The Answers You Wanted On Sepsis – Episode 3

Welcome. Welcome to episode three of the Michael Rothman, a new podcast about health, about life, about saving lives, about other things. And let me start with a disclaimer. The opinions expressed in this podcast are of mine are mine alone. They may or may not be the opinions of those that work for my company, is Space Labs Healthcare. Welcome to episode three. So let's continue the story of sepsis, An important topic, one which occupies the minds of many in the healthcare industry and especially in quality functions and what do we do about it, how do we understand it. In the first two episodes of the Michael Rothman, I talked about sepsis, what it is, what it isn't, how hard it is to get your hands around it, and the miss misunderstanding that many people have about sepsis. And I ended episode two by asking the question, is there a way to reduce sepsis mortality? Is there something that can be done? I said that the sepsis functions at a lot of hospitals are really misdirecting their energy, and is there something that can be done differently? And I gave an answer. I said the answer is yes. Yes. You can reduce sepsis mortality. Okay. Well, how? How do you do it? Well, in order to answer that question, I wanna take a step back and look at some data. We did a study using data from a large hospital in the northeast and we looked at patients who were admitted. I think we had a year's worth of data and some of those patients had been coded at admission that they were septic. And if you may recall from the last episode, one of the things I noted was that about eighty five percent of patients who have sepsis in the hospital arrive at the hospital with sepsis. Eighty five percent. So for right now, we're gonna focus on that eighty five percent. They arrive at the hospital and they're identified as being septic at admission. Well, how do you characterize those patients? One way is to quantify the acuity level of those patients. Now some of you may be aware of a patient acuity score called the Rothman Index, which was created by my brother Steven and I a number of years ago after the death of our mother. And you can read about it in the literature. I'm not gonna go into depth in terms of what it is. Let's just take it as a given that it is a patient acuity score. And let me calibrate you as to what that score means. A score of one hundred means that the patient is unimpaired. It doesn't mean you're gonna live for a long time. It just means you're not impaired. As the score falls, acuity increases. Think of it as a grade on a test. You wanna get a hundred. If your score falls to sixty five, that's the level of acuity that you would see on a patient who is discharged from the hospital to a skilled nursing facility. So sixty five is significantly ill, not someone who is gonna be comfortable at home, they would need extra care. Once the score falls to forty, that's the point at which a physician would consider admitting the patient to the ICU. For those of you who are familiar with the modified early warning system or MUSE score, forty corresponds roughly to a MUSE score of four. As the score falls lower the patient is of increasing acuity, increasing concern. Once you get down to zero there's about a fifty percent chance that that patient is going to die in the hospital. The score does go negative and you sometimes see negative scores of patients in the ICU. Why does it go negative? Why doesn't it go from one hundred to zero? Well that's another topic and we'll talk about that in another episode but let's just take it for now that this is the calibration of the Rothman Index. A hundred unimpaired, sixty five SNF, forty ICU, zero, fifty percent chance of dying. And there will be a citation showing you details of the Rothman Index but let's continue our discussion about sepsis. Alright, so where we are is trying to characterize those patients that arrive at the hospital with sepsis or without sepsis? How do you characterize them? Well, you can calculate this acuity score and get a distribution of acuity scores. And we did just that and we divided the population in two, those patients who ended up dying and those patients who didn't die. And then for the patients who didn't die, we divided that into two groups. Those who were identified as septic and those who were identified or not identified as septic. Okay. And then we look at the distribution of acuity scores at admission for those who are going to live and we compare those who are gonna live who are septic with those who are gonna live who are not septic. And as you can see in this histogram, there is a difference. Those patients who are identified as septic who are going to live are significantly more acute than those who are not identified as septic who are gonna live. So it is clear that that group who is identified as septic who are gonna live looks different than those who are not identified as septic who are gonna live. They are sicker. They are more acute. Not surprising. But let's look at the other side of the coin here. Now we're looking at patients who are going to die. And we look at the acuity level of those patients who are gonna die who are septic and the acuity level of those patients who are gonna die who are not septic. And there should be another histogram which pops up in this podcast. You'll see it right here. What do you see? You see there's no difference in the distribution of acuities for those patients who are gonna die whether they're septic or not. They look the same. The overwhelming piece of information is these are patients who are going to die in the hospital and whether or not they happen to be flagged as septic doesn't seem to make any difference. Okay, we're gonna take another crack at trying to look at the difference between septic patients and non septic patients. Non septic patients who are going to live, and so let's just take a cut point. Let's say what fraction of them have RI scores of less than sixty five? So, you know, significant level of acuity, not deathly ill or anything like that, but less than sixty five is the cut point that we're going to talk about. What fraction of the non septic patients who are going to live show up with scores less than sixty five? Well, the answer is twenty percent. Okay. How about those who are septic who are gonna live? Fifty five percent. This just goes along with that histogram that we saw before, but now we're quantifying it by just taking a specific cut point. And so you can see, if you're looking at patients who are going to live, those who are identified as septic are much more acute than those who are identified as not septic. Okay. But our focus really is on sepsis mortality. So let's look at those patients who are going to die. For the nonseptic patients who are gonna die, eighty six percent of them were admitted with scores less than sixty five. For the septic patients who are gonna die, eighty nine percent of them were admitted with scores less than sixty five. So didn't make any difference. For those patients who are gonna die, the same fraction of them fell under sixty five, whether you were gonna whether you were septic or whether you're not septic. Alright. I think maybe you're picking up a theme here, but let's let's take another couple of steps. And this work was presented at, an AMIA conference, American Medical Informatics Association, and I will provide a reference to the the presentation. And so you can see these, the data, all together in one place. Step number three is we plotted mortality versus RI at