethylated or methylated. And so if you have this unmethylated C-P-G island, RNA polymerase has access to that gene,  and it can be transcribed if the transcriptional environment is such that it is turned on.  However, when it's hypermethylated, so when most of these C-P-G sites have methylated cytosine, RNA polymerase does not have access to those, so that gene is not transcribed.  And so what you can do is you can subject the DNA to a chemical conversion where cytosines that are methylated are protected from this bisulfite conversion of cytosines to uracils.  And so you then compare this sequence to the reference genome to understand what percentage of the cytosines in any C-P-G island are methylated.  And so then you can compare two groups and say, you know, what are the differences in the epigenetic state of these two specimens.  And so we were able to sequence a total of 89 specimens, 10 of them were tracheal. And so when we compared the tracheal, actually it's bronchial, not tracheal.  When we compared the bronchial and the nasal methylomes of these subjects, what we see is that the majority of the differentially methylated sites  are in the intergenic region of the genome. And so this is an important analysis. So if differential methylation is important and it's changing transcriptional sites,  you expect that a lot of the differences are going to be in the promoters themselves or in the intronic or exon surrounding the unit.  And you can see only about 5% of the differentially methylated regions are in these areas. And it's only about 7% are in known Cp2 diluents.  And so this is consistent with there not really being a lot of important differential methylation between the bronchial and the nasal specimens.  And that's what you see on the principal component plot here in 3D space. So there's really no pattern to the tracheal, bronchial, or the nasal specimens.  And when you color these specimens by subject, you can see that subjects tend to cluster closely together.  So you have subjects that are more closely correlated by who they came from than by the site from which they are collected.  So this is indicative of the nasal methylome being a good surrogate for the bronchial methylome. And so if you take just the nasal specimens and then you cluster them,  you see that they cluster really nicely into these two different groups. And so you have cluster 1 and cluster 2.  And unlike with the nasal bronchial comparison, when you do this comparison, almost 20% of the sites are in regions that would be expected to be transcriptually important.  So most of them are in the intronic regions with a little bit in the promoter. And 22% are in Cp2 islands or their shores.  And the shores of a Cp2 island are 1000 base pairs on either side of that known Cp2 site.  And so when you look at the principal component clustering of those specimens that were classified as methyl group 1 or 2, you see that they cluster up pretty nicely.  And so you have these two different groups. Methyl group 1 has 10 subjects and 2 has 11 subjects.  And unlike what happened with the transcriptomic data, the group to which the subjects were assigned did not change over time.  So over the course of the 14 days on the four specimens, there was no change in the assignment over time. So a subject in group 1 stayed in 1, 2 stayed in 2.  And when you look at the genes that were differentially methylated, a lot of the gene set enrichment was for processes that you would expect to be important in inflammation.  So integrin binding, binding of either epithelial cells or immune cells to the extracellular matrix. Oxoreductase activity, so making superoxide like neutrophils would do.  And also binding to cell surfaces and binding to extracellular matrix.  The other interesting gene set enrichment was all of the mouse phenotypes of these genes were related to some sort of increased susceptibility.  So perhaps this represents some sort of a stable alteration in the epigenome that predisposes or protects from acute lung injury.  When we compared this to whether this correlated with the transcriptomic subgroups A, B, or C,  there tended to be a little bit more of group C in methyl group 2, but none of it was officially detected.  So to summarize this part of the talk, the DNA methylation doesn't exactly match the transcriptomic data,  but these two patterns of methylation near transcription start sites seem to indicate a predisposition  to some sort of inflammatory process in a subject in the methylomics subgroups. And unlike the transcriptomic, this is stable over time.  Now we don't have samples from before the onset of ARDS, but you would suspect that since it didn't change resolution,  that perhaps this was present before the onset, and that the bronchial and the nasal patterns of methylation are similar.  So we then tried, we moved our attention to some proteomic analysis.  Several groups have shown that you can extract proteins from the fluid that is embedded in heat and moisture exchange filters, so on the exhalatory limb of ventilators.  And when we did this, what you could see is that the fluid extracted from ARDS subject filters had much more solute in it, or something in it, than the control.  