Prognosis of cirrhotic patients admitted to intensive care unit: a meta-analysis

The natural course of liver cirrhosis is often punctuated by life-threatening complications requiring admission to an intensive care unit (ICU) [1]. Critically ill cirrhotic patients were reported to have poor prognosis [2–4], but recent data suggest improvements due to significant progress in the management of general in-ICU populations, and a better understanding of the pathophysiology of cirrhosis [5–7]. The generalization of standardized interventions, such as management of portal hypertension-related bleeding, has contributed to a survival gain in this population. However, survival after hospital discharge remains low in cirrhotic patients [3, 8–10]. Numerous studies have focused on prognostic factors of cirrhotic patients admitted in ICU but provided short-term and controversial results, possibly impacted by selection biases. In the more recent studies on the topic, the performances of some isolated characteristics such as acute kidney injury [11–14], need for mechanical ventilation [3, 15, 16], septic shock [5, 17], and of liver-specific or general ICU scores [2, 7, 18–21] to predict mortality have been compared but the design was often retrospective and reflected a single-center experience. As a consequence, no evidence-based recommendation regarding the admission of critically ill cirrhotic patients to ICUs can be made [22]. More importantly, the long-term prognosis of the patients who survived to intensive care and its determinants has been poorly documented.

The aim of this meta-analysis was therefore to compare the prognostic value of a large set of variables on in-ICU, in-hospital, and 6-month mortality in patients who survived the ICU phase.

Our meta-analysis investigated numerous characteristics of a large population of cirrhotic patients from different centers. It enabled us to provide original data regarding 6-month mortality in more than 400 patients who survived intensive care. It yields four major findings: (1) despite in-ICU mortality rates below 50% and decreasing in the more recent studies, medium-term prognosis remained poor and this finding was reproducible between studies. One possible explanation could be that a very small proportion of patients underwent liver transplantation; (2) short-term mortality was well predicted by ICU scores, which reflect the degree of organ dysfunction; (3) bad ICU score results at baseline did not predict 6-month mortality in ICU survivors. In those patients, the major determinants of outcome were liver and renal failure, underlining the value of the MELD; the CLIF-SOFA, that incorporates parameters of liver function, was the only score able to predict both in-ICU and mid-term survival; (4) some events considered alone, mostly related to SIRS and sepsis, were strong predictors of poor short-term prognosis, with performance similar to that of composite scores.

This analysis confirms that in-ICU survival was significantly better in patients admitted after 2004. This improvement in short-term prognosis over time suggests that substantial progress has been achieved in the management and/or selection of critically ill cirrhotic candidates for ICU. However, the dramatic decrease in medium-term survival illustrates the fragility of these patients and the limits of available efficient therapies beyond the initial control of organ failures allowing ICU discharge. Liver transplantation was rarely performed, even in centers with liver transplantation program available, and to date, this option appears to be inadequate to solve the problem. This finding is in keeping to what shown by a recent US study, in which only a small proportion of liver recipients (8.1%) comes from ICUs [41]. This is surprising, since liver transplantation of critically ill patients has been reported to provide acceptable results in some critically ill patients [42, 43]. Hence, the “sickest-first” policy applied in the USA and Europe to allocate liver grafts should give high priority to this population, providing the rigorous selection of patients in the context of organ shortage, by discarding those with disseminated hepatocellular carcinoma, severe comorbidities, uncontrolled sepsis, and presumed non-compliance, especially regarding alcohol consumption. All these points justify referring ICU survivors to a liver transplant center for careful evaluation.

In most studies, age was not reported to have an independent influence on mortality in this population, appearing only as a component of the APACHE score. In our study, one spectacular finding was the better prognosis of patients <40 years, who represent a small proportion of each cohort but can be adequately studied through meta-analysis. This finding is contrary to a recent study in which younger patients were supposed to have more severe acute on chronic liver failure (ACLF) [44]. Patients admitted to ICU for variceal bleeding had better short-term outcomes. This probably results from the more general use of pharmacological and endoscopic therapies, and TIPS, which efficiently control variceal bleeding. Pooled analyses of other reasons for admission suggested poor outcome, but were heterogeneous. Hence, indications other than variceal bleeding cannot be considered as deleterious per se. Similarly, results were heterogeneous regarding the impact of organ replacement therapy on ICU mortality. In particular, mechanical ventilation per se did not equal death, contrary to what Rabe et al. suggest [15]. The MARS system, perhaps because it was allocated to the sickest patients, was a strong surrogate for in-ICU mortality. This indicates that MARS is not powerful enough to cure multi-organ failure in cirrhotic patients, and should rather be considered as a bridge to liver transplantation.

