Hello and welcome to today's Journal Club Critical Care Medicine webcast. This webcast, hosted and supported by the Society of Critical Care Medicine, is part of the Journal Club Critical Care Medicine series. This webcast features two articles that appear in the June 2023 issue of Critical Care Medicine. The webcast is being recorded. The recording will be available to registrants on demand within five business days. Log in to mysccm.org and navigate to the My Learning tab. My name is Tama Zagmani and I'm a professor of intensive care at Cardiff University in the United Kingdom. I will be moderating today's webcast. Thank you for joining us. Just a few housekeeping items before we get started. There will be a Q&A session at the conclusion of both presentations. To submit questions throughout the presentations, type into the question box located on your control panel. If you have a comment to share during the presentations, you may use the question box for that as well. You will also have the opportunity to participate in interactive polls. When you see a poll, simply click the bubble next to your choice. And finally, everyone joining us for today's webcast will receive a follow-up email that will include an evaluation. Please take five minutes of your time to complete this. Your feedback is greatly appreciated. Please note the disclaimer stating that the content to follow is for educational purposes only. And now I would like to introduce today's presenters. Dr. Andrea Rossetti is an associate professor of neurology at the University Hospital in Switzerland. He has also served as an adjunct professor at the University of Copenhagen in Denmark. He's an author of more than 240 peer-reviewed articles and his research interests include status epilepticus classification and and his research interests include status epilepticus classification and management, pharmacological treatment of people with epilepsy, and early prognostication of coma especially after cerebral anoxia following cardiac arrest. Dr. Lars Welthuis is an anesthesiology resident at Erasmus Medical Center and a PhD candidate at the Department of Anesthesiology Amsterdam University Medical Center. Dr. Welthuis studied at the University of Amsterdam where he started doing research at the emergency department of the Amsterdam University Medical Center. During this research, he worked as a resident at the intensive care unit of the Red Cross Hospital in Deverik and the intensive care unit at Amsterdam UMCs. Lars continued his research at both the anesthesiology and emergency departments of Amsterdam's university medical centers. In May 2022, he began his residency training in the Department of Anesthesiology at Erasmus Medical Center. He will have his PhD defense this October. And I would like to hand over to our first presenter Andrea Rossetti. Many, many thanks for the introduction and also for giving us the opportunity to present the work. It's really nice. So I'm going to talk about multimodal prediction of favorable outcome after cardiac arrest. That was our paper. And these are my disclosures that are not pertinent to the talk tonight or today for the US colleagues. And I'll start with a couple of introductory thoughts. As we know, cardiac arrest is one of the most frequent diagnostic admissions in the ICU departments. You see some epidemiological data here. And the bottom line is that at the end, it's only a tiny fraction of those that suffered a cardiac arrest that will leave the hospital living. More or less 8 to 10 percent at best of the total patients. And three quarters of these will have a so-called favorable outcome, which will correspond to about 5 to 7 percent of all patients. The idea is to try to prognosticate as early and as reliably as possible this patient, both on the poor outcome and on the good or favorable outcome side. To this intent, there are several modalities that have been developed and studied over the last decades. You see in green clinical neurophysiology, in red clinical examination, and then biomarkers, mostly neuron-specific analyzer, maybe neurofinement light is also a measurement not so widely used so far, and also in violet neuroimaging. And the idea is to be multimodal to increase the performance of prediction. These are the last European guidelines. I think these are the most comprehensively studied guidelines worldwide. It's a European publication of 2021. And you see there are six elements at 72 hours in patients that are still comatose that will predict a poor outcome with high specificity. Once again, it's important to be highly specific in order not to make false predictions of poor outcome. That's unacceptable, of course. And you see you have a lack of return of brainstem reflexes, the lack of response on the cortical level to somatosensory potentials, high neuron-specific analyzer in the serum, a so-called highly malignant EEG, which is an EEG that is flat with or without prior discharges, or inverse suppression, status myoclonus or prolonged myoclonus that's hardly amenable to improvement with medication, and final extensive injury on neuroimaging. And you need at least two of these elements in order to prognosticate a poor outcome, meaning death or a high degree of handicap. Conversely, there are some groups that have directed their attention to prognostication