Impact of ammonia levels on outcome in clinically stable outpatients with advanced chronic liver disease

Background & Aims Ammonia levels predicted hospitalisation in a recent landmark study not accounting for portal hypertension and systemic inflammation severity. We investigated (i) the prognostic value of venous ammonia levels (outcome cohort) for liver-related outcomes while accounting for these factors and (ii) its correlation with key disease-driving mechanisms (biomarker cohort). Methods (i) The outcome cohort included 549 clinically stable outpatients with evidence of advanced chronic liver disease. (ii) The partly overlapping biomarker cohort comprised 193 individuals, recruited from the prospective Vienna Cirrhosis Study (VICIS: NCT03267615). Results (i) In the outcome cohort, ammonia increased across clinical stages as well as hepatic venous pressure gradient and United Network for Organ Sharing model for end-stage liver disease (2016) strata and were independently linked with diabetes. Ammonia was associated with liver-related death, even after multivariable adjustment (adjusted hazard ratio [aHR]: 1.05 [95% CI: 1.00–1.10]; p = 0.044). The recently proposed cut-off (≥1.4 × upper limit of normal) was independently predictive of hepatic decompensation (aHR: 2.08 [95% CI: 1.35–3.22]; p <0.001), non-elective liver-related hospitalisation (aHR: 1.86 [95% CI: 1.17–2.95]; p = 0.008), and – in those with decompensated advanced chronic liver disease – acute-on-chronic liver failure (aHR: 1.71 [95% CI: 1.05–2.80]; p = 0.031). (ii) Besides hepatic venous pressure gradient, venous ammonia was correlated with markers of endothelial dysfunction and liver fibrogenesis/matrix remodelling in the biomarker cohort. Conclusions Venous ammonia predicts hepatic decompensation, non-elective liver-related hospitalisation, acute-on-chronic liver failure, and liver-related death, independently of established prognostic indicators including C-reactive protein and hepatic venous pressure gradient. Although venous ammonia is linked with several key disease-driving mechanisms, its prognostic value is not explained by associated hepatic dysfunction, systemic inflammation, or portal hypertension severity, suggesting direct toxicity. Impact and implications A recent landmark study linked ammonia levels (a simple blood test) with hospitalisation/death in individuals with clinically stable cirrhosis. Our study extends the prognostic value of venous ammonia to other important liver-related complications. Although venous ammonia is linked with several key disease-driving mechanisms, they do not fully explain its prognostic value. This supports the concept of direct ammonia toxicity and ammonia-lowering drugs as disease-modifying treatment.


Introduction
Advanced chronic liver disease (ACLD) is a major source of morbidity and mortality worldwide, with non-alcoholic fatty liver disease (NAFLD) emerging as the predominant cause of ACLD in several regions. 1,2 The first development of hepatic decompensationmost commonly ascites, although hepatic encephalopathy (HE) is the predominant first event in NAFLD 2denotes a watershed moment in the natural history of ACLD, as it is accompanied by a substantial increase in long-term mortality. 3 Those with decompensated ACLD (dACLD) are at risk of acuteon-chronic liver failure (ACLF), which is defined by extrahepatic organ dysfunction (including acute encephalopathy) and high short-term mortality. 4 Portal hypertension, which is accompanied by portosystemic shunting, and bacterial translocation-induced systemic inflammation are considered as the main drivers of clinical deterioration. 5 In people with ACLD, hyperammonaemia is driven by microbiome changes and ammonia overproduction, 6,7 decreased hepatic/extrahepatic metabolic capacity (urea cycle and glutamine synthetase 8,9 ), and portosystemic shunting via collaterals. 10,11 The diagnostic utility of ammonia testing for HE is controversially discussed, as hyperammonaemia is not the only mechanism for HE development, and thus, people with HE may present with normal ammonia levels. 10 Nevertheless, in those with altered mental status, values within the normal range may question the diagnosis of HE. 12 Notably, experimental studies indicate ammonia toxicity beyond its role in HE (e.g. liver fibrogenesis and immune dysfunction 13,14 ). In a recent landmark study, Tranah et al. 15 evaluated the impact of ammonia on liver-related outcomes in clinically stable individuals with ACLD and values > − 1.4 × the upper limit of normal predicted liver-related events in stable outpatients with ACLD. However, it remains unclear whether this association is independent of portal hypertension/systemic inflammation, that is, well-established disease-driving mechanisms. Moreover, the findings of experimental studies, that is the link between hyperammonaemia and liver fibrogenesis, remains to be confirmed in humans to support the potential role of ammonia-lowering drugs as a disease-modifying treatment.
