So, good morning, everybody. Thank you for being here. The faculty that you saw, we have discussed a lot this session, and we really hope you enjoy. And Chal has, of course, spearheaded and lead this group on these presentations today. So, my part is going to be what is acute kidney injury, why it's relevant, a little bit of new concepts on risk classification and management, and set a little bit the foundation of this discussion about organ crosstalk. My disclosures, and first, I'm going to discuss why AKI is relevant, a little bit of epidemiological concepts, then a little bit about risk classification and subphenotyping, and finally, some clinical applications. So, why AKI is relevant? It's very frequent. AKI incidents, particularly in hospitalized patients, it's about 20%. If we go to ICU settings, this doubles to 50%. Now, AKI also can have complications as acute kidney disease. Acute kidney disease is a concept when AKI persists more than seven days in the first 90-day window after the initial onset. And the epidemiology of acute kidney disease is about 25%. That means one of four patients that develop acute kidney injury, they have this complication of acute kidney disease. And of course, there is the complication of chronic kidney disease, that is about one in three patients, if they survive long enough in the next five years, can develop either incident or progression of chronic kidney disease. This is important because we know chronic kidney disease is clearly associated with cardiovascular health. A very important concept here is that a lot of these patients, their morbidity and mortality is not necessarily related to the kidney problems, but the cardiovascular health. And this is something we're going to discuss with detail later today. Now, important is also that patients that suffer from AKI in the hospital are very debilitated, and particularly the elderly population. So this is USRDS data, and when you see patients that are 65 or older, you can see that not all these patients that survive the hospitalization go home. A lot of these patients, they go to facilities like nursing homes and LTACHs. And this is relevant because this creates a burden in the healthcare system, because what type of rehabilitation facilities we're going to send these patients after these type of events. Now, also important to recognize that less than 10% of AKI survivors receive guideline-based care after discharge. Right? That is a whole other discussion, but even what we do with those that go home and survive in terms of follow-up, it's suboptimal with what guidelines will recommend. Very important concept, and this link very well with the intensive care rehabilitation is that these patients are also not recovering their physical ability, and not only physical, but also cognitive ability to resume their daily activities. So this is a study our group did with patients that suffer AKI in the ICU only. And as you can see, we classify the patients according to having a stage 2 or 3 AKI, that is severe AKI, and having either no AKI or mild AKI stage 1. And then there were a lot of assessments of physical function in the clinic after discharge, typically around the first two to three weeks after discharge. And you can see that patients that suffer from severe AKI have less capacity when they walk in the six-minute test on the treadmill, and also had the opportunity to evaluate other metrics of quality of life, and there is definitely decreased scores in quality of life in patients that suffer from AKI. And importantly, a lot of the questions that they were very interested about was the ability to resume their work or hobbies, and also the ability to continue to drive for those that already were driving before the events. This is the graphic representation of that in reference to what is predicted in terms of physical distance that they walk on the treadmill for a patient similar sex and similar age. And you can see all these patients with a dose-response relationship are debilitated, and they have less physical capacity after AKI. Very important aspect is that when these patients have been evaluated in different scores of quality of life, they have clearly identified domains where they are affected in mobility as we discussed on the other study, but also pain, inability to resume their cognitive activities or have the same cognitive fitness as they had before, and also the requirements of assisted care. And over one in four patients that suffer from AKI and survive the hospitalization reported health states equivalent to or worse than that. So this is definitely these patients are debilitated if you combine acute kidney injury with critical illness, the rehabilitation process certainly is essential for these patients. Now when we talk specifically about the kidney, the kidney after episode or multiple episodes of stress and injury need to repair. And when we have an adaptive repair in the kidney, this includes multiple levels at the cellular level, like tubular proliferation, endothelial regeneration, mitophagy, and also the effects of the fibroblast macrophage transformations, and importantly, resolution of inflammation. So if these are all orchestrated, the kidney will repair and hopefully recover, but still this kidney start losing the capacity to respond to subsequent stress and can be susceptible to recurrent acute kidney injury. When the kidneys don't orchestrate these processes of repair is what we call maladaptive repair, and the problem could be at different levels on all these domains. And most of these patients are the ones that are at high risk of developing acute kidney disease and later chronic kidney disease. And as you can see in the right side of the chart, recurrent AKI patients can also develop acute kidney disease, and there is an interplay between these recovery levels and complications after AKI. Now