of good outcome. This has received less momentum historically, but it's more moving in the last few years. This is a summary of a recent paper that isolated some elements like localizing motor response, a continuous background with reactivity on painful or auditory stimuli on EEG, somatosensory potential with high amplitude response on the cortex, very low seromimancy, or normal imaging. Most of these elements have a good specificity, but not a good sensitivity. Now, you can argue that if you want to pick up those that will improve, you rather need to be sensitive than specific. And you, as you certainly know, there is a trade-off across these two elements. For the interested audience, this paper that has been published last year in Intensive Care Medicine, it's excellent at reviewing very carefully each element that portends a good outcome in a unimodal manner, meaning there is no combination of predictors. And what we tried to do was to assess a multimodal prediction for a favorable outcome, pretty much like the multimodal prediction of poor outcome, actually. And to that, we put together two cohorts that come from two sides of the Atlantic Ocean. I will go into more detail in a in a short while. So, the aim was to build up a multimodal score to predict a good outcome in comatose patient after cardiac arrest. So, we had a derivation cohort of nearly 500 patients, consecutive patients at the Lausanne University Hospital in Switzerland. This is the SHUV. And we have a registry with a scheduled outcome at three months using cerebral performance categories. And we define 1 and 2 as favorable outcome. And we used also a subsidiary outcome of survival 1 and 3. Just in parentheses, CPC 3 is very heterogeneous. You can have patients in minimal conscious state, meaning barely interactive on the one side. And on the other side, you can have patients that cannot work, have to have some help, for example, for tax declaration or so, but are well interactive. And so, at times, it's important also to include the CPC 3 in some outcomes. We assessed each potentially predictive variable unimodally to the performance of favorable outcome prediction. And we combined the most discriminant items in a multimodal score. And then we validated the score for a cohort from the Brigham and Women's Hospital in Boston, Massachusetts. It's a smaller cohort as compared to the derivation cohort. And the proposed score is a six-item combination with a eponym, which is supposed to be of help. I don't know if it's going to be of help, but no to our CONFOR. So, if we go through each line, we have the early EEG during target temperature management, so between 12 and 36 hours, which has to be not highly malignant, so not suppressed, not in burst suppression. Then the reactivity in both the early and the later EEG and the continuity in the later EEG. So, four EEG parameters. I'm an EEG-er, so that's half of the authors or more are EEG-ers. That's maybe a bias. We can discuss that later. The N is for the neuron-specific NLAs. We identified with an AUC a threshold of 41. So, remember, the proposed threshold before was 17, highly specific, but with low sensitivity. And here we try to optimise the sensitivity. So, we go up to 41 micrograms per litre. And then the same approach for the four score, the full outline of our responsiveness score, that has to be at least five or more at 72 hours of sedation. So, each parameters become a point and have a maximum of six and a minimum of zero points. And you see the area under the curve on the left with an area of 0.88 in the derivation cohort, and on the right, the table with the scoring, and at a scoring of four to six or more, four out of six or more, sorry, you have a sensitivity of 97 and a specificity of 65 to identify patients with CPC1 to two at six, sorry, three months. This is the validation cohort. So, pretty much similar results. Of course, the confidence intervals are bigger, but the point estimates are quite similar. They are bigger, the confidence interval since the cohort was smaller. Then for the subsidiary outcome, you see this is survival. There is another threshold, not four points, but three. So, a different threshold for a more liberal outcome, and you see high sensitivity and reasonable specificity and thus accuracy. This is the calibration plot, and you see that the score tends to somewhat underestimate and somewhat in low scores and somewhat overestimate in high scores. The outcome, so the best calibration is around five, but all in all, the mean error was very reasonable. So, we come to the discussion. So, this is a reflection of the increasing attention that has been devoted to identification of good prognosis, which we think it's a good movement, because it's reasonable to be robust about poor outcome, having a high specificity, high specificity, but is somewhat nihilistic. And so, it's also important to pick up the patient that can really improve and devote the energy to them. And it's also, as far as we know, the first really comprehensive multimodal approach to improve specificity of optimal sensitivity. There have been some few papers before, a couple of approach with two or three predictors, but they were on highly selected populations, and this is a general population of many hundreds of patients. The strengths of the study is the