We're going to close our triad of debriefs with an overview of clinical hepatology. And I think, again, the most appropriate person for this, I think by far, would be Dr. Michael Charlton, who is Professor of Medicine, Director of the Center for Liver Diseases, and Co-Director of the Transplant Institute at the University of Chicago. Michael has been a major contributor to our understanding of NAFLD, viral hepatitis, and transplantation. So, again, difficult to imagine someone more fitting to cover this topic. Michael? Well, thank you. And thank you for, and congratulations for making it to the final official presentation of the liver meeting. It's been the most balanced liver meeting that I can remember for some time in terms of the diversity of subjects and high-impact presentations. There were 2,427 abstracts. The number that were oral was in the hundreds. I recognized very early it would be impossible to do a good job as an individual to pick out the highest-impact ones, so I didn't try. I recruited these three, who are some of our liver fellows, Thomas Cotter, Jenny Wang, and Bea Pierfidit. And they went through all of the abstracts. They presented them to each other and to myself to pick the ones that had the highest impact, and then we selected after discussion and did a tremendous job. They've been at all of your presentations. In one session yesterday, they exchanged over 40 messages. So this year, we're going to divide it in this fashion. So we will present data on preclinical alcohol, autoimmune and cholestatic, cirrhosis and portal hypertension, neoplasia, and transplantation. These are my disclosures. And we'll start with preclinical. The next series includes one part of one slide that was shown by Elizabeth Corey in the previous presentation. I decided to include it here because I feel that the impact of this work is really substantial and beyond fatty liver disease. And this is the machine learning models accurately interpreting liver histology in patients with nonalcoholic state of hepatitis. This work is the brainchild of Andrew Beck from Harvard. The problem that was addressed is that primary endpoints assessing efficacy of new therapies for NASH are histological. All of the phase three studies have as their primary endpoints one or more of the components of NASH histologically. The existing scoring systems are fairly well described. They have moderate to fair reproducibility. The disconcordance rate for the various components of the NASH score, for example, vary on average by about 30%. So different pathologists will go up or down by one stage or grade about 30% of the time. Andrew Beck and his team had hypothesized that a machine learning approach could be utilized to train models to accurately interpret the histology. And they took screening samples, as you heard from the stellar force studies. These are across the spectrum of liver disease, H and E, and trichrome. They had 75 pathologists make over 1,000 annotations each identifying some of the components of steatosis, ballooning, fibrosis, et cetera. And the software would gradually pick up on how the pixels came together to make an annotation of, say, ballooning. It was then pseudo-colorized, and the software was then able to quantifiably measure these endpoints. So this is an example of an H and E slide from a patient with NASH in the stellar force study screening process. And this is the pseudo-colorized version of that. Red is ballooning. Inflammation is shown in purple. This is two examples. The graph on the left-hand side of the slide shows the model score generated by the machine learning tool and the consensus grade by the pathologists shown on the X axis. And these H and E stains that you see were both by consensus graded as one. And you can see the difference in the abundance of steatosis between the two. The model score was able to quantifiably distinguish these, the bottom one being 3% using the model score and 19.9% in the upper score. And remember that the grading by pathologists was the same by consensus. This is another example with ballooning. You can see the H and E on the left. Both were graded as grade two by consensus by the model score using a machine learning 2.8% versus 8%. And the ballooning is shaded in red in the pseudo-colorized version of this. And this is an example of the fibrosis. Now, the fibrosis is key here. If you look at the fibrosis that was identified by the machine learning in the far right-hand slide here, it doesn't just pick up ISHAC F6. It was also able to pick up something that we all see when we gaze down the microscope, that different parts of the biopsy look different in terms of fibrosis stage. So some parts look like F0, some F1, and maybe some others F5, F6. The machine learning tool quantifies the proportion of the biopsy that is represented by these different ISHAC stages. This has profound implications for endpoints that are used to assess the approvability of therapeutic agents. One of the more impressive parts of this was the investigators were able to, having ascertained endpoints to see