The objectives of our study were (i) to externally validate and extend previous findings on the prognostic value of venous plasma ammonia levels in a large, well-characterised cohort, while accounting for portal hypertension and systemic inflammation severity and (ii) to investigate the relationship between venous plasma ammonia and biomarkers of other diseasedriving mechanisms.

Study design and participants
We performed a retrospective, single-centre cohort study in individuals with ACLD who underwent hepatic venous pressure gradient (HVPG) measurement at the Vienna Hepatic Hemodynamic Lab (outcome cohort; Fig. S1). Those from the outcome cohort were included between Q2/04 and Q4/20. Inclusion criteria were (i) liver stiffness measurement > − 10 kPa and/or HVPG > − 6 mmHg, and (ii) availability of venous plasma ammonia levels. Furthermore, individuals were excluded if any of the following criteria were present: those with a history of orthotopic liver transplantation, any active extrahepatic malignancy, non-parenchymal liver disease, non-elective hospitalisation as a result of a liver-related complication at HVPG measurement or within 28 days before HVPG measurement, unsuccessful/unreliable HVPG measurement, bacterial infection, or missing information on important laboratory parameters and/or clinical follow-up. Recruitment and follow-up over the study period are depicted in Fig. S2.

HVPG measurement
Under local anaesthesia and ultrasound guidance, a catheter introducer sheath was inserted into the right internal jugular vein. 16 Subsequently, a hepatic vein was cannulated and the free and wedged hepatic venous pressures were obtained at least as triplicate measurements by a balloon catheter, 17 as recommended by Baveno VII. 18 Measurement of biomarkers Routine laboratory tests, venous plasma ammonia, and biomarkers (von Willebrand factor [vWF], procalcitonin [PCT], IL-6, enhanced liver fibrosis [ELF ® ] test, copeptin, renin, and bile acids [BAs]) were performed by the ISO-certified Department of Laboratory Medicine of the Medical University of Vienna using commercially available methods that are applied in clinical routine and blood samples obtained via a central venous line (i.e. the side port of the catheter introducer sheath) at the time of HVPG measurement. Venous plasma ammonia was sampled and rapidly transported on ice to the central laboratory. In line with the previous landmark study, 15 venous plasma ammonia levels were divided by the sex-specific upper limit of normal of our local laboratory (i.e. 60 mmol/L for males and 51 mmol/L for females).
Clinical stages of ACLD, definition of hepatic decompensation and of ACLF Participants were classified according to recently defined prognostic/clinical stages. The definition was adapted from D'Amico et al. 19 dACLD was defined by the presence or history of at least one decompensating event, that is ascites, variceal bleeding, or HE. ACLF was defined according to European Foundation for the Study of Chronic Liver Failure (EF-CLIF) criteria. 20 Statistical analysis All statistical analyses were performed using IBM SPSS Statistics 27 (IBM, New York, NY, USA), R 4.1.2 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria), or GraphPad Prism 8 (GraphPad Software, San Diego, CA, USA). Categorical variables were reported as absolute (n) and relative frequencies (%), whereas continuous variables as mean ± SD or median (interquartile range [IQR]), as appropriate. Student's t test was used for group comparisons of normally distributed variables and the Mann-Whitney U test for non-normally distributed variables. Group comparisons of categorical variables were performed using either X 2 or Fisher's exact test, as appropriate.
Univariable and multivariable linear regression analyses were applied to evaluate factors associated with ammonia.
Follow-up time was calculated as the time from HVPG measurement to the date of liver transplantation, death, or last followup at one of the hospitals of the Vienna hospital association by the reverse Kaplan-Meier method. Impact of venous plasma ammonia levels on liver-related outcomes was assessed using Cox regression and competing risk analyses considering the removal/ suppression of the primary aetiological factor (as defined by Baveno VII, 18 i.e. initiation of antiviral therapy/reported alcohol abstinence), liver transplantation, or non-liver-related death, as competing risks. Analyses were performed for hepatic decompensation/liver-related death, liver-related death, development of ACLF/requirement of liver transplantation/liver-related death, and non-elective liver-related hospitalisation/liver-related death as outcomes of interest. For the outcome development of ACLF/requirement of liver transplantation/liver-related deathin individuals who had already experienced hepatic decompensation at baseline (i.e. the main at-risk population)removal/suppression of the primary aetiological factor or non-liver-related death were considered as competing risks. For competing risk regression analyses, Fine and Gray competing risks regression models (cmprsk: subdistribution analysis of competing risks, https://CRAN.R-project.org/package=cmprsk) 21,22 were calculated. Univariable and multivariable Cox regression analyses were performed to evaluate parameters independently associated with the events of interest. In a first step, we included all parameters into univariable Cox regression models. Baseline characteristics which we considered of particular importance for the endpoint of interest (i.e. age, indicators of hepatic dysfunction, HVPG, and Creactive protein [CRP]) were further included into two separate multivariable models. The Child-Turcotte-Pugh (CTP) and United Network for Organ Sharing (UNOS) model for end-stage liver disease (MELD) (2016) scores have significant overlap in terms of included variables. Therefore, we generated separate models with either CTP or UNOS MELD (2016) scores.