importantly, if you see in the top of the slide, what are prognostic factors we can use to identify patients that have higher chances of maladaptive repair? The severity of AKI is very important, the duration of AKI as well. And when we evaluate the level of recovery at the time in the first window of 90 days, that is a good surrogate of identifying patients at risk of recurrent AKI and also chronic kidney disease after AKI. Is it frozen? Not moving? Yeah. Thank you. Yeah, I think my mouse probably froze. Yeah, so and importantly when we see here the evaluation of the recovery, it's a good surrogate of chronic kidney disease after AKI. This is a study in septic AKI patients in particular, and you can see if the creatinine in the first 90 days after discharge is still 50% above the baseline creatinine pre-event or pre-hospitalization, you can have a 70% risk of developing chronic kidney disease. And there is a dose-response relationship, right? So this is relevant because now that we are developing post-discharge clinics or rehabilitation after critical illness, the intensivists need to be well-familiarized to evaluate levels of recovery after acute kidney injury and correlate these with the risk of these patients of developing chronic kidney disease. Can you go next? Similar as I showed you the creatinine level help, the level of GFR drop in reference to baseline also relates to the risk of developing end-stage kidney disease long-term. So here you see if you evaluate this GFR at 30, 60, and 90 days after a hospitalization with AKI, the drop in GFR that fluctuates between 20% and 40% increases the risk between five and seven times of developing end-stage kidney disease in the subsequent five years in these patients. Perfect. Now it's working. The other important aspect, as I said, is the cardiovascular health. And this is one of the studies epidemiologically that highlighted the increased risk of major adverse cardiovascular events in these patients that suffer with AKI. This is a study that used VA data and compared patients that developed episodes of AKI and developed an episode of AKI on top of a coronary event versus a patient that only had the coronary event without AKI. And you can see that having acute kidney injury on top of a coronary event that was very significantly a significant factor that increased the risk of major adverse cardiovascular events in the follow-up in the next five years. So one year after admission, 25% of patients in the MI and AKI group had experienced a cardiac event compared with only 19% in the MI group alone and 7% in the AKI group alone. So there is an effect of increased risk of developing cardiovascular events if you develop AKI in the context of MI. Finally, the consequences in terms of healthcare burden and economic consequences. In the left side of this panel, you can see patients that had acute kidney injury not requiring dialysis. There is definitely an increase in cost related to a prolonged hospitalization in this patient. And also on the right side of the panel are patients that developed dialysis. And you can see there is also an increase in the length of a stay in these patients and this convey higher costs. Of course, the y-axis is much higher in the right side because those are patients that require dialysis and therefore more resources. But bottom line, AKI events, because are very frequent, as I said, 20% of all hospitalized patients convey increasing costs in the healthcare system that can go up to $24 billion. And patients that require dialysis, a smaller subset, still increase healthcare costs about $5 billion in stressing the healthcare system. So with that, I wanted to convey the relevance of AKI in terms of frequency, in terms of complications, and the burden not only at the patient level in quality of life, but also in the healthcare system in terms of economic stress. Now what can we do about AKI to provide better care? And one of these things is perhaps it's time to have better tools to reclassify AKI, and at the same time have better recognition of certain sub-phenotypes that can help potentiate novel interventions, including therapeutics. Now when we talk about AKI reclassification, we should recognize that we can reclassify a patient before the event occurs, and even before the stressor appears, to prevent AKI as the goal. We can also reclassify a patient early during the course of AKI to prevent the progression and or complications associated with AKI. And we can reclassify a patient after an AKI event to prevent recurrence and other complications like re-hospitalizations, cardiovascular events, and the transition to chronic kidney disease. This is a very famous diagram that is in the KDGO guidelines that kind of have the spectrum of a kidney from being normal in a nice color, how this kidney can deteriorate according to different stressors in an acute, super acute phase. So if you see here carefully, ideally if we can have a pre-AKI reclassification tool, we can identify how the kidney responds to stress. And the kidney responds to stress with two specific events. Like first, the kidney needs to recognize what the capacity and the reserve to respond to these stressors, and then needs to show how it can adapt to the stressor. So with these, there are certain biomarkers that are very good in kidney health for identification of kidney stress. The problem is what we do when we identify kidney stress, and how timely we can intervene to really have an effect on mitigating the AKI risk in these patients. That's more the challenge, more in the implementation of these tools. Now of course, in the ideal scenario, we want to have tools including biomarkers or including clinical data that provide some type of risk classification score that give us a window early to be able to intervene effectively and mitigate the progression. So that's the problem that sometimes