whether the machine learning tool could identify histological features that would predict clinical endpoints. And it found that the ratio of portal inflammation to lobular inflammation, if that score was greater than 40, all the clinical events occurred in that group. There were almost no clinical events in the less than 40 group. So even developing new indices to help us discern likelihood of clinical events from histology. So I think the implications for this are that machine learning tools can automatically identify and quantify key histologic features. There are already discussions with the Food and Drug Administration about how and whether to incorporate these into the analysis of endpoints for phase three studies that are currently underway. And it reveals heterogeneity within traditional grading systems. I don't doubt that we will be seeing this type of technology used not just in NASH but in other aspects of liver disease. I would love to see it in something like grading, steatosis, and in potential liver donors. Okay, moving on to alcohol. So this was a granular site of GCSF to treat the cutaneous chronic liver failure interim analysis of the first randomized European study, Engelman, first author. Now as everyone in the audience is aware, there have been a host of studies reporting the beneficial effects of GCSF in treating ACLF. Most of them are single-center or few-center studies, relatively small in size. Alcoholic hepatitis is a common cause of ACLF. In this study, alcoholic hepatitis represented about half of the cases. This was their study design, GCSF administered once patients met ACLF, EF-CLF criteria, and they had initiation of standard medical therapy or SMT with or without GCSF. This was randomized on a one-to-one basis. These are the findings. The superior outcomes were achieved in patients who were not randomized to GCSF. So standard medical therapy was superior to GCSF plus standard medical therapy. I would say, undoubtedly, a surprise. The number of patients who deceased during the study period was about 60%, really confirming the degree of sickness of these patients. I think we could conclude from this that GCSF has no therapeutic effect in patients with ACLF and might be deleterious in subgroups with ACLF. The outcome differences in previously published studies might be due to a type 1 error and related to a low sample size. Another feature of the many alcohol abstracts, so I picked one that was a Venn diagram overlapped with liver transplantation. This is liver transplantation for alcoholic hepatitis in the United States. Excellent outcomes with temporal and geographic variation. The problem is there's a growing sense in a number of reports that there's an increasing frequency of liver transplantation for alcoholic hepatitis. I don't think anyone doubts it. Liver transplant centers like to transplant livers, and with the void or the vacuum created by a declining number of hepatitis C patients, there's a capacity to do an alternative group, and alcoholic hepatitis seems to be filling some of that vacuum. The CDC estimates there are 22,200 deaths related to, or liver deaths related to alcoholic or alcohol use disorder in the United States. A good but unknown portion of these are related to alcoholic hepatitis. The frequency of liver transplantation for alcoholic hepatitis is increasing outcomes and thereby utility, however, and inter-regional variants are not well described. So one of the first findings by Carter et al. was that the rate of increase, the rate of transplantation for alcoholic hepatitis has increased five-fold in a four-year, five-year period. The total number, however, is still only 120 per year in the most recent year of note, which is 2018. There was an eight-fold difference between UNOS regions. So if you look at the regions that had 4% or greater of their transplants were for alcoholic hepatitis, they were all in the northeast. If you look at the regions with the lowest rates, they were generally where there is higher access to organs in the southeast and in the midwest. You might hypothesize that this is because there is a higher burden of alcoholic hepatitis in the northeast. So we pulled data from the CDC, and you can see that the northeast actually has the lowest rate of binge drinking of the centers that were reported in this analysis. So it's not on a basis of increased need. In terms of outcomes, patients with alcoholic hepatitis are shown in the highest line here for the Kaplan-Meier survival curve. The survival, and this is for GRAS survival, was as high or higher than any other indication at all of the measured time points, achieving 88% survival at four years and 84% at five years. So excellent overall outcomes. So the implications of this are that liver transplantation for alcoholic hepatitis is increasing rapidly, but with excellent medium-term outcomes. Utility of transplanting for alcoholic hepatitis is superior or equivalent to all other indications. Realizing that we transplant 120 patients per year, 22,000 patients dying, we are probably too selective