Time-dependent area under the receiver operating characteristic curve (AUROC) analyses were performed and the Rpackage 'timeROC' was used to compare the prognostic performances for hepatic decompensation/liver-related death and liver-related death between established prognostic indicators (UNOS MELD [2016] score and HVPG) and ammonia (multiplicity-adjusted p values) over time.
Spearman's correlation analyses were conducted to investigate potential associations between ammonia and biomarkers in the biomarker cohort. A heatmap plot was used for graphical illustration of associations between ammonia and biomarkers.
The level of significance was set at a two-sided p value of <0.05.

Ethics
This study has been conducted in accordance with the principles of the Declaration of Helsinki and its amendments and has been approved by the local ethics committee (EK1531/2022 and EK1262/2017), which waived the requirement of written informed consent for the retrospective analysis of the outcome cohort. All participants included in the prospective biomarker cohort (i.e. VICIS study) provided written informed consent for study participation.
Ammonia levels increase with liver disease and portal hypertension severity in the outcome cohort NH 3 -ULN/NH 3 consistently increased with liver disease/portal hypertension severity, as evaluated by the CTP score (p <0.001) and UNOS MELD (2016) score (p <0.001), as well as severity of portal hypertension (p <0.001; Fig. 1; Table S1). Additionally, NH 3 -ULN also increased across clinical stages (p <0.001).
Univariable and multivariable analyses of factors associated with ammonia in the outcome cohort In univariable analyses, NH 3 (Table 2). Table 2. Simple and multiple linear regression analysis of factors associated with NH 3 corrected for the upper limit of normal includingamong other parameterseither CTP score, as well as serum sodium and creatinine (model 1), or UNOS MELD (2016) score, clinical stage, and serum albumin (model 2) in the outcome cohort.   Impact of ammonia on liver-related outcomes in the outcome cohort NH 3 not only increased with liver disease severity in crosssectional analyses but was also longitudinally associated with liver-related death  (Table 3). Next, we evaluated the prognostic performance of the previously provided cut-off of 1.4 in our outcome cohort. Importantly, this cut-off was not only associated with liver-related death in competing risk regression analysis (Table S2), but also hepatic decompensation (Cox regression Table 4 and Table S3;  competing risk regression Table S4), as well as liver-related hospitalisation (Cox regression Table S5; competing risk regression Table S6), and ACLF in dACLD (Cox regression Table S7;  competing risk regression Table S8).
In addition, we compared the prognostic performance of NH 3 -ULN for hepatic decompensation/liver-related death and liver-related death to UNOS MELD (2016) score and HVPG in time-dependent AUROC analyses for the following time points: 12, 24, 36, 48, and 60 months. Regarding hepatic decompensation/liver-related death, HVPG showed significantly better discriminatory ability compared to NH 3 -ULN at 24 months (NH 3 -ULN vs. HVPG: p = 0.044 after accounting for multiplicity). In contrast, NH 3 -ULN was comparable to time-dependent AUROC of the UNOS MELD (2016) score ( Fig. 2A). Importantly, timedependent AUROC of NH 3 -ULN for liver-related death was comparable to those of UNOS MELD (2016)-score and HVPG at all tested time points (Fig. 2B).
Finally, stratifying the cohort according to NH 3  Associations between ammonia and disease-driving mechanisms in the biomarker cohort Baseline characteristics of the biomarker cohort are provided in Table 1. Participants included in the biomarker cohort had less viral and more ARLD as underlying aetiology and were more often decompensated at baseline (62% vs. 46%; p <0.001).
As demonstrated in Fig. 4, ammonia showed low correlations with UNOS MELD (2016) score, BAs, HVPG, vWF, and ELF-test, as well as associations with CRP, IL-6, and PCT, MAP, renin, and serum sodium in the biomarker cohort. Further results on the correlations of ammonia with these biomarkers among compensated and decompensated individuals are reported in the Supplementary material.