we have with implementing biomarkers at the bedside. But the conceptual part here is that we want to identify this kidney stress and the adaptation of the kidney early on before creatinine goes up and the GFR deteriorates. Now this is an example of some tools, right? So this is, you have heard this concept of renal angina index in the pediatric literature. They use risk classification tool based on clinical data that can be obtained from the electronic health records. But in this study in adults, we wanted to create a modification of this renal angina index to see if we can identify on patients in the first 24 hours in the ICU the risk of developing severe AKI in the subsequent two to seven days during the ICU stay. So with this score, the clinical variables that were included were diabetes, sepsis at the time of ICU admission, the presence of mechanical ventilation or press or inotropes, and also the change in creatinine in that first 24 hours. So if you have a patient already showing elevation in creatinine, then definitely the score will be higher. And importantly, the fluid overload percentage on these patients. So what's the level of change in weight in these first 24 hours in response to fluid resuscitation in a lot of these critically ill patients? So with this score, the performance as you can see was modest to good that the AUCs were 0.78 to 0.76, but definitely much better than an AUC of creatinine that is around 0.6 alone. So the concept here is that using creatinine alone to risk classify a patient is probably insufficient. If you create tools where you classify creatinine changes in addition to clinical parameters, you can have much better performance to identify those patients at high risk of developing severe AKI in the subsequent days in the ICU. Now when we try to create risk classification tools, we want to have something that is practical, right? A clinician want to have a tool that can help identify quickly which patients we can continue to monitor with no major intervention, like low risk of AKI, patients that have high risk of AKI in which we should intervene very proactively, and how we intervene. Trying to follow very basic KDGO recommendations about mitigation of risk, monitoring of not only creatinine but urinary output, minimizing nephrotoxin exposure, trying to avoid hyperglycemia, optimize hemodynamics, stop certain drugs that can stress the kidney further, or preclude the ability of the kidney to respond to stress, optimize fluid management. Now these interventions have shown being efficient in the right population at the right time, and I'll show you an example later. But also, a risk classification tool, when I identify those patients in the gray zone, in the intermediate risk zone, that perhaps you need to do an extra step of measuring a biomarker or trying to do a stress test to that kidney, like furosemide stress test, where you can risk classify low risk or high risk, and determine, move that patient outside that gray zone. So this is conceptually how a risk classification tool may be effective at the bedside. But what we put in the tool, it's important, but most important is how we test the tool to see if we can implement an intervention effectively. So when we talk about AKI, we have susceptibilities and exposures. Susceptibilities on the left side, some of them are non-modifiable, older age, history of chronic kidney disease, we cannot modify. But modifiable susceptibilities are a lot of the time exposure to certain medications, that if we effectively adjust the dose or stop, if the medication is no longer needed or perhaps not that relevant at that moment, the kidney can have better capacity to respond to stress. And then the exposures, we have planned exposures that we can control, and those exposures are sometimes surgeries or interventions, procedures, and unplanned exposures, when the patients get infected, when the patients had episodes that predispose them to be hypovolemic. So those are unplanned exposures that we can reverse with effective intervention sometimes. Now today, if we see how we practice, we do our evaluation at the bedside, we use our tools in the electronic health record to try to identify certain patterns in the patient, and then we try to determine with that what to do with the patient. And in this example of, let's say, what to do with the patient after AKI event has occurred, and we are determining what should we do with this patient. Should we refer to nephrology, outpatient, should we schedule a visit in the next 20 days or 28 days or a couple of months after discharge? So we basically use a lot the history of the AKI, the baseline status of chronic kidney disease, and some of these patients already followed by nephrology, and the level of AKI recovery, right? So a good clinician will use all this information and try with this to determine, yes, this patient probably should need a close follow-up after discharge to reassess among many things the level of kidney recovery. But imagine if we have a risk classification tool that can tell you this is a patient at high or intermediate risk of developing complications after AKI where you not only need to see this patient, but you need to see this patient early after discharge to prevent re-hospitalizations. So that's the idea, right? How the risk classification tool can help also for the post-AKI management of these patients. As I told you, only less than 10% of AKI survivors receive appropriate care after discharge. So there's definitely an area of improvement, and this is an area of improvement with collaboration with the intensivist because in the post-discharge critical care clinics, all of these patients are being evaluated. So this orchestration between nephrology and critical care is very relevant in the early post-discharge time. Now the problem with risk classification tools at the moment is