Ladies and gentlemen, good morning, and thank you for being here at the last session of the Critical Care Congress. My name is Dr. Ryan Maves. I'm a critical care physician in San Diego, California. Today we're going to be doing a pro-con debate about the electronic diagnosis of sepsis, getting it right or overkill. So we are very fortunate to have two fantastic speakers with us today. First up is Dr. Jonathan Savransky from Emory University, who will be telling us about how electronic sepsis alerts improve outcomes. Dr. Savransky? Oh, and one quick comment. We're going to save questions for both speakers after the conclusion of both presentations, so thank you. So good morning, and I very much appreciate you being here so we're not talking to each other. I was charged with arguing that electronic sepsis alerts improve outcomes for patients with sepsis, and the two points that I'd like to make are the first is that when you have a pro-con, it's usually because there's some evidence and the evidence is not clear, that reasonable people can look at the same data and come to different conclusions. And both based on my own experience in my own institution and some biases that I have, I am going to argue for the pro side. But again, when you have a pro-con, it's usually because the evidence is not clear, the signal is not clear. And for some reason, I ended up with my last slide first, so I guess I have a challenge on going backwards, which I don't think has ever happened to me. So I have some intellectual and academic conflicts. They're not directly related to this, but the intellectual conflicts, I do need to mention in that I am a strong believer in the use of standardized care, and again, that's a bias that I have. I use it clinically, and that's going to turn out. I think that's going to inform my discussion here. What I don't think that people will be conflicted about is that sepsis is a major problem for our patients. And we saw some data presented with simultaneous release in critical care medicine earlier just showing how large a problem this is. The incidence is going up, the mortality over the last few years in a Medicare database is high, and there was an estimate, again, in this series of patients, in this series of articles in critical care medicine, suggesting that we spend about $80 billion a year on patients with sepsis. We have a new definition for sepsis. I think everybody's aware of that. And it's better than the old definition in that there's data behind it, but it doesn't really help us much with treating patients. It's a great research tool. And really, unfortunately, since sepsis is a syndrome rather than a disease, it's dependent on large part on clinicians recognizing a pattern. Some of you may recognize the last of the Rorschach blots. If you see a caterpillar or a lobster, you're in the majority. But with patients with sepsis, we need to recognize that this is a patient at risk, and this is a patient who might be infected. And many clinicians, many nurses, many physicians, many pharmacists are good at recognizing it, but it'd be awful better if we did have a biomarker. Since we don't, sometimes we're challenged, especially when the patient's symptoms are atypical. And this is just one study from a couple of years ago where a group at Harvard showed that patients with atypical symptoms, with atypical symptoms, not the usual ones, are harder to recognize as having sepsis, have process measures such as antibiotics given less frequently, and have a worse mortality. So not all of our patients show up and have a big, I'm infected thing stamped on their forehead. This leads to some sepsis patients not getting desired care. And there are a number of reasons for this. We're all busy. Sometimes people think that they're smarter than what a protocol might show, and so they sometimes opt, they recognize a patient as sepsis, but they think, well, that's probably not right for my patients. And that leads sometimes to either failure to recognize patients, failure to implement things that we know help patients with sepsis, and really that requires additional measures. And one way that you could do this is to have an electronic alert to remind clinicians that this patient might have sepsis, and to ensure consideration of desired care. And you notice I didn't say to ensure that everybody gets treatment X, but to ensure that the clinician considers that they might need to get certain types of care. And finally, to remind busy clinicians that you need to treat patients timely. I think, and I think most people would agree with me, that once you identify patients with sepsis, it's not challenging to treat them. You need to give them some antibiotics, get source control, give some supportive therapy of some kind, and prevent doing harm, prevent nosocomial complications. This doesn't require special expertise, but sometimes identifying them does. And just as an evidence that probably we could do really well comes from the usual care group in the meta-analysis of the three largest trials of goal-directed therapy, early goal-directed therapy, process, promise, and arise, in which the usual care group there got antibiotics in about an hour, and had two liters of fluid given prior to randomization. So they were watered, and they were given antibiotics. And most of them did pretty well. And so I'd urge you to think about whether your patients get this type of care. And I have to say that sometimes patients seen in my own institution don't. Sometimes patients seen in my own institution don't reach that antibiotics in an hour and