Liver Failure/Cirrhosis/Portal Hypertension
Survival in infection-related acute-on-chronic liver failure is defined by extrahepatic organ failures
Partly supported by NIH grant NIDDK RO1DK087913 and UL1RR031990 from the National Center for Research Resources.
Potential conflict of interest: Nothing to report.
Infections worsen survival in cirrhosis; however, simple predictors of survival in infection-related acute-on-chronic liver failure (I-ACLF) derived from multicenter studies are required in order to improve prognostication and resource allocation. Using the North American Consortium for Study of End-stage Liver Disease (NACSELD) database, data from 18 centers were collected for survival analysis of prospectively enrolled cirrhosis patients hospitalized with an infection. We defined organ failures as 1) shock, 2) grade III/IV hepatic encephalopathy (HE), 3) need for dialysis and mechanical ventilation. Determinants of survival with these organ failures were analyzed. In all, 507 patients were included (55 years, 52% hepatitis C virus [HCV], 15.8% nosocomial infection, 96% Child score ≥7) and 30-day evaluations were available in 453 patients. Urinary tract infection (UTI) (28.5%), and spontaneous bacterial peritonitis (SBP) (22.5%) were the most prevalent infections. During hospitalization, 55.7% developed HE, 17.6% shock, 15.1% required renal replacement, and 15.8% needed ventilation; 23% died within 30 days and 21.6% developed second infections. Admitted patients developed none (38.4%), one (37.3%), two (10.4%), three (10%), or four (4%) organ failures. The 30-day survival worsened with a higher number of extrahepatic organ failures, none (92%), one (72.6%), two (51.3%), three (36%), and all four (23%). I-ACLF was defined as ≥2 organ failures given the significant change in survival probability associated at this cutoff. Baseline independent predictors for development of ACLF were nosocomial infections, Model for Endstage Liver Disease (MELD) score, low mean arterial pressure (MAP), and non-SBP infections. Independent predictors of poor 30-day survival were I-ACLF, second infections, and admission values of high MELD, low MAP, high white blood count, and low albumin. Conclusion: Using multicenter study data in hospitalized decompensated infected cirrhosis patients, I-ACLF defined by the presence of two or more organ failures using simple definitions is predictive of poor survival. (Hepatology 2014;60:250–256)
Provide feedback or get help You are viewing our new enhanced HTML article.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
acute physiology and chronic health evaluation
chronic liver failure-sequential organ failure
infection-related acute-on-chronic liver failure
mean arterial pressure
Model for Endstage Liver Disease
North American Consortium for Study of End-stage Liver Disease
proton pump inhibitor
Research Electronic Data Capture
spontaneous bacterial peritonitis
urinary tract infection
white blood cell
Patients with cirrhosis are prone to infections as a result of bacterial translocation and impaired immunity. This infection risk is increased further in those who require multiple hospitalizations, antibiotics, and instrumentation. Infections are a leading cause of death in this population and despite advancements in critical care and prophylactic antibiotic use to prevent recurrent spontaneous bacterial peritonitis (SBP) and infection during/after variceal hemorrhage, infected patients have a high risk of 30-day mortality. The increased mortality is often related to the development of nosocomial or recurrent infections, renal failure, and the exacerbation of liver dysfunction. In a recent large European study of cirrhosis inpatients with and without infections, acute-on-chronic liver failure (ACLF), defined as two or more organ failures, especially if one organ involved was the kidney, predicted short-term mortality. However, the definitions of organ failures used were complex and also not specifically focused on infections, which comprised a large percentage of that study population. Therefore, a simple definition of ACLF to predict survival using widely available criteria, especially in patients with infections, is still needed to facilitate bedside decision making.
The aims of this study were to use the infected hospitalized cirrhosis population in the North American Consortium for the Study of End-Stage Liver Disease (NACSELD) database to 1) identify a group of patients with high mortality at 30 days and 2) to develop a simple bedside scoring index composed of user-friendly definitions of extrahepatic organ failure to determine survival probability in admitted infected cirrhosis patients.