However, when we did protein analysis, what we found is that these proteins were severely degraded. So this is a silver stain showing the proteins with molecular weight high to low.  And so it should look like this mouse bronchial lobog fluid with discrete bands that are proteins that migrated at different rates.  In the heat and moisture exchange filters, all we got was 70 kilodalton per hand, which was not albumin. It was the same with the mouse, the AL fluid.  And so what this is telling us is that the proteins from heat and moisture exchange filters,  you can analyze them by techniques such as liquid chromatometry and mass spectrometry, which other groups have done,  but you don't really get any protein, so antibody-based assays are probably not going to work.  So instead what we did is we asked, you know, is there differences in the peripheral blood,  or this immune proteome, of these different pediatric and ARDS subjects? Does that change over time? So we used an antibody-based microarray panel called a SIO-CD,  and then we just did k-means clustering and compared them to the variables that we already had discovered.  And so when we look at this, the immune proteome, it clustered really nicely into two different groups.  So we have cluster one, which is the material, and cluster two, which is the thread.  And so when you compare that to whether there is an infection, so whether there's a viral, bacterial, or both infection going on,  most of these are pneumonia or lung infections. Some of them are sepsis. There is a high degree of correlation between whether you were in cluster two and you had an infection.  This is pretty much what you would expect, right? If you're doing a peripheral blood immunomic assay, you would hope that you could detect the presence or absence of infection.  It was also correlated with the presence or absence of direct lung injury.  And so what this is telling us is that there's probably covariance, since most of these were pulmonary infections.  And then if you'll notice, the subjects clustered very closely together. So here, it doesn't really matter what your lung injury state is.  So if you have ARDS-10 and you have ARDS-1, so the peripheral blood immune proteome doesn't really change with resolution of ARDS.  So it matters what the process is. It doesn't really matter what the lung injury state is.  And then lastly, we compared whether or not these serum immune proteome states correlated with either the nasal or methylomic subgroups, and they did not.  So there was no correlation with either of those epithelial cell subgrouping strategies.  So just comparing cluster one and cluster two, cluster two is TH1-related inflammatory processes.  So a lot of the processes that go along with TH1-mediated inflammation are upregulated in two compared to one.  And then in one, you have upregulation of the TH2 in the inflammatory processes, so interleukin-4 and TH10-associated pathways.  So to wrap up this part of the talk, this pediatric ARDS serum immune proteome, it really correlates with the presence or absence of infection.  We were hoping to be able to see changes with ARDS resolution or be able to drill down a little bit more granularity into the processes going on, but really we couldn't.  It was just sort of indicative of what is going on in the body, whether you had activation of TH1 or TH2 slash inflammatory pathways.  The association with direct and indirect is probably more related to the presence or absence of infection than whether or not it differentiates immune processes related with direct lung injury versus indirect lung injury.  And most importantly, that serum immune proteome doesn't change with time. It's really stable in the subject, regardless of their ARDS status.  So while it might be informative of what's going on, it doesn't really reflect any inflammatory or immune programs that are associated with ARDS resolution or worse.  So the take-home points.  So the nasal transcriptome, it identifies three different pediatric ARDS subtypes, the homeostatic, the inflammatory, and the hypoinflammatory slash prociliary.  And the inflammatory subgroup transitions to the hypoinflammatory and the homeostatic group with resolution of the ARDS.  The nasal epithelium is a good surrogate for what's going on in the distal conducting airways.  So both the bronchial and the nasal transcriptomes and methylones correlate by subject more closely than they do by collection site.  And the methylone identifies two what we think are stable groups of epigenetic states.  And this possibly represents some risk loci that might be important in risk stratification, but it also could be important in unraveling some of the pathologic processes of how the epithelium acts in the development of acute lung injury.  And lastly, the serum proteome is highly correlated with the presence of dexamethasone, but it doesn't really correlate with the resolution of the pediatric ARDS.  So using the peripheral blood to understand ARDS resolution might be less helpful than understanding what's going on in the epithelium. So what does this all mean?  