fluids prior to them coming up to the ICU from the emergency room. So if some of my patients aren't getting desired care, what tools do I have to make care better? Well, yelling at people usually doesn't work. It might in your own institution, but I can tell you in mine lecturing them and telling them you have to do this tends not to work well. There is pretty good evidence that standardizing care improves outcomes in sepsis. And this has been shown in the Intermountain system in Utah, where over a five-year period, the proportion of patients that got desired care went from 1 in 20 to 3 in 4 by standardizing the therapy, and mortality was cut in half. It happened in Spain with edrosepsis, where even though most people didn't get desired care, they moved the needle a little, and the post-intervention cohort did better than the pre-intervention cohort. And more recently in New York State with Ryan's regulations, mortality went down after these were introduced. So in most cases, for most patients, standardizing care improves outcomes. But even with standardized care, we don't do as well as we could. Here's some older data out of Project IMPACT, suggesting that it takes a while to give many patients antibiotics. And again, in the New York State data after Ryan's regulations, a number of people did not get their three-hour bundle completed in time. So if we're not doing as well as we could, we ought to be trying hard to do better. I've mentioned before that there are some patients at my own institution, even with standardizing care, even with a lot of us there thinking hard about sepsis, we've known for a number of years that we had some challenges in getting patients desired care. At one point a couple of years ago, we had five separate protocols for sepsis, and none of them talked to each other. Patients didn't always get early antibiotics. Patients always didn't get checks of lactates. So we had plenty of opportunities for improvement. And we did a couple of things that were simple in process, but hard to happen, hard to make happen because we had a lot of smart people who didn't always agree on what we ought to do. And we set a single protocol up that talked to each patient who were either in the emergency room or the floors. So if something was started in the emergency room, the protocol would carry over to the floors. We set up standard sepsis alerts, and the nurses, the bed-sized nurse, were prompted to ask, does this patient have sepsis? And if they thought it likely, they would then call the clinician and say, do you think this patient had sepsis? And if they did, then they'd start the protocol. So this graph has data from the emergency rooms of four of our institutions, and I don't think you need to see which one it is, but on your left, you can see when we started out that many patients were not started on a sepsis protocol. But after a couple of years of having these alerts, again, delivered to the nurses, and then they'd ask the clinician, now routinely in our emergency rooms, and one of them was down at about 1 in 10 patients, we're routinely up to about 90% in the emergency rooms. They're actually better in the emergency rooms than we are in the ICUs. So in our institution, four hospitals, using a sepsis alert, an electronic sepsis alert, teams to help patients. We're still not perfect. We still have room for improvement, but we're getting better. So it's useful, since most of you probably don't work at my institution, I don't see any familiar faces out here, what happens in other institutions. There's not a tremendous amount of data, and much of it actually comes out of the UK. Earlier this year, a group of five hospitals, with 20,000 patients with sepsis, started a tiered program where they slowly introduced electronic prompts to clinicians, first on the wards, then in the emergency room, then in the ICUs. Initially, the prompts were sent out silently, so the system recognized it, but the clinicians didn't get that. And then, with the second part, the clinicians ended up getting the prompts. And with that, and again, 20,000 patients, about 10,000 in each group, the before and after, with the institution of the prompts, mortality went down, and time to antibiotics was faster. 10,000 patients, five different institutions. I mentioned that the data isn't all clear. Well, here's a second British institution, a teaching institution, in which it turned out many of the clinicians, who were mostly residents, did not believe in the therapies that were set up with their sepsis treatment plan, and so many of them didn't give fluids or antibiotics early. It turned out that only about two in three patients who got the prompts had changes in their therapy. But here, it did not improve outcomes. So is this perfect? Probably not. This is a picture of Narcissus, who was supposed to be a perfect person, and he spent a lot of time staring at himself in a pool. This is not a thing of beauty. Sepsis alerts clearly have their challenges. It's not perfect. An ideal alert would have high specificity and high sensitivity. So you'd have few false negatives and few false positives. And I can tell you, we don't have such an ideal electronic alert in our institutions, and I think most places don't. So it's not all good. I think most people would argue that it's not an unmitigated success. But I think, and certainly in our own institution, it's mostly good. The grumbling from clinicians has went down, and the number of patients who get desired care has went up. So one of the things that using an electronic alert, I hope that I've showed