large derivation cohort, the standardized EEG scoring according to American Clinical Neurophysiology Society guidelines, and the Withdrawal of Life-Sustaining Therapy guidelines. There is some external validation, there is a suggestion of thresholds that are more liberal, especially for the neuro-specific NLAs than those that have been proposed so far. And the two different cut-offs for different outcomes may guide the clinicians to counsel families also. And this is, we think, a plus of our analysis. Of course, each study has limitation, among them the sort of bias towards EEG as compared to other modalities. There is no MRI in this course, since at the SHUV we don't use routinely MRI in each patient. We use MRI in patients that are in a clear prognosis, and especially we suspect a poor prognosis, since so far the MRI, the normal MRI, has not a really recognized value to portend a favorable prognosis. There has been no data to the N20 amplitude, there is some literature on that, maybe something that we can look in the future. And finally, the sedation was somehow standardized across the centers. Of course, one center uses very different sedation protocol, that could be an issue. Now I come to the polling question. There is no right or false answer just to get the pulse of the audience. If one looks at favorable outcome in this scenario after cardiac arrest, what should she or he ideally target? High specificity to minimize false positive, high sensitivity to minimize false negatives, high accuracy, so meaning a good combination of both, or none of these are really useful. So what do you think? There's no examination, go ahead. Almost, yeah, 100%. My goodness, okay. Well, I agree, but of course, it's open to discussion. I like to finish here, thanking all the collaborators that made this study possible. On the left top, you see a view of Lausanne with the Lac Le Mans. On the right side, you see a view of the Charles River and Boston. In the US, at the SHUV, we had a neurology department, the adult ICU department, and in Boston, mostly neurologists and urine intensivists working together, a nice collaboration, and I hope an interesting study. Many thanks again to the audience for the attention, and now I would like to invite my colleague, Dr. Lars Weltheus, from Amsterdam for his presentation. Thank you very much for this nice introduction, and thank you for inviting me to present this article. I will talk about the recognition of critically ill patients by acute healthcare providers, and indeed, I work as a resident of the anesthesiology department in Rotterdam, and performing my PhD at the anesthesiology department in Amsterdam. So what we've all seen, first, I do not have anything to disclosure, and I will talk about the issue we are facing at the emergency department at the moment. We will talk about the possible solution, about the methods we use to perform this study, what our results are, discussion, conclusion, and further perspectives. So the main problem we face globally is that at the emergency department, there are an increasing amount of patients needing help from our physicians, but also from nurses. This leads to less time for each patient, and a higher patient-to-nurse or patient-to-physician ratio. So we have less time for each patient, and we can see them less, but also less frequently, and we know that if you see your patients just a couple of times and just a couple of minutes, this will lead to a less true flu investigation of the patient, which leads to increased mortality, more cardiac arrest, and the need for more intensive care unit admission. So a possible solution in Europe is instead of just triaging the patients, we also use like an early warning score. An early warning score is mainly developed for the ward to to recognize possible deterioration of patients, mainly surgical patients, to ask the intensive care unit to just assess the patient and to see whether the patient needs intensive care unit admission. So these early warning scores are developed to predict deterioration in hospitality, but also mortality. But these are not developed to use at the emergency department. However, they are used, and in the Netherlands we use the modified early warning score, which is just what it says, like a modified score of the originally early warning score. So how does it look? This is the early one, the modified early warning score, and it's basically just a score based on vital parameters, such as the pulse rate, the systolic blood pressure, level of conscious temperature, etc. And further away from the normal values, you will get more score. And at the threshold score of three or more points, you are likely to develop critical illness or likely to deteriorate. So, despite having our use that the modified early warning score is used more frequently in the emergency department, also pre-hospital, there's no proof that it actually is better than clinical judgment or clinical gestalt. And clinical gestalt and clinical judgment is basically the gut feelings of us as physicians or nurses after we see a patient and we say, well, I don't think this patient is actually well, so we should check up more frequently. So what we've done, we performed a prospective multicenter study in two academic teaching hospitals, and we included all adult patients that