that we have access to the data, but importantly, we need to create pipelines of this data to create tools that are sustainable and can be implemented in a wide scale. Now this is an example of another risk classification tool. This is a study that our group worked for identification of major adverse kidney events after AKI. So major adverse kidney events is an epidemiological outcome that includes a composite of mortality, performance on dialysis, and also level of kidney recovery in reference to baseline. Typically, and what we use here was if the patient had 50% decrease in kidney function in reference to baseline, that will be considered a major adverse kidney event. We evaluate these events at 90 days after the discharge from the hospital where they had the episode of acute kidney injury. So we labor EHR data. A lot of parameters, including characteristics of the patient, hemodynamic parameters of the patient during their stay, and also the laboratory data on these patients. And as you can see, the mortality in the hospital in patients, we have two cohorts, one in the University of Kentucky, one in the University of Texas Southwestern in Dallas. The mortality in these patients in the hospital fluctuated between 10% and 20% overall. The frequency of make events in the first 90 days was around 30% in both sides. So one out of three patients developed major adverse kidney events after an AKI event. That's very relevant, right? So what to do with these one in three patients is what we need to work on. These are the one in three patients that perhaps in some of them we can mitigate some of these episodes. Now we use, of course, machine learning and some sophisticated algorithms that the computer scientists now like to use, and every time they keep improving. But to identify with different classification tools, going from logistic regression to XGBoost, the top features in this clinical data that can classify a patient for the risk of developing major adverse kidney events, we selected at the end 15 and 14 features for different outcomes, mortality and make, and we create the interpretation of the model with some visualization. Just for you to have an idea which of these parameters came in this tool, for the outcome of major adverse kidney events, and on top are the most prominent features that were important for the model, and then on the colors, on the X-axis, you can see was the directionality of that parameter with the outcome. So for example, LASK ADEGO. LASK ADEGO means that what was the severity of the patient in terms of AKI during the window of observation that was the first three days during the ICU admission, and the higher the level of severity of AKI, of course, was very important for the model for the risk of major adverse kidney events of 90 days, not surprising. Importantly also urinary output was that the lower the urinary output, the higher the risk of the patient to develop major adverse kidney events based on the model, and so on, right? So the purpose here is not to say these are the variables you need to focus on, this is just to understand how you can interpret a model after you apply certain machine learning tools for classifying certain parameters. But the important thing is what's the performance of this type of models. The performance, again, is modest to good. That means that the AUCs were around 0.79, 0.75 in the external validation to have a prediction of the major adverse kidney event outcome. So this is a tool that is good enough to identify certain patients, so not to miss patients at high risk, but perhaps it's going to have some false positives too in the classification, and this is where we need to develop implementation science steps to determine what to do and how not to burden or stress too much the provider that is going to see more patients than often. And of course you can create certain ways to apply the tools and input online, so to create the classification of these patients, and this is available. You can play around with it if you want, and some people told me how to use it, and this is by no means a finalized tool, but this is just an example of how we can start reclassifying certain patients. Now of course in this context the role of artificial intelligence is very relevant, and what can artificial intelligence do to help in our day-to-day operations? If you evaluate how we do things right now, we do it a little bit sporadic, right? Because we are busy and sometimes we need to interrupt our activities because we have an acuity that we need to address, or we are interrupted for different things, so our evaluation is typically sporadic. Sometimes we have incomplete data because either tests are in progress or the test has not happened yet, and certainly our workflow is not integrated and we just need to come back and reassess continuously according to the new data that's being generated. So using all these with our clinical evaluation, our EHR tools, we try to make decisions. But imagine if we have tools that can help us integrate this data, and not only the data that we see in the electronic health record, but the data of the imaging test, the data of the monitors, the data of the devices, ventilator, dialysis device, et cetera. And these AI tools can help creating this data integration and data visualization and perhaps even some data classification to, as I mentioned before, identify patients at high risk versus low risk and create this process more dynamic. So this process is happening continuously while you are doing anything else. And it has this capacity to leverage on multi-modal data and can be integrated in your workflow. Why? Because when you are rounding on that patient, you can have a tool in your hands that is doing all this integration and risk classification so you can make the decision with all the data available at the moment that is already