So if we're going to move into targeted therapies for pediatric ARDS and adult ARDS, it's going to require identification of underlying pathobiology in the different areas.  And we're going to have to consider how we can use the respiratory epithelial cells to provide a little bit more granular insight into the pathobiology of ARDS.  Maybe this can sort through some of the noise of what's going on in the peripheral blood because that, of course, is impacted by other processes that are going on in the body.  And then we might be able to identify risk loci. So this could be important in risk stratification, perhaps for ARDS.  So I'd like to thank the Discovery and Critical Care Research Center and the SCCM Research Trust for funding this work.  Cincinnati Children's Hospital Center for Pediatric Genomics also helped to fund it.  And my lab members, especially Rashika Joji-Shih, Garrett Williams, and Mikey Oha were important in doing a lot of the work here.  Our collaborator at Children's Hospital of Philadelphia, Nader Yeha, who tried to mesh together the peripheral blood and nasal transcriptomic endotyping work.  Ronnie Schmidt at Sciomics, who did the proteomic work for us.  And then members of the Critical Care and Perinatal Institute, who provided a lot of learning for myself, but also support in enrolling patients.  And of course, most of all, the patients and their families who participated in this work. So I look forward to hearing from you in the question and answer session.  Good morning, everybody. My name is Nandita Nidig. I am a critical care physician at the Medical University in South Carolina.  I'm going to be talking to you today about inter-ICU transfers. I have no disclosures to note.  So the objectives for today's talk are to review the current state of inter-ICU transfers and regionalization in critical care.  And then moving the focus to families and how they deal specifically with inter-ICU transfers.  So let's talk about inter-ICU transfers. Every year, over 5 million Americans are admitted to an ICU with annual costs exceeding $80 billion.  And where does inter-ICU transfers play into the equation?  The only paper that has looked into this is from a group in Michigan where they looked at Medicare claims and found that 5% of critically ill patients are transferred between hospitals.  But mind you, this is not generalizable. What happens to people who are younger, who are on the ventilator, who have other forms of insurance, or no insurance for that matter?  We don't know. In addition, we have no idea how these transfer patients do clinically, let alone know how much they cost the system.  So this figure here showcases a hub-and-spoke pattern for inter-ICU transfers, with the bigger circles being the hub.  And it's actually quite interesting to see how different the East Coast of the country is compared to the rest of the country.  Let's now talk about regionalization in the context of inter-ICU transfers.  Regionalization is the systematic concentration or transfer of selected patients to centers of excellence.  They're well-known triage and transfer systems for trauma, myocardial infarction, preterm infants, stroke, all having better outcomes in higher volume centers.  Easier, right, to some extent, for certain specialized conditions. But what about critical care? It's so varied. We see so much more.  But one potentially common umbrella could be, in critical care, acute respiratory failure. Why choose acute respiratory failure? Because they're worth half the admissions to an ICU.  And in terms of clinical mortality, these patients can experience mortality of about 60%. On the economic front, critical care costs at about 1% of GDP.  And much of this is associated with being on the ventilator due to acute respiratory failure.  Additionally, these patients with respiratory failure not only experience clinical morbidity, but also psychological impact with higher rates of anxiety, depression, post-traumatic stress.  Almost 10 times that of the U.S. population.  In addition, the volume outcome relationship that we talked about in the previous slide for the other condition holds good for acute respiratory failure, particularly on the ventilator.  And the Sentinel paper about this was published in the New England Journal by Khan and colleagues, favoring high volume centers for acute respiratory failure.  Since then, there has been a lot of debate about regionalization.  The pros for regionalization include access to specialists and expertise, improved outcomes, hence economically more advantageous.  In addition, motivation for performance improvement and being a center of excellence. But on the other hand, you worry about events during transport.  How do you deal with the no-bed situation at the referral center, particularly now during COVID? Delays in treatment while we may miss the golden hour of critical care.  As a community hospital, how do you keep up with your schedule? And the impact on families, the travel burdens imposed, how do we deal with all of that?  And what do we as clinicians and physicians think about regionalization?  To address this question, a group of researchers at UPMC performed an observational study that involved administering semi-structured surveys to critical care specialists across the country.  