admission separately for septic and non septic patients. And what you see is, of course, lower RI scores of higher mortality. You get a curve which curves down. Low RI is higher mortality. High RI is lower mortality. But the lines for septic patients and non septic patients, basically fall on top of each other. And the last piece of data that I will submit to you along this same vein is we created a couple of linear logistic regression models. The first one used RI at admission to predict mortality in the hospital and the AUC area under the curve of the receiver operating characteristic. I'm sure many of you are familiar with that measurement. It's sort of a measure of how successful your model is in ordering patients in terms of riskiness. Anyway, the AUC for a logistic regression model using one variable only and that is the RI at admission was zero point nine zero seven. Zero point nine zero seven. If we add another variable, and that is just a flag to indicate whether the patient was identified as septic or not septic, what do we get for the AUC? Instead of nine zero seven, we get nine zero nine, a change of two in the third decimal place. Basically, flagging that patient as septic did not improve the model's ability to predict whether you were gonna live or die. So what is this all leading to? What it's leading to is if you wanna focus on preventing death, don't focus on whether that patient is flagged as septic at admission. What you should focus on is how sick that patient is. And so the the answer to the question, how do you reduce sepsis mortality, is is kind of something obvious. What is the answer? How do you reduce sepsis mortality? You pay attention to those patients who are sick. You don't focus on whether or not they're flagged as septic. You focus on how sick they are. And one way to focus on their acuity is to use a score that quantifies acuity. And in this case, what we've been talking about is this score called the Rothman Index. So let me just cover one more topic on this, podcast, And that is what I call the trifecta of warning. Well, a lot of systems will alert when a patient is particularly acute. You can use a MUSE score or a NEWS score or an eCART score or an EPIC deterioration index score or a number of other scores out there. There's Pugh score, there are CRWS scores, there are lots of early warning scores. And these scores tend to be tested with retrospective studies. And that gives you an indication of how effective the score is in identifying patients who are likely to have a bad outcome. But retrospective studies are limited. And that is they don't distinguish between flagging a patient for attention for whom the doctor and the nurse already are aware of the problem versus flagging a patient for attention who the doctors and nurses are unaware of. And so what is the trifecta of value? The trifecta of value is, first of all, if you are trying to warn the doctor, the nurse, you want to direct their attention to this patient, the first thing you have to be is correct. If you say the patient is sick, they have to be sick. That is really the least difficult part of this trifecta. If your blood pressure is plummeting, if your heart is racing, you're sick, you need attention. The second element of the trifecta is this alert has to be timely. And that is, I can't give you an alert, ten minutes before the bad outcome occurs, the transfer to the ICU, patient dying. You have to provide that alert with enough lead time so that you can intervene and make a difference. So it has to be correct and timely and then you get to the most difficult part of the trifecta. And that is this. The warning has to be new. This is what people tend to forget when they're building models, especially when they're being tested retrospectively, which is the vast majority of modeling papers that you'll see in the literature. It has to be new. What does that mean? What that means is well, why why are you in this business to begin with? Are you in the business to write papers? That is the case for some folks. That is why they're they work in the field to write papers. It's not a bad thing, and you can advance the state of knowledge, but that is that's not what motivates me. I'm in this field because I've always wanted to make a difference. And the Rothman Index started with a story about my family, which I will not relate today. That'll be in an upcoming episode. But the situation is doctors and nurses, by and large, do the right thing. They work hard to make sure that the patients in their care get the best possible care, have the best possible outcomes. So any system that you're adding to that hospital has to be an adder. It has to provide value beyond the best efforts of doctors and nurses. That's why a retrospective study really doesn't give you the answer, is this system useful? The only way to know whether a system is useful is to implement it, integrate it into workflow, and see if it makes a difference. Does all cause mortality go down? Do unplanned transfers to the ICU, the the rate of unplanned transfers, do they go down? And if the answer is yes, then yes. Your system has made a difference, and you have hit the trifecta. You're correct. You're timely, and you're new. You're adding new information to the story. You're adding information that the doctor or the nurse can act upon to intervene, to make a difference. And I think that is where we're going to end our podcast today. So just to recap, we've been talking about sepsis. Yes, sepsis is associated with a third of hospital deaths. Yes, if you don't provide IV antibiotics to patients in septic shock. If you delay that, you increase mortality. But those don't tell the story. Those numbers are widely misinterpreted. The real answer is yes, patients do die of sepsis. How much of it is avoidable? Some. Some is avoidable. The vast majority is not. Some is avoidable. But if you want to reduce mortality, you focus on patients who are sick. I mean, it seems so obvious. Of course, you know, that's what you would do normally. Yes. Okay. Well, it's not always that easy to know just how sick someone is. It's easy to know when they are, you know, extremely sick. It's easy to know when they are relatively in good shape. The tricky part, the important part is understanding when they are starting to deteriorate in the when the early stages of, decline show up. That's when you have an opportunity to intervene. That's when you can affect mortality. And I talked about how do you reduce sepsis mortality? Well, the answer is you reduce overall mortality. You focus on those patients who are sick and you intervene when there's still an opportunity to prevent that bad outcome. You hit the trifecta. And in our next podcast, I will tell you how I came to the story of the trifecta and how we achieved hitting the trifecta and how we've actually saved lives and and where the Rothman index came from. Okay. Well, with that, I will, I guess I will end, episode three. Maybe not quite as much humor in this one as, in the others, but, I'll look for opportunities to lighten the story. It is a serious story. It is a serious we're in a serious business and I have the utmost admiration for the doctors and nurses who take care of millions of patients in hospitals in this country and throughout the world. It's a tough job and it is one where doctors and nurses could use just a little bit of additional help. And that's what we'll be talking about in episode four. And until then, thank you very much for listening to the Michael Rothman, and hope you have a great rest of your day.