Discussion
Although routinely measured ammonia is of limited value for diagnosing HE, a recent landmark study highlighted its prognostic implications. In our study, venous ammonia increased across clinical stages of ACLD as well as with more severe hepatic dysfunction and portal hypertension. Importantly, timedependent AUROC values of ammonia for liver-related death were similar to the laboratory-based composite score UNOS MELD (2016) and HVPG, which can only be measured invasively. Ammonia was not only independently associated with liver- Table 3. Uni-and multivariable Cox regression analyses of factors associated with liver-related death includingamong other parameters -CTP score, serum sodium, and creatinine (model 1) or UNOS MELD (2016) score, clinical stage, and serum albumin (model 2) in the outcome cohort. related deathas shown previouslybut also with other liverrelated outcomes including ACLF, even after adjusting for liver disease, systemic inflammation, and portal hypertension severity. Notably, our findings support the use of the previously proposed cut-off of NH 3 -ULN of > − 1.4 for risk stratification. Finally, we have provided information on associated pathophysiologic mechanisms that may explain the prognostic value of ammonia, that is, liver fibrogenesis/matrix remodelling and endothelial dysfunction.
Interestingly, diabetes was independently associated with venous plasma ammonia levels, even after adjusting for various co-factors (adjusted B: 0.134 [95% CI: 0.016, 0.252]; p = 0.026; Table 2). Diabetes may increase venous plasma ammonia levels by autonomic dysfunction, extended gastrointestinal transit times, and bacterial overgrowth as well as increased protein catabolism and accelerated muscle-breakdown. 23 Even in earlier stages of fibrosis in individuals with NAFLD/metabolic-associated fatty liver disease (MAFLD), deficiencies in urea synthesis (in part caused by impaired liver-a-cell axis with glucagon resistance and impaired ureagenesis 24 ), microglial activation, astrocyte swelling, and possibly even neurodegenerative changes and brain atrophyall caused by elevated ammonia levelshave been reported. 25,26 From a clinical perspective, Haj and Rockey 27 have found that in individuals with cirrhosis hospitalised with HE, ammonia was often normal and did not impact treatment decisions, thereby arguing against the routine use of ammonia as either an initial diagnostic test or for guiding medical therapy.
However, moving to risk stratification, ammonia is increasingly acknowledged as a prognostic biomarker. Vierling et al. 28 have shown that ammonia was able to identify people at risk for HE-related events. Moreover, a recent multicentre study by Tranah et al. 15 indicated that ammonia is predictive of hospitalisation/liver-related complications and mortality, showing a better prognostic performance than traditional scores. To corroborate the main finding of this study, we have externally validated the cut-off of > − 1.4 NH 3 -ULN. However, the authors did neither adjust their multivariable models for systemic inflammation nor portal hypertension severity, which has been accounted for in our work. Notably, ammonia levels varied significantly throughout the study centres, with only one centre reporting ammonia levels that were comparable to our cohort, which may be explained by differences in patient selection and characteristics. Interestingly, ammonia levels reported by Gairing et al. 29 were similar to ours and only a minority of individuals had ammonia levels above the ULN (13% vs. 19% in our cohort).
In people admitted because of acute decompensation of cirrhosis, ammonia levels have been found to be independently associated with mortality. 30 Our study extends these findings, as it demonstrated that stratifying individuals according to the proposed cut-off of > − 1.4 NH 3 -ULN identifies decompensated individuals at risk for ACLF, even if they are still outpatients/clinically stable. Therefore, it may provide the opportunity for the timely initiation of disease-modifying interventions that are currently under investigation (e.g. LIVERHOPE, NCT03150459).