you, it can improve the recognition of sepsis. It can improve process measures and just raise awareness of time to therapy. You have to worry about some of the downstream complications, which include alarm or alert fatigue. And every institution has lots of alarms that go off. And you need to be worried that your clinicians might completely turn them out. Initially, our clinicians did. We had to revise our thing a little bit so that it wouldn't fire too often. We made some minor adjustments, and now I think most people are comfortable with it. And one potential complication, and I am unaware of any data on this, although it's certainly a potential problem, is that if you look for sepsis under every bed and you find it more often, it's possible you could give too much antibiotics. And it's possible that you could have complications of antibiotics. Again, it's a theoretical problem. I'm unaware of any data that shows that this does happen, but that's probably because nobody's looked really hard yet in a fashion. And finally, and this to me is the biggest challenge, whenever you tell people you need to concentrate on X, that you need to pay attention to sepsis, that means that maybe they're paying less attention to something else, to early mobility or less attention to stopping sedation, that people's brains can only concentrate on so many performance improvement projects. For us, this is a very important one. We've continued it. It's added value at our institution. But people at the different institutions need to decide which is the most important one for us, for you. So I think certainly at our own institution, sepsis electronic alerts helps patients. And one of the things that's most important is that you don't take one from somebody else and smack it into your own electronic medical record. You've got to adapt it. You've got to talk to your clinicians, find out, make sure that they're not seeing things and having unintended consequences of things that you didn't think of when you set it up. And finally, the thing that's most useful is we feed back data to individual units, to individual clinicians. And so if there's somebody who's not getting it right, we let them know. I can find out if I'm on service for a couple weeks and none of my patients have sepsis protocols or responded appropriately to alerts. I can get that data and say, by goodness, I've got to do better. So with that, I've reached the end of my talk. And at the end, we'll have some questions. And if people want some more information or have violent disagreements with what I've said, please feel free to email me at that address. Thank you very much, Dr. Savransky. I would discourage everyone from violence in this room, however. Our next speaker is Ms. Rhonda D'Agostino, acute care nurse practitioner from Memorial Sloan Kettering in New York. And she will be speaking to us on problems with electronic sepsis diagnosis. Thank you. So it's interesting. And thank you for allowing me to speak today. But it is about perspective. And before we even decided, we didn't care what side we got, whether it was the pro side or the con side. So yeah, I'm here to talk about the con and that it is electronic alerts are overkill. I recently left in physical form Sloan Kettering but remained there in spirit. And so I have to still thank my colleagues who work there, Kate Taban, who is their sepsis clinical program manager, William Zhu, our clinical informatics specialist, and Dr. Neil Halpern, because I still needed all the data to present today. My colleague did a great job about talking about the background, but I'm from New York State. And so New York State in 2013 had implemented mandated protocols. These protocols included all institutions to have an evidence-based protocol. At the time, it was based on the 2012 guidelines. They still have that in effect. It was to report your adherence to these protocols. And this was supposed to be publicly reported. And then they were going to track outcomes of these protocols. This data was going to be collected from our ICD-10 codes. And so as an organization getting this information, we were very nervous to have all of this out in the public and all of our results now evident. So we thought to ourselves, how do we ensure protocol compliance when data is collected off ICD-10 codes? How do we ensure that we're going to get to these patients in time, have an effect on what we're reporting to New York State? And the only answer that we came up with at the time was to have real-time electronic alerts. I'm not here to talk to you about whether or not protocolized systems in sepsis work. We know they have some meaningful impact. I'm not going to break down the research, but I want to show you and break down what the issues are with when we do a mandated reporting structure. I'm going to present to you just one organization and how we dealt with New York State. And I'm going to use 2018. You might ask yourself, why it's 2020? Am I using 2018? Well, New York State is very behind. And so the only data that I have to give you that is risk-adjusted, one is 2017. But the full data cohort is the 2018 data. So in the year 2018, 546 patients or cases with ICD-10 of severe sepsis or septic shock were discharged and reported to our Department of Health. We measure, as my colleague said, the three-hour bundle and the six-hour bundle. And you have a bundle compliance rate. In the three-hour bundle is your initial lactate, your blood cultures, and your antibiotics. I am going to point out that the devil is in the details. If you miss any part of the