presented by the ambulance of two academic teaching hospitals in Amsterdam for four months. And we excluded patients that had ongoing CPR because they go to the ICU as a standard. And we also excluded patients that were intra-hospitally transferred, as they were already seen by a physician and treatment has already been started. So the outcome we looked at is the accuracy of the MUSE, but also of the clinical judgment of emergency medical service personnel, emergency department nurses, and ED physicians in predicting critical illness defined as intensive care unit admissions or as a serious adverse event, such as cardiac arrest or delayed need of intensive care unit admission. This all had to be in within 72 hours, as this more reflects the actually state of the patient at the emergency department, rather than the influence of the disease itself for a longer period. So our secondary outcome was accuracy in predicting 28-day mortality, and also the severity of illness as a proxy of the development of critical illness. So the severity of illness was divided into four groups, like not ill at all, minority ill, moderately ill, or severely ill. So in total, we included 800 patients of whom 10 we needed to exclude as they were transferred to a different hospital, and we did not have the follow-up data. But of the 790 patients, 14% became critically ill, where the majority was directly admitted to the intensive care unit, but also patients developed sepsis, while it was not stated as the diagnosis at the emergency department. So they became septic within 72 hours after ED admission. So when we look at the primary outcome, with the prediction of, on the left side, the sensitivity, specificity, negative predictive value, and the positive predictive value are stated, for each of the groups studied, which were the emergency medical service nurses, the ED nurses, and the physicians, compared to the modified early warning score. So when we zoom in a little bit, we see here the score and the sensitivity and the specificity of the emergency medical nurses, and we see that the sensitivity is quite low, only 41%. However, the sensitivity of the MUSE score, with the cutoff value of three or more, was significantly higher compared to the MS nurses. However, specificity was significantly lower for the MUSE compared to the AMS nurses. When you look at the negative predictive value, you see a similar result. However, positive predictive value was like twice the amount compared to MUSE. Similar results are found when we look at the predictive value of the emergency department nurses, with a sensitivity which was also significantly lower compared to MUSE, but a specificity significantly high. However, interestingly, the sensitivity of physicians, ED physicians, was similar compared to the modified early warning score cutoff. However, specificity was significantly higher. So basically, the physicians, ED physicians, have a similar sensitivity as MUSE, but however, significantly higher specificity. So, when we look at the primary outcome, we state that the EMS and ED nurses have a lower sensitivity compared to MUSE, but the ED physicians have a sensitivity which is similar to MUSE. All healthcare providers have a significantly higher specificity of MUSE compared to MUSE. So, I think here it gets really interesting, because our secondary outcome was the prediction based on the scale 1 to 0 to 3. And when we look at the area and the receiver operating characteristics, we see that the physicians and the ED nurses have a significantly higher score compared to the modified early warning score. And the area and the receiver curve is basically the sensitivity and specificity of each of the threshold score. So when we look at the cutoff points, which are relevant of our study, which were the highest score possible, and the score of three or more for the Modified Early Warning Score, we see that the EMS nurses and the ED nurses have a similar sensitivity compared to the Modified Early Warning Score. Also, the physicians have a similar score. However, once again, the specificity of all of the health care providers is significantly better compared to in use. So basically, what we see is that we find it hard to tell whether the patient becomes critically ill or not. But we do find it easy, or at least possible, to predict which are the patients who are severely ill. 28-day mortality, in short, we cannot really predict well enough. Because the sensitivity is really low, however, specificity is high. This is the same for physicians, ED nurses, and EMS nurses. So despite having a multi-center prospective study, we do have some biases introduced. Because we only included patients presented during the day, as it was not possible to have students during the night. And we did not take into account if patients had a do not resuscitate policy. Also, a majority, a major limitation is the imputation of data. Up to 60% of the data was imputed. However, the majority of the data not available was the temperature of the patients or the level of consciousness. However, these were measured prospectively. And up to three hours after ED presentation, we used the same score measured. So we did not make up any random imputation of data. So in conclusion, this is the first study that actually compared a vital parameter-based