They surveyed on three domains, the general beliefs, attitudes, barriers, and solution to these barriers as it relates to regionalization. What did they find?  They found that clinicians had mixed views about the survival benefit and the healthcare efficacy of regionalization.  However, one particular group, the academic physicians, favored regionalization more so than community physicians.  What struck me the most about the study was that 70% of physicians surveyed believed that transfers would cause a significant strain on patients and families. Let's step back.  What is it like to be a family member of an ICU patient? Intense? Stressful, to say the least?  And it's not just the acute phase, but the implications are much long-lived, and hence a recognized syndrome now called post-intensive care syndrome family.  These family members have rates of post-traumatic stress as high as 30 to 80%. Rates of anxiety and depression as high as 50%.  What is even more scary is that one year after the event, up to 40% of these family members present with criteria for psychiatric illness, such as generalized anxiety and major depressive disorder.  And to all of that, add the variable of inter-ICU transfer. And this is a slide illustrating all of the burdens associated with transfer. How did they get here? Where do they eat?  Where do they stay? Do they comprehend what we're saying? What do they do with their kids, their own jobs?  And unfortunately, it's not something that many of us think about while practicing in the ICU, because the focus is often the patient and not so much the family.  But let's change that. Let's try to understand experiences of families with inter-ICU transfer.  In order to do that, our team at MUSC designed a qualitative study and utilized semi-structured interview guide among three stakeholders involved in transfer.  The transferring clinician, the receiving clinician, and the family member undergoing transfer with the patient.  We utilized content analysis techniques to identify themes in these interviews. An outcome of interest, of course, was inter-ICU transfer experiences. What did we find?  What we found was that patient, family, clinician, and system-level factors were key influences with inter-ICU transfer.  And this is just a sample of the most provocative quotes that I have here. As a family member, the doctor said that this is where we need to be. And I said, let's go.  Let's transfer. As a transferring physician, acknowledgement that we're good at transporting patients but not so much families.  As a receiving physician, again, acknowledgement and embarrassment that we haven't thought about this before or they hadn't thought about this before.  And also a feeling that patients and families have no clue why they get transferred from one hospital to another. Like I said, this is just a sample of the codes.  But collating all of these codes are key themes and findings in the study was that there was lack of engagement of family in the decision to transfer.  Two, lack of standardized communication between the transferring and the receiving clinicians during care transitions.  And finally, irrespective of all of this, family members were reassured with quality of care and expertise during transfer.  And in spite of having multiple practical and financial challenges, they were reassured with the quality of care at the tertiary-level center.  In fact, in this particular study, clinicians also uniformly acknowledge the lack of a systematic approach for engaging and meeting the needs of families whose loved ones undergo transfer and suggested a number of resources.  Some innovative video meetings prior to transfer. Some, you know, a lot more financially intensive and not, you know, maybe practical on a day-to-day setting.  What was interesting, though, was families rarely asked for anything for themselves. It was all about the patient and mostly communication and seldom anything for themselves.  I think this slide is also important because there are potential intervention targets here that can be tested in future studies.  So the last few slides discussed experiences of families with inter-ICU transfer. Granted, you know, experiences are important.  But as physicians and healthcare professionals, we're trained to think about healthcare outcomes rather than just experiences.  So next question, more important question, I should say, is what is the psychological impact on families due to inter-ICU transfer?  The design we used to investigate was a single center, cross-sectional study, prospective study, survey-based study here at the Medical University of South Carolina.  We surveyed 82 family members of patients with acute respiratory failure, of which 39 had undergone transfer from a community hospital and 43 were directly admitted from the emergency room to our ICUs.  The survey was to be completed by the family members during their stay in the ICU.  The outcomes we were interested in were anxiety and depression measured by the hospital anxiety and depression scale and post-traumatic stress measured by the post-traumatic stress scale 10, both well-validated scales in the ICU setting.  