three-hour bundle, your bundle compliance goes to zero for that case. So even though our bundle compliance was 69, we were above state, we would like a higher percentage. I could tell you also that the reason why we are so low is because our blood culture rate. And it comes down to the details. And as you report and as you do chart abstraction, when you're looking at timestamps of each of these components, it's very difficult to get the timestamp that the blood culture is actually drawn. You don't have a computerized system standing at the bedside. Maybe some do. So what we would do is a septic patient, their blood pressure is low, you're drawing the lactate, you're drawing the blood cultures, and then you're ordering the antibiotic. And the nurse now triages what she's going to do first. So she did the blood. We drew the blood. We drew the blood cultures. And now she's hanging the antibiotics. She's getting everything settled. Those blood cultures were actually sitting at our scanner to be scanned in. They were drawn before the antibiotics, but unfortunately, we were reporting the time that they are scanned in. So that long delay actually resulted in lower. And so when we reported this to the state, you have to be just, it's what we could actually, the data we could abstract from the chart. This is the New York State Department of Health sepsis reporting website. Anybody has access to it, the 2017 risk-adjusted mortality is there. And if you click on it, there's this graph and nobody can read it. You have to really zoom in. So it wasn't as prevalent as we thought. But it shows all the hospitals in New York State with their risk-adjusted mortality. We also got a 15-page report every quarter on how well we did compared to benchmark and went over all the different bundle compliance rates and so on. And so going back to when we decided to implement this, our quality department said, you know, this really belongs with critical care. You're doing a great job with your rapid response team. And so we decided to take this as part of our critical care center. In doing that, we had to hire, we didn't have the resources to do all of this. So we had to hire a programmer to input the data, to build a system in which we could track the compliance, to maintain the program, and to troubleshoot the program. And I'm going to take the perspective of just looking at the cost, right? We're all about a value and looking at, you know, what the value is and over the cost. And so every time you see a dollar sign, it's about, again, we're in New York State, so there's a higher cost of living. And so we're about $100,000 in just getting the initial programmer. We then had to use our CIS system to create an electronic alert. This took a lot of time, a lot of effort, and build in an alert that will then feed into an alert that goes to the provider at the bedside. We had to use middle rail called Conexul so that the alert from the system that we said that we programmed would then go to Conexul to filter the alerts. This again was another $100,000 or so. It would then push the alert over to what we called the Vault application. This application sat on any mobile device. Again, we had a lot of issues with privacy, so we spent money on getting mobile devices for the provider that they can then get the alerts. So our real-time sepsis alerts, we trialed a lot of different tools to get to these patients quicker. And every time that we added another variable, the amount of alerts increased tenfold. And so this is the balance of alert fatigue versus trying to get a high sensitivity and specificity for the patients. And we kind of came up with this formula of two abnormal vital signs plus altered mental status plus raggers or and or raggers would trigger alert, or three abnormal vital signs would trigger an alert. Of course, if any clinicians suspected sepsis, they could manually trigger an alert or call Critical Care Consult. We did a lot of hospital education, a lot of nursing education. Again, all of this was a lot of time and resources. We wanted to create a note that would be able to help us capture all of the data that we had to feed to your Department of Health. The note had to be extremely comprehensive, and we wanted to make sure we were hitting all the bullet points, because when you do a chart review after the fact, it's very hard to pull out the timestamps, like I said, for the blood cultures and so on. And so we created a very specific note that did help us in the end with some compliance issues. Once we got to understanding the system, building the alerts, and testing the alerts, we tested around 1,000 alerts or so per year is what we came up with. And we thought that the sensitivity would be around 72 percent or so, and we were okay with that. But by the time we added a variable and increased the sensitivity and specificity, that 1,000 alerts went to 10,000 alerts, and we couldn't handle the 10,000 alerts. 