classified. So that will be what we call augmented clinical decision capacity for a clinician. Is it too utopic to think about this? Not at all. But the problem that we have with this is the need for external validation, implementation science with these tools. The tools exist. The tools can be refined. But we are still in infancy steps of implementation of how to use them. And this is an example, right? So this is a patient that you see very often. This is a patient in the ICU that requires CRRT for the support of AKI. You can see that if you apply an AI tool to predict, for example, mortality, the AI tool is able to detect the risk of mortality in a window of 24 hours before it actually happened. So this can be applied. Can we do something? That's a different question, right? If I give you your patient has a 90% risk of mortality in the next 24, 48 hours, are you going to be able to do something? I will argue yes, because sometimes doing something is a stop doing advanced support. And that liberates a lot of stress from the patients. Some families can benefit from you telling them the risk is too high. Perhaps this is time we can stop CRRT. And some of them will be agreeable to that. And that itself is doing a lot for this patient. And of course, the tools can be refined. The risk is very dynamic. As you can see, this is real-time examples of patients that are surviving in the ICU and they will end surviving the ICU process. And you can see how the tool reclassify risk of mortality going down as time goes on. And the opposite for patients that die in the ICU on CRRT, the tool reclassify this risk going up as expected. Again, AUCs are modest to good. This means that this is not the tool that is going to mandate what you do. It's going to be just some type of support for you to make the right decision for the patient, hopefully at the right time. As I said, the problem with this is data harmonization, the use of deep learning algorithms that are continuously being modified and improved. But the importance of addressing biases on these tools, and not all populations are represented, it's very important for the field and also the implementation testing, as I mentioned. In the last few minutes, I'm going to talk about sulfenotyping. Sulfenotyping are two concepts here. So sulfenotyping can identify a subset of patients, in this case with AKI, at high risk for an outcome. And this will be the context of prognostic enrichment for a patient. And then sulfenotyping can also help to identify a subset of patients likely to respond to a therapy due to a common underlying biology. And this is the predictive enrichment. With the prognosis, you can have interventions, right? You can tailor certain interventions to minimize risk. With the predictive enrichment, you can identify sulfenotypes that may be responsive to specific therapies. And I think this is very important for industry to be able to have tools, not only that we can use for prognostic enrichment that can enrich a clinical trial, but also for predictive enrichment to understand how the patient can respond to specific interventions, specific therapeutic angles. Of course, clinically, there are many phenotypes of AKI. This is examples. For example, in letter A, you can have a patient with hypovolemia, very clearly, where kidney function improves after administration of IV fluids. But you can also have, in letter C, a patient with hepatorenal syndrome, where the creatinine will remain elevated according to the context of the hepatorenal syndrome and according also to the prognosis and the liver transplant candidacy. The other concept here is that sometimes, when we evaluate independently sulfenotypes of AKI, we come to these patterns, right, patterns of biomarkers reflecting inflammation and endothelial injury. And this has been reproduced in certain studies. Of course, these studies are limited by the data that was available. But one important aspect that always captured my attention is that a lot of the time when you do sophisticated analysis to sulfenotype AKI based on biological patterns, inflammation, endothelial injury comes up. And these are very similar to the sulfenotypes of hyperinflammation that have been associated with increased risk of adverse outcomes like mortality in ARDS. So this is relevant for this session, an organ crosstalk. Because during injury and repair, perhaps, we can no longer see the organs in an isolated fashion. We need to start to see the organs in an orchestrated fashion. Because perhaps, we need to have a phenotype of high risk of organ injury globally. And the intervention can not necessarily target a specific organ, but can target multi-organs, right? So this is a concept that we need to start thinking about because we are all coming back to this identification of these patterns. So how we treat? Should we treat AKI alone or should we treat AKI and ARDS together, right? Perhaps we will need to focus our treatments more multi-organ in the near future. Of course, phenotypes also apply for the recovery capacity of the kidneys. Typically a kidney at risk of recurrent AKI is a patient that have hypertension, microalbuminuria, and biologically hyperfiltration patterns, tubular dysfunction if we measure certain biomarkers, and decreased renal reserve if we have the chance to measure it. Now patients that progress to chronic kidney disease are patients that typically have already underlined CKD, and if we evaluate those kidneys, there is a certain degree of fibrosis and small kidneys overall. So there are patterns that we can start identifying to predict this risk. And of course, the patterns of recovery, as I said, that have very clear relationship with the outcomes have also very importantly been linked to the ability of the kidney to recover. And this is a nice study on the right that evaluated the duration of AKI and also the levels of