risk stratification score with clinical judgment. And interesting enough, we find that the healthcare providers can actually perform pretty well in predicting short-term critical illness. Also, stating your concerns about the patient is highly relevant. And I think we should do that more often. Because two weeks ago, one of my colleagues called me. And he said, well, I do not feel very good about this patient. So we assessed the patient. And we both had something like, well, the gut feeling, yeah, there's something wrong with the patient. And up to one hour after that, the patient had a cardiac arrest. So we can, despite not having any vital parameters that suggest that we have a major incoming serious adverse complication, we can predict it. How, we do not know for certain. And we need to study that in the future. But at least we can. So the main question I have for you as a colleague is, can your junior doctors, can they actually call the supervisors and tell them they are worried about the patient? Or is that not possible in your hospital due to the hierarchy in the hospital? You can type in the chat and just want to discuss it. Because I think it's really important. Also, medical students should always be able to say that they are worried about the patient. But I find it difficult in the hospital I work at. Because I'm not always sure that the students will discuss it with the supervisor when the vital parameters have been assessed. Thank you for your attention. Yeah, I see that some people can actually, young colleagues can and should pop up. Yeah, they should. Well, thank you for your attention. If there are any other questions, please let me know. I'm happy to answer them. Thank you very much for both of you for the really interesting presentations. And I'm just looking at the questions and the chat. I think if I may answer with our experience, Lars, I think we are trying really hard to encourage a culture where the hierarchy is flattened and the junior colleagues can speak up and tell us if they are worried about patients. And I think it is more successful in some specialties than in others. But actually, to that, there is one question that I would like to ask. Do you think that in your study, the seniority of the doctor has any potential modification effect on the results? Is it possible that you had a doctor who is a very experienced physician and they could be more accurate? Or those experienced physicians might be at this moment more burnt out and they are less accurate in their predictions compared to a score? Yes, thank you for your question. One of the sub-analyses we performed in the study is the correlation between experience in years and the accuracy of the health care providers. And we see that there is indeed a correlation between experience. So more experienced, but also elderly nurses who have seen lots of patients and had a different teaching do have a significantly higher accuracy in predicting the deterioration or the need for intensive care units. So there is indeed a correlation between experience and predictive accuracy. Thank you. The next question goes to Dr. Rossetti. And you stated that you are an EEG enthusiast. And obviously, you've got a lot of experience and knowledge on this field. Do you need a trained neurophysiologist or neurointensivist, an EEG buff to interpret the results? Or do you think that there is a scope for a non-trained intensivist trying to interpret the waves? That's a very good question. So if I answer as pertains to the presented study, we use the American Clinical Neurophysiology Society criteria. And of course, you would need some sort of training in order to identify correctly the patterns. Having said that, of course, the more widespread some eugenology is, the better it will be. One has always to be humble and know her or his limits, of course. I can think of several colleagues in my multidisciplinary ICU that perform, for example, cardiac ultrasound without being cardiologists. They can find quite easily the ejection fraction, for example. But of course, if you need to have a decision on a replacement of the mitral ball or so, you need a specialist. So that's just an example in another field. And I think for having some ideas of how the patient is doing, an interested intensivist, meaning an intensivist that has some exposure to the EEG, is fine, of course. I hope I answered to your question comprehensively. I think, yes. And I've got a follow-on question on that, if I may. Sure. And again, it is our own experience. And I think I can speak of many other institutions that it is OK that we obtain the EEG, because we could get hold of the technicians. But the interpretation of that is difficult. Do you think that, let's say, we train the intensivist to recognize some basic patterns? That is one approach. The other approach could be to use telemedicine and telecritical care to support the units which don't have this expertise. Is this something that you think is feasible when you are talking about neuroprognostication of cardiac arrest patients? I honestly don't know. But if you allow, I will step aside. So there has been already some interest in trying to instruct in a relatively efficient manner, in a fast manner, ICU nurses and ICU physicians, and to big families of patterns of EEGs, like encephalopathy, status