1,000 alerts is still a lot to handle. And so who's going to now respond to these alerts? And so when we went to the primary teams to say, hey, these alerts, we're going to push them to you, they were too busy. Overnight teams were too overwhelmed. This had to be a provider. I want to say that because it can't be a nurse that responds to these substance alerts. New York State does report that you have to have a reassessment done by a provider. So this had to be some sort of provider. We decided to add this to our rapid response team, APP-led team. They were already overwhelmed with the number of rapid response team and ICU consults, and so we had to add an adjunct and add another team 24-7. So we had one provider 24-7. And in New York State, again, we don't come cheap. This is just to show you how comprehensive the Department of Health web tool that we built. You have to track comorbidities, including chronic respiratory failure, HIV on admission, so that we could do the risk-adjusted. And so this took, for every chart that we reviewed that had an ICD-10 code of severe sepsis or septic shock, we had to do an extensive chart review that took, I wouldn't say, 12 to 15 hours per chart. And again, we are looking back at identification, because identification of these patients is so important. But the problem with the New York state mandate is that the reporting comes from ICD-10 codes. So just a little bit, I'm going to walk you through this graph right here. So here, what I'm pointing out, do we have a pointer? No, maybe not. At the bottom of the screen, we have what was reported to New York state. And at the top of the screen, we have our sepsis alert. And hopefully, we would get them to match. In 2018, we had 1,925 alerts. We did thought it was going to be around 1,000. We quickly realized it was double that. Those alerts, then, we broke them down into unique and non-unique. And then we had about 1,245 uniques. We did get multiple alerts on some patients. And we had ways that we had to shut off the alert if they were, whether they had a DNR, or they were giving treatment, and so on. Of the 1,200 alerts that we got triggered that were on unique patients, only 278 of them actually were reported to New York State Department of Health. And so our sensitivity was much lower than expected. But really, it's about identification, like my counterpart had said. And so even though the number was small, were we identifying these patients in time? Were we getting to that patient when they developed and had the first signs of severe sepsis or septic shock? And we looked at what we call, and there's lots of definitions in the Department of Health, is time zero plus three hours. And unfortunately, only 130 patients triggered alerts during time zero and three hours. The rest were late. They alerted much later, the 148 patients. And of those, plus the 268 patients did not alert. And of those, we now have the 546 patients that reported to Department of Health. We went further and broke down of the alerts that we weren't getting, what was their mortality. And so of the 9% of the alerts that we did not get who had severe sepsis, they completely fell off the radar. And they had a higher mortality, which is not surprising, about 47%. These results are from New York State, and they're not risk adjusted. We're still waiting for the risk adjusted. But our mortality, I can tell you, for the last three years did not change. Our risk adjusted mortality that we computed did not change, and our raw mortality did not change. So I want to go back, and this is where I say that this may be just a bit of overkill. The problem with what we had identified throughout the years with our electronic alerts is that it relies on proper documentation. We saw a high percentage rate, which has been spoken about before, of respiratory rate. And the most common respiratory rate documented was 20. When in fact, the patient's respiratory rate was 40 or 50. And that really led to an issue with our being alerted within the right time. We did tons of education to our nurses. And again, my colleague pointed out that clinicians are, I think, sometimes smarter than the electronic system. And so as we stressed, this is what we're monitoring. This is what we're looking at. Nurses would actually free text documentation, which we couldn't pull, and free text altered mental status. Because they knew if they clicked that altered mental status button, and they had two abnormal vital signs, this would then feed alert. This would trigger order sets and more work. And so there was a lot more free texting. And we realized that late in the game that we could not pull from. And then we looked at alert sensitivity. We're currently exploring machine learning models. And we have, in the background, run some of these machine learning models that are taking a little bit more time. The problem is that any time we implement this, it slows down our computer system where it's inoperable, the entire computer HIS system. So in summary, it was about, give or take, a million dollars plus an exorbitant amount of time to implement the real-time sepsis alerts. We're not sure if we would have the same results, whether or not we did implement the electronic or not. It did, however, help to track those patients that we identified as severe sepsis or septic shock, and follow up with them in a timely fashion. And that's why we're so good with our 98% compliance rate with the New York State follow-up assessment. I don't like to always complain and not have a solution. But I think I'm going to echo the same as my colleague did, and that we should really look at exploring machine learning in real time, have a better integration with our HIS systems. And then if we want to move the New York State regulations to other states, I would say that we need to look at institutions' risk-adjusted mortality and compare that against benchmark. And those that are doing well, let them continue doing well. Don't put the rigor and all the details associated with going through all the different compliance bundles. And then potentially look at places that are not doing well and then say, OK, let's break it down. Let's look at your three-hour bundle, your six-hour bundle. I