in identifying patients who can benefit from this procedure. The practice is not standard, raising equity concerns and highlighting the need to optimize selection criteria. Moving on to autoimmune and cholestatic diseases. Now this abstract sort of gave the game away, but their title, Bezafibrate, is more effective than placebo in pruritus of chronic cholestasis, the Fitch study. The problem, however, of course, is that pruritus is a vexing clinical issue. It can be highly distressing for patients, and it's a common feature of cholestatic liver diseases. The therapies have variable, typically poor efficacy. It's a genuinely miserably making condition. I had a patient tell me that they felt like the subject of Edvard Munch's Scream painting. So this was a study using Bezafibrate, which is a PPAR agonist, and it apparently decreases the production of something referred to as biliary itch factor. This study was to assess the effect of Bezafibrate with a 400 milligram once-daily dose on pruritus in patients with PSC, PBC, or secondary sclerosing cholangitis. It was a multi-center investigator-initiated RCT between the Netherlands and Spain. The primary endpoint was a 50% reduction in a visual analog score for pruritus. The frequency that the endpoint was met was four-fold higher in the treatment group, the Bezafibrate group, than it was in the placebo group. On the right-hand side, you can see the actual mean VAS score in the two treatment groups, the placebo and the Bezafibrate group. So this is, I think, a very encouraging study for this very difficult clinical issue. I think the implication is that Bezafibrate is, as of this meeting, superior to placebo in treatments of pruritus and cholangiopathies and should be considered as first-line treatment for pruritus in PSC and PBC. I can change the picture here. Moving on to cirrhosis and complications of portal hypertension. So this is a study describing Rifaximin for the prevention of hepatic encephalopathy in patients treated by TIPS, a multi-center RCT. There are very few things in guidelines that, across multiple societies, produce exactly the same verbiage. But hepatic encephalopathy, prophylaxis at the time of TIPS, is one of those. So this is a quote from ASLD, EASL, and the French Society for the Study of Liver Disease, AFEF. And they say, pharmacological prophylaxis is not recommended for the prevention of post-TIPS HE. Grade II. Strong agreement. With absolute disdain for these guidance statements, a French group went ahead and studied it for themselves. They performed a randomized, double-blind, placebo-controlled, two-arm study with rifaximin 600mg twice daily or placebo, started 15 days before the TIPS was performed and for six months after the procedure. And their primary endpoint was a treatable hepatic or clinically significant hepatic encephalopathy. These are their primary findings and this is for their primary endpoint, which is hepatic encephalopathy that was included. Even simple asterixis. And if they subtracted out the patients who just had simple asterixis, the magnitude of the effect was greater. It was two-fold the untreated group. So I think they have added through a prospective multicenter RCT new data that will, I think, lead to the change of guidelines for at least the three societies that I referenced. So rifaximin is effective in preventing post-TIPS encephalopathy in selected patients. So I congratulate the authors on this Joan of Arc-esque tilt at a set of established guidelines. Moving on to efficacy of carvedilol for endoscopic varicea ligation. The management to prevent first variceal bleed for patients with varices, high-risk varices, is difficult. We have two primary choices, beta blockade versus endoscopic variceal ligation. Carvedilol is fairly good at diminishing portal blood pressure. It is, however, problematic in the sense that it can exacerbate some aspects of portal hypertension, including severe ascites and produce symptoms including some lightheadedness, et cetera. And EVL, or endoscopic variceal ligation, increases the incidence of post-bleeding, perhaps. So the objective of this study was to evaluate the efficacy of carvedilol, EVL, or both in prevention of first variceal bleed in advanced cirrhosis. And the high-risk varices were defined by large varices or small varices with red color signs. This is their primary finding. There was a five-fold difference between the treatment groups in this prospective randomized controlled study. So the most effective was carvedilol and endoscopic variceal ligation. The least effective was carvedilol alone. And the difference between these treatment arms was, as you can see, clinically significant. So the obvious impact or implication of this is that the combination of carvedilol and endoscopic variceal ligation is more effective than either therapy alone for prevention of first variceal bleed in patients with large varices or small varices with red color signs. Moving on to the confirmed study, this is a North American randomized controlled trial of terlipressin plus albumin for the