Hyperammonaemia in ACLD is driven by ammonia overproduction/altered microbiome in the intestinal tract, 6,7 decreased metabolic capacity in the hepatocytes leading to a reduced ammonia metabolism in the urea cycle and via glutamine synthetase, 8,9 and portal hypertension through the development of portosystemic shunting via collateral flow. 10,11 Effects of hyperammonaemia include immune dysfunction and sarcopenia as well as direct negative implications on liver disease progression. 10 Recent works on the impact of hyperammonaemia in animal and in-vitro studies on fibrosis demonstrated the induction of oxidative stress and apoptosis and the activation of hepatic stellate cells (HSCs). 31,32 Intriguingly, we observed consistent (i.e. both in cACLD and dACLD) positive correlations between ammonia and the ELF-test, which has been shown to reflect fibrogenesis/extracellular matrix remodelling irrespective of the stage of ACLD and indicate HSC activation. 33 Moreover, there were also positive correlations with systemic inflammation, vWF as a marker of endothelial dysfunction, and severity of hepatic dysfunction and portal hypertension. Finally, ammonia   also correlated with BAs, which may be interpreted as a biomarker for portosystemic shunting, 34 which has also been linked with disease progression. 35,36 The main limitation of our study is its retrospective design. However, participants were thoroughly characterised at the time of HVPG measurement. For our multivariable models, we were unable to consider several potentially important prognostic indicators (e.g. sarcopaenia/frailty), as they have not been recorded systematically. Nevertheless, participants included in our study were extensively characterised in terms of portal hypertension severity, prognostic scores, and routine laboratory parameters including markers of systemic inflammationimportantly, all of these aspects have been considered in our analyses. Model selection was based on expert opinion/biological relevance. Applying backward elimination for variable selection yielded a 'slimmer' model for predictive purposes, that still included NH 3 . Furthermore, we cannot exclude that some hepatic decompensation events have been missed. However, we have thoroughly reviewed electronic health records of the Vienna hospital association and nationwide electronic health records. Moreover, we have also performed searches of the liver transplant database of our institution (i.e. the only transplant centre in eastern Austria) and examined the nationwide death registry. As complete information on (reason of) death is guaranteed by the latter measure, we included liver-related death in all composite endpoints to ensure the ascertainment of the most severe disease courses. We cannot rule out selection bias because we only included people undergoing HVPG measurement. However, haemodynamic evaluations are routinely performed for risk stratification and treatment monitoring purposes at our centre, and thus, we are confident that our study population is quite representative of clinically stable outpatients with ACLD treated at our centre. Finally, ammonia testing has several limitations. There is substantial laboratory variability 37 and arterial ammonia might be preferred over venous ammonia. [38][39][40] However, it has been shown that venous ammonia closely correlates with arterial ammonia in people with cirrhosis 41 and venous sampling substantially increases feasibility and therefore clinical utility in outpatients. We have provided a detailed description of the measurement of ammonia in the Materials and methods section of this study and are confident that our results are reliable, because preanalytical and analytical conditions were highly standardised.
In conclusion, venous ammonia predicts hepatic decompensation, non-elective liver-related hospitalisation, ACLF, and liverrelated death, independently of established prognostic indicators including CRP and HVPG.
Although venous ammonia is linked with several key diseasedriving mechanisms, its prognostic value is not explained by associated hepatic dysfunction, systemic inflammation, or portal hypertension severity, suggesting direct toxicity.

Financial support
The authors received no financial support to produce this manuscript.

Conflicts of interest
The authors have nothing to disclose regarding the work under consideration for publication. LB, JK, GS, MJ, LH, AFS, PS, and TS have nothing to disclose. The following authors disclose conflicts of interests outside the submitted work. RP received travel support from AbbVie, Gilead and Takeda. BSc received travel support from AbbVie, Ipsen and Gilead. BSi received travel support from AbbVie and Gilead. MP served as a speaker and/or consultant and/or advisory board member for Bayer, Bristol-Myers Squibb, Eisai, Ipsen, Lilly, MSD, and Roche, and received travel support from Bayer and Bristol-Myers Squibb. MT served as a speaker and/or consultant and/or advisory board member for Albireo, BiomX, Falk, Boehringer Ingelheim, Bristol-Myers Squibb, Falk, Genfit, Gilead, Hightide, Intercept, Janssen, MSD, Novartis, Phenex, Pliant, Regulus, Siemens and Shire, and received travel support from AbbVie, Falk, Gilead, and Intercept as well as grants/research support from Albireo, Alnylam, Cymabay, Falk, Gilead, Intercept, MSD, Takeda, and UltraGenyx. He is also co-inventor of patents on the medical use of 24-norursodeoxycholic acid. TR received grant support from AbbVie, Boehringer-Ingelheim, Gilead, Intercept, MSD, Myr Pharmaceuticals, Philips Healthcare, Pliant, Siemens, and W. L. Gore & Associates; speaking honoraria from AbbVie, Gilead, Gore, Intercept, Roche, and MSD; consulting/advisory board fees from AbbVie, Bayer, Boehringer-Ingelheim, Gilead, Intercept, MSD, and Siemens; and travel support from AbbVie, Boehringer-Ingelheim, Gilead, and Roche. MM served as a speaker and/or consultant and/or advisory board member for AbbVie, Collective Acumen, Gilead, Takeda, and W. L. Gore & Associates and received travel support from AbbVie and Gilead.
Please refer to the accompanying ICMJE disclosure forms for further details.               Table S8   Table S8. Competing risk regression analyses of factors associated with the development of ACLF/requirement of liver transplantation/liver-related death including -among other parameters -CTP-score, serum sodium, and creatinine (model 1) or serum albumin, and UNOS MELD (2016)-score (model 2) with non-liver-related death/removal of the primary etiological factor as competing risks in patients with who have experienced decompensation prior to study inclusion in the outcome cohort.