Our hypothesis was that families of transferred patients would have worse psychological distress and measures. Moving on to results. Here's our Table 1.  As it relates to patient demographics and illness severity characteristics in Column 1, Column 2 and 3 are the cohorts of interest, directly admitted patients, and the transferred patients.  Of note, some differences between the two cohorts. Over 60% of transferred patients were admitted to the new ICU.  Transferred patients had longer length of stay, hospital stay, and also longer ventilated days.  Additionally, transferred patients here were more likely to be discharged to rehab or an LTAC facility.  To summarize this table, one could say that although the two cohorts, transfers and direct admits, seem similar based on the Apache score and Charleston score, which are acute and chronic illness severity measures,  transferred patients were still inherently different in terms of where they get admitted, how long they're admitted, how long they're on the ventilator, and where they're discharged to.  Then we looked at family demographics, a host of family demographic measures between the two cohorts, age, race, religion, education, employment, prior history of psychological disorders.  But the only notable difference between the two cohorts was in the distance traveled. The distance traveled for the transferred group of family members was much higher.  And we also have here an unadjusted outcome of interest. Anxiety and depression and post-traumatic stress scores were higher in the transferred family cohort.  And as background for folks who are not well versed with the HADS questionnaire, it ranges from 0 to 42, and higher scores reflect higher distress.  Similarly, for PTSS, the score range is between 0 to 70, and higher scores reflect greater distress.  Finally, we ran regression models of outcomes. So to walk you through this table, the HADS and the PTSS outcomes.  In the first fully adjusted model of anxiety and depression measured by HADS, we controlled for patient APAGI score and length of stay,  and found that family members of transferred patients on average had a 1.7 higher HADS score than those of directly admitted family members.  This did not reach statistical significance.  While in the adjusted model for PTSS, controlling for APAGI score, patient female sex, and length of stay,  we found that family members of transferred patients had a 5.1 higher post-traumatic stress scale score, which was in comparison to directly admitted family members.  And this was significant with a P of 0.03.  Thus, our main finding from this study is that transferred families have worse scores on psychological questionnaires in comparison to directly admitted families.  However, there are limitations. Higher scores on questionnaires do not equate to actual clinical diagnosis,  and our measurements were only at one point of time,  and therefore do not account for the dynamic and the evolving nature of these outcomes in family members who have loved ones in the ICU.  We also found that transferred patients were inherently different with higher days on the bend, higher length of stay, more likely to be discharged to rehab,  which could all be responsible for the adverse psychological outcomes in families.  Finally, this was an exploratory study, and we did not investigate cause for directionality or any specific intervention strategies.  And particularly for me as a clinician, the biggest take-home point is that families that undergo transfer are an at-risk population.  And future studies should focus on resource provision in the form of, let's say, a care navigator.  As well as, you know, we as clinicians can take a look at our communication strategies that we employ when we're talking to transferred families a little bit more carefully.  There's much more to do, but this study highlights the need to heighten our family assessment, the setting of transfer to truly provide patient and family-centered care.  Coming to our summary slide, we know inter-ICU transfers happen on a day-to-day basis.  Yet, there is no organized structure for it, and we still have many questions to answer and weigh the benefits and the burdens.  We've also learned that there's lack of engagement of family members and resources during inter-ICU transfer.  And finally, transfer families report higher scores on psychological questionnaires, thus bringing to the forefront the need for future longitudinal studies  and to focus on evaluating needed resources and interventions to address psychological well-being in family members.  My research is supported by my mentoring team and collaborators listed here.  I am truly indebted to all of them, and I've been fortunate to have support from the above funding agencies.  And, you know, finally, this is the Cooper River Bridge in Charleston. Come visit when the pandemic is over.  I hope my research can contribute and bridge the knowledge gap with an ultimate goal of allowing clinicians, patients, and families to make more informed decisions about transfer  and help streamline future transfer networks by improving outcomes and family experiences. Thank you.  ♪

