This was a retrospective non-randomized human study based on the analysis of 777 consecutive patients admitted to the ICU in the hepato-gastroenterology department at Theodor Bilharz Research Institute (TBRI), Giza, Egypt. All were cirrhotic patients with either HE and/or GI bleeding admitted during the one year period i.e., from January 1^st to December 30^th, 2015. These are the two types of patients, we used to admit to our hepatic ICU. All patients were anti HCV-antibody positive with liver cirrhosis and portal hypertension as evidenced by clinical examination, laboratory, ultrasonographic and endoscopic features. None of them received recent oral antiviral treatment. All patients with upper GI bleeding were exposed to upper gastrointestinal endoscopy after they have been resuscitated and stabilized by administration of intravenous fluids, packed red blood cells, octreotide and antibiotics. Most of them underwent upper GI endoscopy within 24 hours of admission. The Oesophageal Varices (OV) were banded, fundal varices were injected with histoacryl and bleeding peptic ulcers were treated with two methods, mostly diluted adrenaline injection and argon plasma coagulation when detected. If bleeding was observed in the ulcer base, it was treated by clips. No liver biopsy was done for any patient because of prolonged prothrombin time and/or presence of ascites. The laboratory data were determined within two hours of admission to ICU and the most recent findings were considered when patients were readmitted.

Exclusion criteria include: *Age <18 years.

The study had been approved by the ethical committee of TBRI and follows the Helsinki rules.

Where the modified terms bilirubin_C = bilirubin bounded below by 1 mg/dl, creatinine_C = creatinine capped by 0.8 mg/dl below and 3 mg/dl above, INR_C = INR bounded by 1 below and 3 above and Na_c = serum sodium bounded by 125 below and 140 above.

Kolmogorov-Simirnov testing was performed to evaluate the normality of the distribution of each variable. Continuous variables were compared by Mann-Whitney test. Categorical variables were evaluated using a Chi-square test or Fisher’s Exact test as needed. The accuracy of prognostic models was analyzed by Receiver Operating Characteristics (ROC) curves and by calculating the sensitivity, specificity and 95% CI for chosen cut offs. Logistic regression was performed to identify risk factors for mortality (dependent variables) by backward stepwise regression. Kaplan-Meier curves were plotted to display the impact on survival. All statistical analyses were performed by the statistical software SPSS version 22.0 (SPSS, Chicago, IL, USA). A p-value of less than 0.05 was considered statistically significant.

A total of 777 patients, admitted to the ICU either due to hepatic encephalopathy and /or GI bleeding were included in our study; 72.8% survived and 27.2% died. Mean age was 57 years (range 23-80 years) and 66.4% were males. The demographic, clinical characteristics and laboratory results of the study groups are shown in [Table/Fig-1]. Both age and sex did not show any statistical difference between survivors and non survivors. Serum bilirubin, creatinine, INR and hyponatremia were significantly higher in non survivors compared to the survivors. Non survivors showed significantly higher proportions of ascites (p<0.01), hepatic encephalopathy (p<0.01), Child C (p<0.01), Spontaneous Bacterial Peritonitis (SBP) (p<0.01) and hepatorenal syndrome (HRS) (p<0.01) compared to the survivors [Table/Fig-2]. Associated co-morbidities did not show statistically significant difference between the two groups [Table/Fig-3]. Mean values of the MELD, MELD-Na, MESO, iMELD, Refit MELD and Refit MELD-Na models are shown in [Table/Fig-4]. Significantly higher means values (p<0.01) were observed for all models among individuals who died compared to survivors. No significant differences were observed when model equations compared with each other (p>0.05). ROC curves of all studied models at admission was plotted to evaluate their performance in predicting ICU mortality [Table/Fig-5,6]. MELD-Na had the highest area under curve and was highly predictive of mortality at an optimized cut-off value of 20.4. Logistic regression analysis showed the number of transfused blood units and hyponatremia as additional independent prognostic factors that predict mortality [Table/Fig-7].

[Table/Fig-6]:

ROC curves of all studied models: (a)-MELD-Na, (b) MELD, (c) MESO, (d) iMELD, (e) Refit MELD-Na, (f) Refit MELD.

Many prognostic models and scores have been proposed in the last two decades to predict prognosis in patients with ESLD and to determine the most appropriate therapeutic option [17]. The ICU mortality rate of cirrhotic patients ranged from 34% to 68% [2]. Overall mortality in our patients was 27.2%. This is contrary to other reported mortality 47% [18], >55% [19] and 57.5% [20] in cirrhotic patients admitted to ICUs. The explanation can be attributed to variability within the stages of the disease, associated complications and number and type of non-functioning organs among different studies. Also, most (nearly 63%) of our patients were admitted to the ICU because of GI bleeding for whom emergency therapeutic intervention is available with estimated mortality of 15.19%. This is in concordance with Carbonell N et al., who reported improved survival after variceal bleeding in patients with cirrhosis over the past two decades from 42.6% to 14.5% [21]. In our series, 37% of our patients admitted to the ICU because of HE, of whom 47.4% died. This is in concordance with other studies where HE comprised the single most common presentation of liver disease in 47% of patients, of whom 50% died in hospital [18] and was the most common complication of cirrhosis requiring admission to the ICU (33%) where it was also was the most common cause of death (50%) [20].

