Good morning. It is such an honor to be here today. Here are my financial disclosures. I want to introduce you to Mr. Smith. Mr. Smith is a 56-year-old man with decompensated cirrhosis due to NASH who's admitted to your hospital with jaundice, worsening ascites, and altered mental status requiring intubation. He is hypotensive, requiring vasopressor support. His MELD score has risen from 9, which it was four months ago, to 32 on admission. And he now requires renal replacement therapy for acute kidney injury. His workup has revealed spontaneous bacterial peritonitis and E. coli bacteremia. Now, I think we can all look at this case and agree that at a minimum, Mr. Smith has presented to you today with acute hepatic decompensation. But increasingly, his clinical presentation is being recognized as a syndrome that is really distinct from acute hepatic decompensation and one that has more recently been termed acute on chronic liver failure. My objectives for today are to define acute on chronic liver failure, or ACLF, describe the conceptual underpinnings of the clinical prediction models that are available to assess severity and prognosis, and to apply clinical prediction models in a variety of clinical scenarios. So what is acute on chronic liver failure? Well, in the broadest of definitions, one that was developed by consensus, by experts within the World Gastroenterology Organization, acute on chronic liver failure is a syndrome in patients with chronic liver disease with or without previously diagnosed cirrhosis, characterized by acute hepatic decompensation, resulting in liver failure as evidenced by jaundice and INR prolongation, and one or more extrahepatic failures associated with increased mortality up to three months. Importantly, it is distinguishable from chronic liver disease without cirrhosis, compensated cirrhosis as well as acute decompensated cirrhosis that we traditionally think of in our patients. Now much of what we know about the clinical aspects of patients presenting with ACLF has been derived from three multi-center cohorts. The first is the Asian Pacific Association for the Study of the Liver, or APASL, ACLF Research Consortium, which has enrolled over 5,200 patients from 43 centers in 15 countries representing the Asia Pacific region. The European Association for the Study of the Liver, Chronic Liver Failure Consortium, or ESL-CLIF, that has enrolled over 1,300 patients from 29 liver units in eight countries within the European region, and the North American Association for the Study of Liver Diseases, or NACCELD, which represents over 2,600 patients in 14 centers within the U.S. and Canada. Now all of these three cohorts really have enrolled and represent the patients with underlying liver disease who really are the sickest of the patients we manage, but the Eastern cohort differs from the two Western cohorts in two important aspects. The first is the requirement for cirrhosis for enrollment into the cohort. The APASL cohort did not require patients to have underlying cirrhosis, and so in fact it did include patients who had acute alcoholic hepatitis without known advanced fibrosis, as well as patients presenting with acute hepatitis B reactivation without cirrhosis, whereas the Western cohorts did require all patients to have cirrhosis as an enrollment criterion. And the second difference between the Eastern and the Western cohorts is the primary driver of liver injury, where the APASL cohort, primarily the patients presented with a direct liver toxic injury, such as alcohol or acute viral hepatitis, drug-induced liver injury, or autoimmune hepatitis, whereas the Western cohorts, the primary driver of injury was non-liver. Many of them had infection, although it's worth noting that 40% did not have an identifiable precipitating event. Now from these three cohorts, three different clinical prediction models have been developed to predict outcomes in this acutely ill setting, and the links to the three calculators are shown here. Now I'm not gonna go into the specifics of how to calculate clinical prediction scores using each of these three calculators, because they're easily accessible online, but I do want to provide you with a broad overview of the clinical prediction model so that you can make the decision in your own clinical practice which clinical prediction model to use in which clinical scenario. The general underlying principle of the three clinical scoring systems is that the components represent dysfunction of multiple organ systems, whether it be the liver, renal, neurologic, circulatory, or respiratory systems. Now let's get a little bit more specific into the key components of each prediction model. Liver dysfunction, as represented by total bilirubin and INR, were components of the APOSL-ACLF scoring system and the ESL-CLIF scoring system, and then lactate was also included in the APOSL scoring system, whereas liver dysfunction, you'll note, was not a component of the actual scoring system in the NAC-CELD model. Kidney dysfunction was represented in all three clinical prediction models, ranging from subcategories of creatinine to simply