Materials and Methods
The NACSELD database comprises prospectively collected data following informed consent from patients with cirrhosis hospitalized in 18 hepatology referral centers across the USA and Canada. Patients with an infection or who subsequently develop nosocomial infections are included. The study was approved by the respective Institutional Review Boards of the participating centers and uses a REDCap (Research Electronic Data Capture) tool that is located at Virginia Commonwealth University. REDCap is a secure, web-based application designed to support data capture for research studies, providing: 1) an intuitive interface for validated data entry; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for importing data from external sources.
Cirrhosis is diagnosed with a combination of biochemical, radiological, and endoscopic findings if liver biopsy confirmation is not available. Patients who had infections but did not require hospital admission were excluded, as were patients who did not have an infection during their admission. Other exclusion criteria include immunocompromised patients with human immunodeficiency virus (HIV) infection, prior organ transplant, and disseminated malignancies.
Once informed consent is obtained, data collection begins with patient demographics, vital signs, baseline full blood count, biochemistry, and assessment of liver and renal function. Details of the infection include antibiotic treatment as well as documentation of a second infection when applicable. Data regarding intensive care unit admissions, organ failures, liver transplantation, and length of hospital stay are also collected. Patients who are discharged alive are contacted at 30 days postenrolment to determine survival.
We define infections according to standard criteria4: 1) spontaneous bacteremia: positive blood cultures without a source of infection; 2) SBP: ascitic fluid polymorphonuclear cells >250/μL; 3) lower respiratory tract infections: new pulmonary infiltrate in the presence of: i) at least one respiratory symptom (cough, sputum production, dyspnea, pleuritic pain) with ii) at least one finding on auscultation (rales or crepitation) or one sign of infection (core body temperature >38°C or less than 36°C, shivering, or leukocyte count >10,000/mm3 or <4,000/mm3) in the absence of antibiotics; 4) Clostridium difficile Infection: diarrhea with a positive C. difficile assay; 5) bacterial entero-colitis: diarrhea or dysentery with a positive stool culture for Salmonella, Shigella, Yersinia, Campylobacter, or pathogenic E. coli; 6) soft-tissue/skin Infection: fever with cellulitis; 7) urinary tract infection (UTI): urine white blood cell >15/high-power field with either positive urine gram stain or culture; 8) intra-abdominal infections: diverticulitis, appendicitis, cholangitis, etc.; 9) other infections not covered above; and 10) fungal infections as a separate category. Nosocomial infections were those diagnosed after 48 hours of admission while second infections were those that were diagnosed after a separate first infection had been documented.
We used standard organ failure definitions as 1) hepatic encephalopathy >grade III or IV by West Haven Criteria; 2) shock: (mean arterial pressure [MAP] <60 mm Hg or a reduction of 40 mmHg in systolic blood pressure from baseline) despite adequate fluid resuscitation and cardiac output; 3) need for mechanical ventilation; and 4) need for dialysis or other forms of renal replacement therapy. These simple definitions are used to ensure generalizability.
Categorical data are presented as a percentage in addition to the actual numbers used to calculate the percentages. Continuous data are presented as means ± standard deviations while discrete data are presented as medians with the accompanying interquartile ranges. Group comparisons for categorical variables were done using the χ test with the corresponding degrees of freedom while group comparisons for continuous variables were done with either a two-sample t test or a one-way analysis of variance (ANOVA) if more than three groups are compared. Group comparisons of discrete data were done using nonparametric Wilcoxon Rank-Sum tests (Mann-Whitney U test) for two groups or Kruskal-Wallis tests for >2 groups. For all analyses, P < 0.05 was considered statistically significant.