Our study confirms the power of composite scores to predict in-ICU mortality. Among them, the SOFA was largely used and the recent CLIF-SOFA [45], although not used by the majority of authors at the time of their studies, provided strong predictions. The APACHE-II score was not widely used in cirrhotic patients, probably because of its complexity, and/or because of liver failure, which renders the interpretation confusing. These general scores were better than the liver-specific scores for the prediction of in-ICU mortality. It is acknowledged that short-term prognosis mostly depends on the number of organ failures and that scoring liver failure provides limited prognostic information in patients with multi-organ failure. Conversely, MELD and Child–Pugh, even measured in the context of organ failures, were able to predict 6-month mortality among ICU survivors. Interestingly, the APACHE-II and CLIF-SOFA, more influenced by liver failure than the SOFA, maintained their predictive value after ICU discharge. Moreover, we found that among the extrahepatic organ failures, renal failure was more powerful for the prediction of 6-month mortality in patients who survived ICU than for the prediction of in-ICU mortality. This is in keeping with results already reported by Fede et al. [14] and reinforces the value of the MELD score. The question of repeated evaluation of patients at day three, supported by some landmark papers [8, 46, 47], was also evaluated by our meta-analysis. Only a small proportion of patients (10%) had available SOFA at day three, thus decreasing the power of our pooled analyses. As a consequence, changes in SOFA between baseline and day 3 did not provide better predictions than baseline evaluations. This raises the question of the acceptability of this procedure, which supposes the interruption of organ support in the event of aggravation.

Another major result of our study is the impact of SIRS and sepsis in the outcome of cirrhotic patients admitted to ICU. The pejorative influence of SIRS and infection has been shown in non-ICU populations of cirrhotics [25, 48]. Our meta-analysis demonstrates that this is still true in the most severe cirrhotic patients regardless of the reason for admission in ICU. In particular, fungal infection, an event often associated with a delay in initiating the appropriate antimicrobial therapy, had a strong negative influence. Gram-negative infections were also significantly associated with mortality, highlighting the deleterious impact of endotoxinemia in the natural history of cirrhosis [25]. Pneumonia and its consequences were clearly a challenge for intensive care specialists, whereas spontaneous bacterial peritonitis, a surrogate of end-stage liver disease, had a deleterious influence in the longer term.

Our study acknowledges several limitations. First, the need of original data led to lack of exhaustiveness since a substantial proportion of authors was unable to share their data and answer our questionnaire. Second, because of missing data and/or recall biases, our meta-analyses of events that occurred during ICU stay were probably inaccurate. Third, among the 301 pooled analyses performed in the present work, 108 provided heterogeneous results and the source of heterogeneity was unknown. This is why we decided not to take heterogeneous results into account in our conclusions. Fourth, we did not study prognostic scores recently described, such as the quick SOFA [49] and the Besançon’s score [21], because they were not available at the time of study conception. The prognostic impact of ACLF was also incompletely studied because this condition was not defined at the time of some studies. Only a retrospective assessment of ACLF as a reason for admission could be studied, with no significant impact on mortality. Fifth, the design of this meta-analysis only allowed separate assessments of weight-adjusted pooled odds ratios for a single variable at different follow-up timepoints, but did not allow us to combine variables for analyses. However, correct multivariate analyses must include truly independent variables, and we assume that, in the context of multi-organ failures, clinical condition, biochemical parameters, and interventions are intricate. Sixth, because individual data were not available in the present study, it was not possible to provide new information regarding which patients could be considered futile for intensive care, and to confirm or not the conclusions from the CANONIC cohort recently published [44].

Our analysis indicates that admission to ICU should be envisaged as soon as possible for critically ill cirrhotic patients, before too many extrahepatic failures compromise short-term survival. ICU survivors without heavy comorbidities can be identified early and may be systematically considered for transplantation. The number and intensity of organ failures on ICU admission does not mask the power of MELD score measured concomitantly for predicting 6-month mortality and thus immediately discriminates the best candidates for liver transplantation. The CLIF-SOFA is relevant for discriminating both in-ICU and mid-term mortality in critically ill cirrhotic patients.