the presence or absence of dialysis. Cerebral dysfunction is represented in all three clinical prediction models by hepatic encephalopathy grade in the APOSL and the ESL consortium models, and simply the presence or absence of severe hepatic encephalopathy in the NAC-CELD model. Circulatory dysfunction, as represented by mean arterial pressure or vasopressor use, was represented in the ESL and the NAC-CELD scoring systems, and not a component of the APOSL scoring system, although in the right clinical scenario, of course, lactate can represent circulatory dysfunction. And then you'll note that respiratory dysfunction was represented in the ESL consortium model based on oxygenation deficits, and in the NAC-CELD model, simply by the presence or absence of mechanical ventilation. Respiratory dysfunction was not a component of the APOSL scoring system. So now that the components of each of the models is represented here in front of you, I think you can see that actually, they're relatively similar with a few differences, such as respiratory dysfunction and liver dysfunction here. But I also hope that you can appreciate that perhaps these models don't actually represent three different models for the same patient, but perhaps represent models for patients within different places within the spectrum of the presentation of acute decompensation of cirrhosis. On the one end of the spectrum, the APOSL-ACLF research consortium score really represents the initial liver injury, as well as the sequelae that result from the direct liver injury. And then on the other end of the spectrum, the NAC-CELD scoring system represents the end organ manifestation that results from the systemic inflammatory response that occurs as a result of the liver injury in these patients. Select test characteristics are shown here. The APOSL-ACLF research consortium score ranges from five to 15, so it has a moderate range. And the Eazle-Cliff score has a very broad range from zero to 100, which is important if you're gonna be trying to assess change. Whereas the NAC-CELD scoring system is much simpler, but has a much narrower range of zero to four. All three scoring systems allow for categorization of patients within three to four classes of ACLF. And all three scoring systems have very good prognostic accuracy for the prediction of mortality within the short term, with C-statistics ranging from 0.78 to 0.85. But perhaps more importantly than the actual prognostic performance of these metrics, these metrics have been shown to predict mortality in these acutely ill patients, independent of MELD or even better than MELD. For example, here are two area under the receiver operating characteristics curves for the Cliff score here shown in the red line, and the Apozzle-ACLF score shown in the solid line here. And you can see that in both instances, the ACLF models predict mortality at 28 days better than our traditional metrics of liver disease severity, such as the MELD score. Really providing additional support for the construct that ACLF is a distinct clinical entity in our patients. So now that you are informed with this information about these three clinical prediction models, how can we apply this in our clinical practice? Well, selection of the prediction models really depends upon the scenario. I would recommend that if you have a patient who's presenting to you in the hospital without cirrhosis and with a liver-related, directly liver-related injury, such as acute viral hepatitis, alcoholic hepatitis, drug-induced liver injury, that the Apozzle-ACLF scoring system may represent the best prognostic marker for you. Whereas if your patient has known underlying cirrhosis or is presenting with a non-liver-related injury, such as infection or surgery, perhaps the easel scoring system or the NAC-CELD scoring system may be more appropriate for that clinical scenario. Now, what is your clinical need, though? If your clinical need is to assess response to therapy, identify patients for liver-supportive options, perhaps identify a window for transplant in these patients, the Apozzle scoring system and the easel scoring system with its broader range and more flexible and dynamic scoring ability perhaps may be more useful. Whereas if you are looking for a rapid bedside test that you can just calculate in your head, you don't need a calculator, and you want to just make rapid decisions about sort of which way to go with care, if you want a tool to help facilitate conversations about futility with patients and their family members, a simple, easy-to-calculate metric, such as the NAC-CELD score, may be more useful for you. So let's bring this information back to the bedside of Mr. Smith. If you'll remember, Mr. Smith was intubated. He was on vasopressors. His MELD score has risen acutely to 32. He's on renal replacement therapies. He is actively infected. Based on these clinical parameters, Mr. Smith achieves the highest level of scores in all three clinical prediction models and is in the highest ACLF grade, which gives him a predicted probability of death that is greater than 90% at one month and a near certain risk of death at