think now we have time for questions. Thank you. So before we do any questions, I'd like to thank both of our speakers. These are two great and provocative talks. If anyone would like to come up and have any questions, I'd ask you to come up to the microphone. Sir. I'm John Sternegel from Baltimore at Hopkins. And we have a very similar system, early warning system, used through that type of. As I find myself as a day-to-day practitioner in the intensive care unit, the false positives that I see a lot from these early warning systems are largely due to, as you hinted at, inaccurate and imprecise data in the electronic medical record. Chronic kidney disease patients being flagged as acute kidney injury because no one documented their CKD. Chronic liver patients with thrombocytopenia and coagulopathies not being defined in the electronic. How do you fight provider error or omissions in the electronic medical record, which is we're basing these systems on garbage data. So we have a definition problem, and we have a data problem. And those aren't fixed by electronic systems. It makes them worse. Any thoughts? So that's a great question for which I have an imperfect answer. I would reframe the question. And I think you heard a little bit in the last talk. We're not giving patients routinely at a higher level life-saving care. We're not getting them early antibiotics and early supportive therapy and getting them to a place where they can be closely monitored in case they go south. One tool potentially to do this is an electronic alert. And at our institution, it's gotten us closer to that point. But I can't argue that it's a perfect system or that it works in all institutions. So if this doesn't work in your institution, you need to find something else to help more of your patients get the right care. And I wish I could give you a better answer than that. But I think it's important not to let perfect be the enemy of good. And if there's a problem, our biggest challenge with the alerts initially was getting it to the right person. And that's why we decided that we could, with our call schedule, there was no way that we could have it go off that the person who put in the order 30 minutes later might be done with a shift if they were in the emergency room. And that's why we ended up targeting the nurse who we thought would be more likely to know who the clinician that they needed to reach. Again, it's not a perfect solution, but it worked for us. And I wish I could tell you what's going to work for you, but I don't. Maybe my colleague can. I wish I could say that I had a good answer for that. And that's something that I think a lot of organizations still struggle with. We did a little bit more with accountability. So when we had a patient that did an alert or alerted late, we would then go back to those units and say, we can't prove that the respiratory rate was higher, but if you go from 20 and then you're intubated, it's like, OK, let's look at this case a little bit better and see. And so we did a lot of re-education and really stressed the importance of an accurate. And it was respiratory rate that was our issue. And we did see some changes in some of the units. If there's anyone in the audience who has a wonderful solution, I would encourage you to come up to the microphone and share it with your colleagues. My name is Hassan Al-Durzi. So I work in Saudi Arabia. So it's not a question. It's like a couple of comments. So we had a sepsis alert system, electronic, that we implemented in the emergency department, I think 2012, for I think 44 years. And we demonstrated improvement and compliance with the sepsis bundle, as well as improvement in outcomes. And nowadays, we started a trial on an electronic sepsis alert using QSOFA in the wards. The previous study we did was in the ER. And now we started it in the wards. It is a stepped wedge cluster randomized control trial in the context of a quality improvement project. And so we started the trial in the beginning of December of last year. And so we are gradually enrolling wards in five hospitals in Saudi Arabia. And so we expect, and so every two months, we add five wards with the sepsis alert. And eventually, we're going to look at the outcomes of patients. And I know the problems with the electronic alert. We have similar problems with the electronic alert, which we try to work on. But I think this trial may give an answer. And we expect to finish the trial in 20 months. And we will be recruiting, by the end of the 20 months, 45,000 patients. So this is, thank you. And that trial protocol is available on clinicaltrials.gov. I was going to put a slide in that. So thank you for correcting my absence. I think with more data, we will be able to tell better whether this is something that more people ought to try or fewer people. We, what we're trying right now in the background is using some of the QSOFA, but really looking at the changes from patient's baseline and doing, again, looking at just the variables within that change. And we've added a couple more variables. But it's still in trial period. But it'd be interesting to see. I think that's where the future's going to be. Hi, I'm Patrick, Bridgeport Hospital, Connecticut. What changes would you want when it comes to Department of Health from New York? What changes would you want them to make? Everything? No, I would want more. One, I would not want to use, I would want a better feedback system of our benchmarks. We get it two years after, a year after it changes. And so it's very hard for us to implement change on old data. And