To define the criteria for infection-related ACLF (I-ACLF), the determinants of 30-day mortality were calculated using logistic regression. A multivariate logistic regression model, with backward elimination, was used to arrive at a parsimonious model to determine baseline predictors of the development of ACLF. The variables analyzed were Model for Endstage Liver Disease (MELD) score and its components: age; gender; SBP versus other infections; gram-positive organism versus other organisms for the first infection; nosocomial first infection; alcoholic versus nonalcoholic cirrhosis etiology; and admission white blood cell (WBC) count, serum sodium, serum albumin, MAP, and heart rate. The resulting model was then pared down by eliminating, one by one, covariates that were not significant at the 0.05 level, and the final model where all covariates were significant at the 0.05 level was identified. Similarly, a multivariate logistic regression model, with backward elimination, was used to arrive at a parsimonious model to determine predictors of death. The variables analyzed were the same as those used for the prediction of I-ACLF with the addition of the presence or absence of I-ACLF, and the development of second infections. Other models were created to predict survival based on individual organ failures and combinations of two or three organ failure types to determine their probability of impairing survival compared to patients without any organ failures.
At the time of this analysis, 507 patients with cirrhosis were enrolled from 18 sites. Whereas most included patients who were admitted with an infection, 15.8% of patients developed their first infection while hospitalized. The details of the patient demographics, severity of cirrhosis, and medications used are enumerated in Table 1.
Table 1. Baseline and Admission Variables for the Patient Population
|Age (years)||55 ± 10 years|
|Gender (male/female)||298 (58.8%)/209(41.2%)|
|Etiology of cirrhosis (HCV/HCV+Alcohol/ Alcohol/NASH/Other)||124/138/74/78/93|
|History of complications|| |
|Variceal bleeding (%)||30%|
|Hepatic encephalopathy (%)||62%|
|Hyponatremia (Na <130) mmol/L (%)||59%|
|Hepatic Hydrothorax (%)||13%|
|Large volume paracentesis (one or more in the previous 6 months) (%)||45%|
|On liver transplant list (%)||34%|
|TIPS in place (%)||11%|
| || |
|Outpatient Medications at the Time of Admission|| |
| Proton pump inhibitors||59%|
| On SBP prophylaxis (all ciprofloxacin)||21%|
| On beta-blockers||41%|
|Within the previous 6 months|| |
| Received beta-lactams||35%|
| Median hospitalizations (interquartile range)||1 (1-3)|
| Median hospitalizations for infections (interquartile range)||1 (0-1)|
| || |
|Admission Vital Signs|| |
| Heart rate (beats/minute)||87.3 ± 16.8|
| Systolic blood pressure (mmHg)||116.4 ± 22.3|
| Diastolic blood pressure (mmHg)||64.9 ± 14.0|
| Mean arterial blood pressure (mmHg)||82.3 ± 18.2|
| Temperature (degrees Fahrenheit)||98.2 ± 3.1|
| || |
|Admission Laboratory Values|| |
| WBC count (/mm3)||8,334 ± 6,897|
| Platelet count (X100/mm3)||127.4 ± 59.5|
| Serum creatinine (mg/dl)||1.6 ± 1.1|
| Albumin (g/dl)||2.7 ± 1.4|
| Sodium (meq/L)||131.8 ± 9.3|
| Child Pugh score, mean ± SD [median (IQR range)]||10.3 ± 2.2 [10.0 (9.0, 12.0)]|
| Child Pugh score (proportion > 7)||96%|
| MELD score, mean ± SD [median (IQR range)]||19.9 ± 8.3 [19.0 (14.0, 26.0)]|
Types of Infections
In the 507 patients included, the most prevalent infections were as follows: UTI (28.5%), SBP (22.5%), spontaneous bacteremia (13.2%), skin/soft-tissue (12.2%), respiratory (9.9%), others (9.6%), and C. difficile (4.1%). Gram-positive infections were the leading isolate (32.9%), followed by gram-negative (26.8%), no organism (22.7%), and fungi (17.6%).