treatment of hepatorenal syndrome type 1. Now, hepatorenal syndrome is a near-ubiquitously fatal condition if patients are not able to undergo liver transplantation. There is no approved pharmacological therapy specific for hepatorenal syndrome in North America. This was a 300-patient randomized controlled study where they took terlipressin. Now, terlipressin has the relative advantage of being relatively more specific for the V1 receptor than, say, vasopressin. They had a 90-day follow-up. Patients were randomly assigned on a two-to-one basis, so two to terlipressin versus one to placebo. And they received albumin and standard of care in both arms. These are their primary endpoint finding. And so the confirmed study looked at verified hepatorenal syndrome reversal. And there was about a two-fold benefit, so 29% versus 15.8% in verified hepatorenal syndrome reversal in this study. These data actually vary similar to the five other prospective randomized controlled studies with terlipressin. The reason why we haven't had this drug approved in the United States for this indication has been that the number of mortality or the survival rate was actually not improved by the addition of terlipressin. In this study, they observed exactly the same phenomena, where despite the improvements in hepatorenal syndrome and a superior frequency of verified hepatorenal syndrome improvements, the overall survival was numerically less in the terlipressin arm than in the placebo group. And transplant-free survival completely superimposable Kaplan-Meier curves. So unfortunately, the confirmed study will join a list of studies whose names were optimistic, I think, in comparison to the outcomes. Moving on to another study that was presumably being carried on at the same time as the confirmed study by a group in India. Jayachandran et al., perhaps in anticipation of these results, looked at a combination of terlipressin and noradrenaline in non-responders to terlipressin in hepatorenal syndrome. This is a much smaller study. In this case, they evaluated 60 patients and they gave them 48 hours to show some response to terlipressin therapy for hepatorenal syndrome type 1. The methods were open-label, two-armed study, N of 60. And the primary outcome was assessed in 15 days as opposed to the 90 days for the confirmed study. And I don't show a p-value here because the numbers are small. I don't think that a p-value is helpful. And they're looking at percent survival. This isn't a response to hepatorenal syndrome. This is a percent survival. And they show at least a trend for improved survival in the combination study arm of this analysis. The control arm continued to receive terlipressin. So I think this at least opens the door to a follow-up study from the Herculean efforts of the confirmed study and the studies that preceded the confirmed study that perhaps because of the tolerability issues that go hand-in-hand with terlipressin, there are GI side effects, mesenteric ischemia is an issue, cardiovascular effects of administration of terlipressin, et cetera. Neuroadrenaline is a little bit milder in side effects and that was the hypothesis from Jai Chandran et al. And I think it will hopefully lead to a combination study that is larger, multi-sensor. So in combination, I think the implications of terlipressin and albumin are significantly better than albumin alone in achieving verified HRS1 reversal, but it does not confer transplant-free survival or overall survival advantage. But because of the potential for lower adverse effects, the combination of terlipressin and neuroadrenaline merit further investigation. Moving on to neoplasia. So this is again an overlap with a transplant study and looking at the U.S. multi-sensor analysis of 2,500 patients undergoing liver transplantation, a 10-year outcome assessing the role of downstaging within Milan criteria. As everyone in the room who manages patients with HCC knows, most patients with HCC present outside of Milan criteria. The UNOS Liver and Intestine Committee has done a very good job at developing criteria that may start outside of the Milan criteria, but then be downstaged to within the Milan criteria and still achieve or still be allowed exception points for organ allocation. And the upper limit that has been developed by the UNOS Committee has been one lesion greater than five, but less than or equal to eight, two to three lesions, at least one lesion greater than three, but none of them greater than five, and a total tumor diameter less than or equal to eight, or four or five tumors less than or equal to three centimeters, but not greater than eight. So the objective was to compare liver transplant recipients who were downstaged from within those criteria that I just mentioned to within the Milan criteria versus patients who were not downstaged. And this is a retrospective review of adult liver transplants undergoing liver transplant for HCC between 2001 and 2015 in five large centers. This was the disposition of patients. You can see the relatively large number of patients, over 2,000, were