Both age and gender did not predict outcome in our study. Mean age was 57 years; 66.4% of whom were males. This is in concordance with other studies. Gunnarsdottir SA et al., [22] reported mean age around 60 years; whereas O’Brien AJ et al., [19] reported the mean age was 52.5 years with male preponderance (60%). Parkash O et al., reported more men with liver disease (60%) were admitted to the ICU compared with women (40%) [18]. Similarly, Nafeh H et al., in 2014 [20] reported mean age of 56 and the majority were men (85%).

In our study, INR, serum creatinine, bilirubin (MELD components), white blood cells count and hyponatremia (<126 meq/l) were significantly higher in ESLD patients who died in the ICU. This is similar to other report [23]. Bilirubin was a good parameter to predict patients’ survival although various limitations should be kept in mind. Even in advanced hepatic disease, bilirubin levels may be normal [24]. In our study, 73 (19.3%) patients under child score C had normal serum bilirubin level. On the other hand, elevated levels are often very unspecific and can point to many different diseases, severe or harmless [25]. Still, in patients with liver disease elevated levels of bilirubin point to a worse outcome [24]. Thus, bilirubin levels may at least provide some estimate of mortality in patients with liver disease admitted to an ICU.

We did not find statistically significant value for hypoalbuminemia; although, it was low in patients who died in ICU. Hypoalbuminemia has been shown to be a frequent finding in hepatic patients on ICU admission, because it may result from hemodilution, capillary leak, or a cytokine-induced decreased synthesis of albumin by liver rather than pre-existing liver disease alone [26]. Similarly, INR may increase because of hypoxic hepatitis, disseminated intravascular coagulation, or bleeding-induced coagulopathy. The absence of a prognostic impact of liver disease severity with the prognostic impact of high INR or low serum albumin, suggests that INR and serum albumin might not reflect liver disease severity in the setting of ICU admission [27].

Studies showed that patients with decompensated cirrhosis had a poor prognosis, particularly when they develop complications related to ESLD, such as HE, SBP, or GI bleeding with or without diminished renal function [28]. The latter is a well established important predictor of survival in those patients [29]. Pan HC et al., showed that cirrhotic patients with acute kidney injury admitted to ICUs show extremely high mortality rates [30]. This came in agreement with our study as 78 (10%) patients of our ESLD patients admitted to the ICU presented with HRS, of whom 44 (5.6%) patients died. Of note is that the HRS indicates a severe liver failure and has a bad prognosis without liver transplantation [31].

A recent meta-analysis by Arvaniti V et al., showed that the presence of infections in cirrhotic patients increased the mortality four-fold [32]. Indeed, in our cohort one of the main causes of death was sepsis. However, the detection of pathogens may not necessarily be indicative of an active infection, 34 of 48 (70.8%) patients had died from SBP.

We compared the efficacy of the liver specific scoring system in predicting the outcome of critically ill cirrhotic patients admitted to the ICU. In our study, patients with cirrhosis who died during their ICU stay displayed significantly higher values on all prognostic scores at admission when compared to those who survived. MELD-Na had the highest area under curve comparing to the other scores denoting the importance of adding serum sodium to the original MELD. This is in agreement with previous report of Kim WR et al., [14]. They reported that 7% of waiting list deaths could be averted if the MELD-Na score were used for liver allocation. However, the effect of hyponatremia gradually diminished as the MELD score increased. It is known that serum sodium concentrations may vary depending on factors such as the volume status and use of diuretics [14]. However, published data indicate that hyponatremia in patients with cirrhosis is difficult to alter [33]. It is worth noting that how well the model discriminates between individuals who will live and those who will die is termed discrimination, and how close the estimated probabilities of mortality correlate with the observed mortality over the entire range of probabilities is termed calibration. It is rather impossible for any model to have perfect calibration and discrimination at the same time [34]. We found Refit MELD-Na scoring model to have a mortality predicting value which in contrary to Kim J et al., [15] who reported no difference between both Refit MELD and Refit MELD-Na in predicting mortality in patients with ESLD with ascites. It could be attributed to different type of patients included in our study.

As previously suggested by others [35] the prognosis of ICU admitted cirrhotic patients seems to have improved over time. The hypotheses that may explain this observation include significant advances in medical care of cirrhotic patients and/or in general intensive care, a more strict selection of cirrhotic patients by ward physicians or intensivists for ICU admission, or differences in the characteristics of studied populations (cirrhosis causes, reasons for ICU admission) [21,35].

Potential limitations of our study include the retrospective design, absence of randomization and no follow-up with single time point measurement of blood tests. However, this study for our knowledge is the first to study a homogenous group of patients with single disease; HCV infected patients with and without complications and study most of the liver specific scores. Needless to say, randomized controlled prospective studies with longer follow up and repetitive scoring calculations are needed to confirm our results.

Patients with liver cirrhosis due to HCV infection, who died during their ICU stay displayed significantly higher values on all prognostic scores at admission when compared to those who survived. The addition of sodium to MELD score greatly improves the predictive accuracy of non survival. MELD-Na showed the highest predictive value of all scores. Randomized controlled prospective studies with longer follow-up and repetitive scoring calculations are needed to validate the predictive accuracy of liver specific scoring system.