epilepticus, and so on, and artifact recognition. And the results are really promising. You need, though, some training. And in my sense, you need practice. You need people that keep looking at the EEG. If you look at the EEG twice a year, it's going to be difficult. And that's for, in my sense, sorting out overnight or during the weekend when the experts are not around, the important alterations that can trigger consequences for the patients. For prognostication, I think you should have no room for doubts or for uncertainty. Because if you overestimate a pattern, for example, of poor prognosis, and this pattern is, I don't know, I can find, for example, a calibration error from the technologist at the beginning, and then you are just playing with the life of the patient. So I think also for legal issues, it's always good to have, for such decisions, a formal report. Now, for telemedicine, our French colleagues did that for a while. There are advantages, of course, of availability of reading. And the disadvantage is that the people that are sitting hundreds of kilometers apart have no experience of the way of working of the ICU, the state of the patient, though they can have the video. And they tend, in their experience, to be very descriptive in their report and not helping very much. That could be encephalopathy, but irritative component not excluded. That doesn't help too much. And that's maybe a challenge for the telemedicine. Thank you. That's a very helpful and comprehensive answer. Lars, there is another question about the early warning scores, which were developed to predict 28-day mortality, ICU admission, or cardiac arrest. And we know that there are probably more than 40, but definitely I know of 34 different early warning scores which have been used and, in some way, validated in different populations. Now, admittedly, the MEW score that you were using is one of the weaker performers. Do you think that if you would have used a different score, you would have had different results? Yeah, thank you for your question. Yes, definitely. Because the MEW score is always inferior compared to the NEWS-2 or the NEWS. However, the MEWs originally had two additional points you could give to the score when you're worried about the patient. So that modified early warning score was like a score that, like the NEWS or the NEWS-2, which had an additional option just to get two points because you were worried. So your supervisor had to come and to assess the patient. However, this is not used anymore in the Netherlands. And the MEWs is like the implemented score in many hospitals. So that's why we studied the MEWs. But I think, definitely, when we compare it to the NEWS-2, the NEWS-2 always has like an area under the curve of 0.1 score, 0.1 higher compared to the MEW. So definitely, our results would have been different. Thank you. Going back to Dr. Rossetti and asking about another modality, you mentioned the NSE levels. And a lot of places are using the neurospecific analyzed levels to predict bad outcome, as opposed to good outcome. What is your view on the serial measurement of these levels? Are they any helpful? Do you use them? Yeah, thank you. Well, the biological half-life of NSE is about 24 hours, I think. So it's quite short living in the serum after being produced in the worst-case scenario for the patient, but best-case scenario for the prognosticator by dying neurons, if you exclude formally, for example, amolysis. And actually, by experience, you'll see that you can have an increase up to 48 and at times 72 hours after the cardiac arrest because of secondary injury or apoptosis or whatsoever. So it definitely makes sense to have a serial measure. If you have just one measure, probably the best way of doing this is about 48 hours. But at 24 hours, you're trying to underestimate the peak. And actually, the peak has been repetitively shown to correlate quite robustly with the outcome. Now, on the other side, you see that if the NSE is somehow low or low, yeah, not really normal, but lower than some thresholds, then you have a compatibility of the patient with good outcome. That's also something interesting. But it's important to recognize that the threshold are different. So now, we use about 60 according to the European Society guidelines for poor outcome and should be lower than we are. So we propose 41. Other teams propose 17, but definitely lower for good outcome. So there is a gray zone somewhere. And I think the shadows of gray are really interesting. And we should dig in and try to understand better what happens there. Thank you very much. Going back to Lars again, please correct me if I'm wrong, but your study was mostly looking at the medical personnel ability to predict the disposition of your patients. Would it be worthwhile looking at specifically non-medical personnel? Because in a lot of systems, doctors are quite hard to come by, especially out of hours. So we staff our emergency departments with advanced nurse practitioners, senior nurses, et cetera. And they are making these judgment calls. And they would be the ones who would be raising the flag that this patient needs further attention. What are your views on that? Yes, that would definitely influence the results. But however, the patients we looked at are all patients presented by the