so that was one of the issues that we struggled with. I also, I think we need to be better at looking at how realistic is it to capture the timestamps? Because, again, there was a lot of areas that we were compliant in. But we couldn't prove that to New York, to the Department of Health. And so I'd want to have some looser criteria. Again, because at the end, we still had a good compliance rate no matter what. But I just didn't like that data being out there from an organization when I knew we were better. So I guess those are to start. Thank you for your presentation. I'm on the fence on both of these things. I think it's pretty easy to capture deteriorating vital signs in the EHR if they're recorded in real time. The problem we have is capturing early onset organ dysfunction. How does an electronic system do this? This is where we really have our issues. So there are lots of different ways to do that. And I am not an expert in predictive analytics. There are a number of people who spoke at this Congress who are. But there are some imperfect ways that you can build in either with lab tests or with vital signs that are specific. And in our own institution, we capture the Glasgow Coma Scale, which the nurses chart. Again, these are all imperfect. And my guess is most of you could poke holes in each of those measures that I've just said. But I think in order to get better, we can't just say these are horrible things. They don't work. Rather, we need to try what we have. And if it doesn't work well, then adapt them as necessary. The two British studies that I showed, each of them used a slightly different mechanism to catch organ failure. And each of them noted their own challenges with capturing organ failure. So I would say that it needs to be institution dependent, again, depending on what your nurses, do they put accurately vital signs into the EMR? And do they routinely measure Glasgow Coma Scale? And if either of those don't work, you may need to think about something else, just sticking with labs or sticking with a biomarker. Yeah. Again, one of our machine models was looking at predictive analytics and identifying at-risk patients and, again, still a lot more to do with that. And it's like the Goldilocks. We just haven't found that this is the perfect solution right now. Pretty nice lectures. I have a thought. Instead of using machine learning or any computer system, could we better define which is the model, SERS, or Quicksilver, or a combination of both? Because when we start discussing, we have basically philosophical questions. What are we discussing? So we're in the position of trying to, I'm going to use, I guess, the analogy that comes to me is that we're rowing a boat with a hole in it and trying to get somewhere. We're doing so poorly. Poorly is probably the wrong word. We have so much room for improvement at many of our institutions, including mine, that we're implementing things that have not been well-validated to be used for what we're trying to use them for. And both SERS and QSOFA don't necessarily work well for identifying patients with sepsis. So could we wait until somebody came up with the perfect prediction algorithm? Sure. But what do you tell to your patient who shows up, who comes up to your ICU, who it took 180 minutes for them to get their first dose of antibiotics in the emergency room? You say, I'm sorry. We don't have any great tools. And next time we'll do better. So again, I don't know which one of those two to pick. And people have, for their systems, chosen each of those models. And I would just say, you have to pick something, try it, and see if it works in your institutions. And this is certainly not evidence-based care. It might not even be expert-based care. But it's locally-based care in most cases. And that's, I think, where we are right now, trying to get to evidence-based care. And we heard of a study that is ongoing that might get us a little closer to evidence-based care. Yeah, there's a possibility of no answers for this question, okay? And I think like, quote, God, when you have problems that you have not solution, it could be clear. But indeed, we are in the middle of nowhere. Thank you, sir. Can I make an additional? Please. So we started preparing for the trial before the sepsis-3 definition came out. And so initially, we were using SERS criteria for the electronic sepsis alert. And then when sepsis-3 definition came, then we were debating what to do. And then so we tested both of them. This is in ward patients. And we found out that QSOFA identified possible sepsis earlier than the SERS criteria. And predicted worse outcome more than the SERS criteria. This was our observation. So when you look at the performance improvement data from Intermountain and from the edu-sepsis data, both of those included consensus recommendations that were later proven to be wrong, right? So both of them wanted tight glucose control and use of active or consideration of use of activated protein C. They were in the things. People implemented them. And by standardizing treatment and making people think about it, on average, more patients got desired care. Which I think is, again, it's reasonably strong evidence that doing something and following up on it, even with imperfect models, is better than saying, these models are awful. I'm not going to do anything. Neil Halpern, Memorial. Excellent presentations. Excellent data, I might say. Thank you. In New York State, when the New York State mandates came out, we were faced with complying with the mandate. And our audience and outcome was