within Milan criteria. There were 113 that were not downstaged that were transplanted outside of Milan criteria, and there were 100 and something patients who were downstaged to within, sorry, 330 patients who were downstaged to within Milan criteria. The difference between groups was substantial. If you look at patients who were outside of the Milan criteria, their 10-year survival, 39%, compared to 52% for patients who were downstaged and 61% of 10 years of people who were within Milan criteria at the beginning. Having said that, if you look at the three-year survival time point, the survival for patients within Milan criteria and patients downstaged to within Milan criteria was essentially exactly the same. This is very encouraging. This panel here shows the tumor recurrence frequency, HCC recurrence, over the course of time. So Captain Myers, for the highest frequency of recurrence of 47% at 10 years of people who were transplanted outside of Milan criteria, 20% for people who were downstaged to within Milan, and 14% to people who were within Milan at the time of transplantation. These are very important findings, and I think they show us that the UNOS downstaging criteria for HCC are validated. Now having said that, I'm going to just discuss very briefly what I think is the most impactful, I think, not presented data of the year, which was presented in news release just a couple of weeks ago. And this is the recent success of endpoints of overall and progression-free survival for non-resectable HCC. So this would include, potentially, people who would undergo liver transplantation. And this was as first-line therapy in a 501-patient prospective multicenter randomized controlled trial of a tazolizumab, which is a PD-L1 antibody, and bevacizumab, the VEGF inhibitor. The actual results have not been disclosed. The company did disclose, however, that the overall survival analysis is still evolving. So I think that Captain Myers is probably continuing to separate. This will require us to rethink all that we do in terms of managing patients with HCC before liver transplantation, probably after liver transplantation as well. So what is the right time, for example, for PD-L1 to be administered before you can safely perform liver transplantation? The half-life of these drugs is about four weeks. There are several reports of rejection for people who have received checkpoint inhibitor therapy. So this really restarts the clock, I think, on un-resectable HCC. And finally, transplantation itself. So the donor quality score and prediction of graft failure after liver transplant is one of the abstracts I've chosen for this. And the problem that these authors addressed is that since the development of the original Donor Risk Index, or DRI, primarily generated at the University of California in San Francisco, recipient characteristics have changed substantially. Recipients are older, they're sicker, they have more comorbidities, and we've had the advent of direct-acting antivirals. The objective of this group was to develop and validate a new model to assess donor quality, looking at the new period since direct-acting virals. And so they looked at the performance of DRI in the DAA era and developed a novel donor quality, or DQ, score, validated it using adult recipients between 2013 and 2016, excluding patients with hepatitis C. This is the performance of the Donor Risk Index, the tool that we currently have and that is used to assign donor risk or risk of graft failure. And it generates between the highest quartile of risk and the lowest quartile of risk, only separates out by around 5%. In contrast, the novel DQ score, and these are the elements that go into it, there's only seven of them, they're fairly simple, so diabetes, GFR, and age, donor height, donation after cardiac death, cold ischemic time are the factors that go into this. This separates out the highest and the lowest quartiles by more than twice the degree that the DRI does. And actually it's more nuanced as well, in that things like cold ischemia time interacts importantly with DCD, but doesn't seem to be impactful for donors who are donating after brain death. And age was important for some indications, but not for others. So this is a really new and important development for understanding donor risk. So the implications of this are that DRI performance has decreased, the DQ score may help to optimize risk assessment for donor organs and allow for granular discussion with patients about predicted risk of graft failure. So I'll conclude once again by thanking the three, Jenny, Thomas, and Bea, for their tremendous work on this, and to congratulate everyone for making it to the end of what's been a tremendous LIBRE meeting. Thank you. Thank you so much. Thank you so much, Mike, for that beautiful talk, and I'd like to thank all of our presenters and all of our attendees for sticking around until the very end for what has really been an extraordinarily broad and deep meeting.

clinical hepatology

machine learning models

NASH

liver transplantation

pruritus

HCC