ambulance. And we have this difference between a 112 call, which is similar to the 911 call, and patients presented to the hospital by the general practitioner. So every patient that is presented after a 112 call is directly assessed by an ED physician and a nurse. However, the patients just presented to the hospital by the GP are only seen by the emergency department nurses. So there's indeed a difference between those patients. Thank you. Dr. Rossetti, two more questions for you. One is, again, just thinking about the, well, both of them are thinking about the future. And one is thinking about AI and its potential role. There is a lot of literature already available on AI-supported imaging. And do you think that this could enter the clinical arena and the prognostication model in the next, I don't know, one or two years in a short space of time? Very timely question. I will try to answer in an autobiographical manner. When I was in high school several years ago, I used to play chess. It was in the 80s of the last century. And I was a low-average player for the club. And of course, I had no chance with a computer. But the best players of the club, they always could beat the computer. And even the best computer at that time had no chance with the best player of the club. And talking about club, club was maybe the thousandth player in the world, the best of the club. It was unconceivable at that time that a computer could beat the world champion. Now we know, since 15 years, that the computer has 100% winning percentage against the world champion, 20 years or 25 years later. So it will come. Of course it will come. I don't think in the next couple of years. Why? Because so far, as you rightfully say, there is a bunch of literature, but by several different groups with several different approaches from the mathematical side, from the deep learning side, and so on, and also using different parameters. And so far, the generalization is lacking. And I think we will have to unify a little bit this work and try to generalize in a way that then there is a sort of plug and play approach for centers that don't have the engineer around that can understand what happens. And in my sense, but I'm not sort of God that knows everything, but in my sense, it will last maybe a decade or so. OK, thank you. It is helpful. My own opinion, and I'm no expert at all in that, is that I think it will be less than a decade, because how quickly we have developed these tools, and now as they are entering the more mainstream section of our life and society, I think the acceleration will be even more exponential. And I can fully relate to your chess analogy as my son plays chess. And he's a good average player, but obviously he's got no chance against computers nowadays. So that is something which is a new reality for those who are younger than us. And one last question to you, Dr. Rossetti. And again, it's about multimodal prognostication. Do you think that any new biomarker, let it be protein or maybe an mRNA-based biomarker, could enter this arena? Are you aware of anything which might be interesting? Well, probably the most advanced biomarker, as I briefly mentioned, is the neurofilament light that has better performances than NSE. It's difficult to generalize because of the sort of lab that you need to measure that. And we have to remember that biomarkers have to be there quickly. There is no need for a biomarker that needs two weeks because the patient is then already dead or awake, more or less. So I'm just aware of these biomarkers that there have been many others. But probably the most promising is NFL, I think. Thank you very much. And going back to Lars, a last question for you. There is a trade-off between the sensitivity and specificity. And do you think in this scenario, the sensitivity should be higher? And if yes, then what about the risk of over-treating, increasing costs, and essentially maybe altering the outcome because of the over-treatment of certain patients? Yes, thank you for your question. I think this is the question we should always ask. Like, how can we interpret the results? And why is it important? I think we should not miss any patient that are potentially becoming critically ill. So I think consulting the intensive care unit or consulting your senior is always a good idea. So I think that should be the goal. And of course, we have things like alarm fatigue or a higher cost. But I think having a delayed intensive care unit admission with all the adjusted extra morbidity or even mortality will cost us a lot more than additionally consulting of the intensive care units. Thank you very much. And this concludes our Q&A session. And I would like to thank both Dr. Rossetti and Lars for their excellent presentations. And to the audience, thank you for joining us today. Please take that five minutes to complete the evaluation, which will be sent via an email. Your feedback is really important to us. And I would like to welcome you next month on Thursday, the July 27, 2023, where we will have our next Journal Club Critical Care Medicine webcast. Thank you and goodbye.

Journal Club Critical Care Medicine

multimodal prediction

favorable outcomes

cardiac arrest

EEG parameters

neuron-specific enolase levels

clinical assessment

acute healthcare providers

emergency department