Albany. It didn't necessarily have anything to do with patient care. We weren't sure we agreed with what we were being mandated to do. So a lot of our discussion today revolves around improvements in patient care and outcome. And that is very important. But perhaps equally important so that you don't end up on the front page of the New York Post or the Daily News, is to figure out a way to comply with the state mandate. So as Rhonda talked about, so much time was spent around getting the time stamp correct. That had nothing to do with patient care. The blood cultures were drawn. The antimicrobials were administered. Things were done right as to the current evolution of sepsis care. But it wasn't done right as to the New York State system of data collection. So we were left to have to follow a parallel tracks. Meet New York State stuff, a ton of dollars went in as we saw. Try and improve care in the facility, which is what our job is. And at the same time, test an electronic warning system with the understanding that if we really implemented a souped up system with machine learning, AI, whatever we want to call it, it will break the hospital. So there were three tracks that had to be met simultaneously. And I must say, if we would end up on the wrong part of the New York State compliance and end up in the front page of the Post, none of the hospital administrators would care how the patients are doing. They would flip out as to what the hospital is being portrayed in. So I just wanted to highlight there are multiple tracks going on simultaneously. And a facility in New York, of which there are many, has to figure out how to meet them all. I'm not sure that I have anything to add to that. Other than that, while you have a telegenic face, I'm not sure that any of us want to see it above the fold on the New York Times. I'm not sure. Yeah, I just want to echo. And those are some of the things I didn't want to get too involved in looking at, you know, into the New York State regulations. But New York State is still behind. And they are using surge criteria as your timestamp of zero. So that's why we went with the electronic alert. And despite an oncology patient population, surge criteria did not serve us very well. But that was the timestamp that they identified as this is the first signs of sepsis. And so we struggled with that. I don't have an answer, though, how to fix that. Thank you for this presentation. One of the things I've noticed is that we tend to be pretty good at measuring the population and the interventions and the outcomes that we're interested in and that are relatively easy to do. And just in case none of us think that this comes as a free lunch, and I think you've done a good job of showing some ways in which this is not a free lunch, one of the things we're actually particularly bad at I feel is measuring the unintended consequences of some of these things, particularly both in the populations that are not necessarily being screened in or the effort that's being invested in electronic sepsis screens in this conversation are not being invested somewhere else. And so that's harder to measure. And certainly there's some things we can do in like what kind of nonseptic patients, how much extra fluid did they get, how much antibiotic exposure did they get, was there any consequence of that. But there's this entire other population of patients who we didn't spend a million bucks on, right. And so it's not entirely clear whether like every publication which is going to focus on a sepsis intervention, whatever kind, and then an outcome, including in a trial, okay, is devoting those resources not to these other patients. And so it's hard to measure that. It's hard to know like what are we not doing just because we're spending all this time on sepsis. And so I worry that that's an unanswered question for the broader health of all of our patients. So you're absolutely right that there's an opportunity cost of doing anything. I would say that if your ICU cares for medical or surgical or oncology patients, that sepsis is probably your biggest if not close to your biggest problem. If your ICU cares for a lot of cardiothoracic surgical patients or neurosurgical patients, it might be a less important mandate to do. And yet, all ICUs are asked to perform on this. So again, for many of us, this is a big part of the patients that we care for, but certainly not for all critically ill patients, especially those that have a lower prevalence of infections. Yeah, I mean, this goes to also, again, I don't have the answer, but this is why our quality department was like, we're not, I mean, this is, we think we're doing a good job already. We have other issues that we need to address. And so this is why we think that cruel care should do this. And we don't want to use our hospital-wide resources to do something and put all the time in. I mean, they were very helpful and supportive, don't get me wrong, but that was one of the inclinations why it was pushed to the critical care center. All right. Well, I think that'll do it for our session. Thank you all for coming so much, especially at this late session. And thank you particularly to our two speakers for a great presentation and a great discussion from all of you. Have a great trip home, and thanks for coming.

electronic diagnosis

sepsis

electronic sepsis alerts

improve outcomes

early recognition

challenges

limitations

high costs

patient outcomes

Administration

Informatics

Monitoring

Presentation

Select

Professional

Associate

2020

Intermediate

Advanced

Sepsis Resources