one year in the absence of transplant. Now, perhaps some of you in this audience are thinking, Jen, I didn't need a clinical prediction model to tell me that a patient with decompensated cirrhosis who's coming in acutely ill with septic shock who's intubated in renal failure has a very low probability of walking out of this hospital alive in the absence of a new liver. And you're entirely right. I mean, in this clinical scenario, our clinical intuition really matches and is completely aligned with this patient's predicted prognosis as well as his observed outcome and reality. But here's the thing with clinical prediction models and why ACLF represents the perfect paradigm for how clinical prediction models can be used in our clinical practice. Clinical prediction models don't exist to help us in cases that are black or white. For example, when the total bilirubin, creatinine, and INR are on extreme ends of the spectrum, we don't need to pull out our MELD calculator to tell us if a patient is well or sick. Clinical prediction models exist to provide clarity in the gray. My clicker's not working. Thank you. They exist to provide clarity in the gray. When patients, as they so often do, present with normal function of one marker, mild dysfunction of another, severe dysfunction of a third, and then it becomes extremely useful to know where in the middle of the spectrum the patient lies, such as, is the patient a MELD 13, 19, 27, or 34? That makes a really big difference in whether we hospitalize the patient, do we send them to the ICU, how urgently do we respond, do we prepare the patient for transplant, and how do we communicate risk to the patient and the family members? And so similarly, in patients who present with acute and chronic liver failure with some cognitive dysfunction, at right now normal renal function, some coagulopathy, some respiratory dysfunction, it doesn't help us to know if the patient is ACLF one or zero or four, but it is actually very helpful to know where in the middle of the spectrum the patient lies. And these different ACLF grades really do stratify patients by their risk of mortality. You can see here very clearly by the survival, predicted probabilities of survival, that survival decreases dramatically, not subtly, but dramatically by ACLF grade, regardless of whether you are assessing mortality risk by the APOSL score, the CLIF-C score, or the Naxcelled score. So let's bring this back to Mr. Smith. At the current time, he has a greater than 90% predicted probability of death, but what if instead of requiring vasopressors, you actually resuscitated him with some albumin expansion? His risk of death decreases by a third. And now instead of talking about futility and how perhaps he's not gonna make it out of the hospital alive, we have now opened the door for the family members for that window of hope that maybe if he continues to get a little bit better, we should talk about transferring him to a liver transplant center. What if instead his MELD was not 32 but 25, or he did not need dialysis? His risk of mortality now decreases by another third. And instead of conversations about futility, we're gonna start talking about conversations of we need to try a short term of very aggressive interventions to try to bridge this patient either to response to our therapy or perhaps a bridge to transplantation. This is precision hepatology in clinical practice in 2019. It can be as sophisticated as the omics techniques and the personalized biomarker approaches that the distinguished speakers before me spoke about to you today. But it can be as simple and elegant as the clinical prediction models that I presented to you today that were derived entirely from clinical variables that you are already collecting in your clinical practice. And in this era of the electronic health record, we can now take these clinical variables, apply predictive analytics to now bring them into the electronic health record and make them available to you immediately so that we can integrate these objective data, more accurate prediction of risk into our clinical decision making and take them to the bedside, to the patient, to the family members to provide them with more accurate assessments to facilitate personalized discussions with the patients and their family members to provide care that is concordant with their own values and wishes. In summary, acute on chronic liver failure is a syndrome of acute hepatic decompensation in patients with chronic liver disease associated with very high short term mortality. Three clinical prediction models exist to assess severity and risk of death that are more similar than they are different and selection of the model depends upon the clinical scenario and clinical need. And systematic application of clinical prediction models to patients with chronic liver disease who are hospitalized with liver failure can improve the accuracy of risk prediction to refine our clinical decisions and facilitate discussions with our patients and their family members. Thank you.

acute on chronic liver failure

APASL-ACLF

ESL-CLIF

NACCELD

clinical prediction models