Video Summary

Inter-ICU transfers refer to the movement of critically ill patients between different intensive care units. The exact prevalence of such transfers is not well known, but it is estimated that around 5% of critically ill patients are transferred between hospitals. Regionalization is a strategy that involves concentrating patients at specialized centers to improve outcomes and access to expertise. While regionalization has been successful in other areas of medicine, such as trauma care and stroke treatment, its effectiveness in critical care is still debated. One specific focus within critical care that could benefit from regionalization is acute respiratory failure. Patients with acute respiratory failure account for a significant portion of ICU admissions and often have poor outcomes. Studies have shown that high-volume centers have better outcomes for patients with acute respiratory failure, suggesting that regionalization could improve care for these patients. However, there are challenges with regionalization, such as potential delays in treatment during transfers and limited capacity at referral centers. Additionally, the impact of inter-ICU transfers on families is often overlooked. The process can cause significant stress and financial burden, and families are often not involved in the decision-making process. Studies have found that families of transferred patients experience higher levels of anxiety, depression, and post-traumatic stress. There is a need for improved communication and support for families during inter-ICU transfers. Ultimately, more research is needed to better understand the prevalence and impact of inter-ICU transfers, as well as the effectiveness of regionalization strategies. The goal is to improve outcomes for critically ill patients while also addressing the needs of their families.

Asset Subtitle

Administration, Quality and Patient Safety, 2021

Asset Caption

The Society of Critical Care Medicine's Critical Care Congress features internationally renowned faculty and content sessions highlighting the most up-to-date, evidence-based developments in critical care medicine. This is a presentation from the 2021 Critical Care Congress held virtually from January 31-February 12, 2021.

Learning Objectives

To review current understanding of etiology of anemia in the ICU and to identify future treatment strategies for ICU anemia

SCCM-Discovery Grant Receipient: Nested Observational Cohort to Evaluate Diagnostic Error and Delay in the Critically Ill
Amelia Barwise

SCCM-Discovery Grant Recipient: Multi-Omic Approach to Identifying PARDS Endotypes
Brian Varisco

SCCM-Weil Research Grant Recipient: The Psychological Impact of Inter-ICU Transfers
Nadita Nadig

Meta Tag

Content Type Presentation

Knowledge Area Administration

Knowledge Area Quality and Patient Safety

Knowledge Level Foundational

Knowledge Level Intermediate

Knowledge Level Advanced

Membership Level Select

Tag Informatics

Tag Medical Legal Issues

Tag Transfers

Year 2021

Keywords

Inter-ICU transfers

critically ill patients

intensive care units

prevalence

regionalization

specialized centers

outcomes

expertise

acute respiratory failure

high-volume centers

	Society of Critical Care Medicine 500 Midway Drive Mount Prospect, IL 60056 USA Phone: +1 847 827-6888 Fax: +1 847 439-7226 Email: support@sccm.org	Contact Us About SCCM Newsroom Advertising & Sponsorship DONATE	MySCCM LearnICU Patients & Families Surviving Sepsis Campaign Critical Care Societies Collaborative	GET OUR NEWSLETTER
© Society of Critical Care Medicine. All rights reserved. \| Privacy Statement \| Terms & Conditions The Society of Critical Care Medicine, SCCM, and Critical Care Congress are registered trademarks of the Society of Critical Care Medicine.