relapse and recurrence and how that relate to survival of kidney function in the follow-up in the next year was very clear in a dose response effect. So there are many inputs on the phenotyping, and this is one of my last slides here. This is just to summarize that phenotype AKI in a patient is not only the clinical characteristics we see in the EHR or we evaluate when we examine a patient, but also the pathobiological signals that includes biomarkers as well, the response to treatment, and the genetic predisposition of these patients. And now if we go farther, social determinants of health are also relevant for response to acute illnesses, particularly in critical care patients, and important definition of which are the outcomes that we are prioritizing for phenotyping and the patient-centered outcomes need to be included because, as I told you, these patients have a lot of debility after discharge. Now a couple of studies I'm going to show you quickly here. The first one is when I said that there is examples of successful implementation of KDGO guidelines after AKI, this is one good example. The Privaki trial, this is the original one. It was replicated multi-center, but in this study they evaluated cardiac surgery patients at high risk of developing AKI. How they identified the high-risk patients? Measuring a biomarker that is the cell cycle arrest biomarker, the combination of those two biomarkers you see on the bottom there. If these biomarkers tested positive, these patients were enrolled in this clinical trial, were randomized to usual care versus an intervention with dedicated team applying the KDGO bundle. All the recommendations I mentioned before. You can see that they are doing this in a timely fashion at the right patient population. They decrease the incidence of AKI postoperatively from 70% to 55% and severe episodes of AKI from 44% to 30%. This is very important. So it can work if you apply to the right patient at the right time. Also there's another example of implementation science. Mitigation of nephrotoxicity in the children population in the ICU. This is the NINJA program. This program, it was very simple, initially developed in Cincinnati children, where they tried to identify patients at high risk of nephrotoxic AKI if they were exposed to three nephrotoxic agents or if they were exposed to IV aminoglycosides for more than three consecutive days. They created an implementation science study where a dedicated team identified these patients, notified the primary teams about the risk, tried to mitigate the exposure to nephrotoxins moving forward. They successfully implemented in nine centers in this study and this is growing even farther. It has shown that significantly can decrease the risk of nephrotoxic AKI in children. Very interesting study here. This is a clinical decision support tool to give you insights about which patient can benefit from renal replacement therapy for supporting AKI at that particular moment in a more organized fashion with CDSS or a standard of care. This is not a clinical trial, it's a prospective observation with two arms. One of the outcomes that was improved that I always was intrigued was that when you provide the clinicians a summary of recommendations for initiation of RRT, they decrease statistically significantly on the right side of the panel the number of patients that died in the first 48 hours where the treatment is perceived as futile. So this is very important because if you see the epidemiology of CRRT in the ICU, you will see that there is about 10 to 15 percent of patients that die in the first 48 hours. Yes, some of them were very sick and we try our best, but others perhaps we offer a treatment when it was already futile. We can improve that with clinical decision support. And then, of course, treatment response, right? Treatment response, this is a slide I have because when we were in the times of COVID, very aggressively used different type of filters, and this is a filter we used that have enhanced hemoabsorption in CRRT, tried to remove extra inflammatory proteins. Now as you can see, one of the parameters in the machine, this is the machine on the y-axis is the transmembrane pressure. Of course, if you have a filter that has enhanced hemoabsorption, the pressure on the filter will be higher because it's being saturated faster. Now this is very clearly seen in the curves here, one filter versus the other, and this is a group of about 85 patients. But if you evaluate individually the curve on the top, the curve of the filter that has enhanced hemoabsorption, you can see that the response is very heterogeneous. This is exactly what happens when we give a drug to a patient with AKI. Some of them may respond differently, but this is what we need to move our field towards. Treatment-affected heterogeneity assessment and the phenotyping can help for this. Because if you design this study using, let's say, testing this filter, you want to provide this treatment to the patients on the top, but not to the patient on the bottom. Because the patient on the bottom, for a reason, are not saturating the filter. Perhaps they don't have enough inflammation to saturate the filter, or are not the right patient population to receive the treatment. But the patients on the top, they do saturate the filter. So if you want to test the concept of removing these inflammatory proteins, you will do it on them to see if it works or not. So this is the concept, of course, of adaptive trial design and treatment-affected heterogeneity assessment. So I hope I provide some novel concepts on risk classification, subphenotyping, and also some examples of implementation of these tools in the near future. Thank you very much for your attention.

acute kidney injury

complications

patient outcomes

healthcare costs

post-discharge care

artificial intelligence

risk classification