The 30-day outcomes were not available in 54 patients owing to their being lost to follow-up; the remaining 453 patients had complete 30-day postdischarge follow-up. During admission, 40% of patients required a large volume paracentesis, 40% had a central line inserted, 32% of patients required intensive care, 16% experienced an upper gastrointestinal (GI) bleed, and 6% required parenteral nutrition. Second infections were seen in 21.6% of patients; of the patients with second infections, 32% were UTIs, 25% had respiratory infections, 12.5% had SBP, 12.5% had spontaneous bacteremia, 10% had C. difficile, and 8% had other infections. During hospitalization and 30-day follow-up, 23% of patients died while 4% got transplants.
During the hospitalization, 55.7% developed grade III/IV HE, 17.6% developed shock, 15.1% required renal replacement therapy, and 15.8% required mechanical ventilation. Many hospitalized cirrhosis patients with a bacterial infection did not experience even a single organ failure (38.4%). However, 37.3% experienced a single organ failure, 10.4% had two organ failures, 10% had three organ failures, and 4% had all four organ failures. The risk of death within 30 days of hospital discharge was highest in patients with two or more organ failures (Table 2).
Table 2. Crude Survival Rates With Individual Organ Failures
|HEa||184/255 (72.2%)||186/198 (94.1%)||<0.0001|
|Shock||34/81 (41.9%)||332/367 (90.4%)||<0.0001|
|Renal replacement therapy||46/68 (67.6%)||322/383 (84.1%)||0.002|
|Mechanical ventilation||28/44 (38.9%)||338/378 (89.4%)||<0.0001|
|Any one of the above||74/196 (72.6%)||159/166 (95.7%)||<0.0001|
|Two of the above||56/109 (51.3%)||299/327 (91.4%)||<0.0001|
Multiple logistic regression analyses were done to assess the relative significance of the individual and combined impact of different organ failures on 30-day survival. The number of organ failures ranged from zero to four. The patients with four organ failures had a 23% 30-day survival compared to 96% in those without any organ failures. Individual organ failures and combinations of two or three different organ failures were compared to the group of patients without any organ failure. The need for dialysis alone was not an independent predictor of death in multivariate analysis, while the other three, especially shock and the need for mechanical ventilation, were independent predictors of death. Using these combinations to estimate survival, any two or any three organ failures severely reduce the survival compared to one or no organ failure (Table 3).
Table 3. Univariate Logistic Regression Survival Probability With Combinations of Failures
|No organ failures||0.96||—||—|
|Single Organ Failure Only|
| Dialysis only||0.93||0.62||(0.33, 1.16)|
| HE only||0.88||0.35||(0.17, 0.71)|
| Ventilation only||0.85||0.26||(0.13, 0.56)|
| Shock only||0.84||0.23||(0.12, 0.48)|
| HE and dialysis||0.82||0.21||(0.09, 0.53)|
| Ventilation and dialysis||0.78||0.16||(0.07, 0.39)|
| Shock and dialysis||0.76||0.15||(0.06, 0.37)|
| HE and ventilation||0.67||0.09||(0.03, 0.24)|
| HE and shock||0.64||0.08||(0.03, 0.21)|
| Shock and ventilation||0.57||0.06||(0.03, 0.13)|
| HE, ventilation and dialysis||0.55||0.06||(0.02, 0.16)|
| Shock, HE and dialysis||0.52||0.15||(0.02, 0.15)|
| Shock, ventilation and dialysis||0.46||0.04||(0.02, 0.09)|
| Shock, HE and ventilation||0.32||0.02||(0.01, 0.06)|
|Combination of All Four|
| Shock, HE, ventilation and dialysis||0.23||0.01||(0.01, 0.04)|
Determinants of 30-Day Mortality
We performed multivariate regression analysis with ACLF (defined as two or more organ failures) as the outcome using MELD: age; gender (male); SBP versus other infections; gram-positive organism versus other organisms for the first infection; nosocomial source for the first infection; alcoholic versus nonalcoholic cirrhosis etiology; and admission WBC, serum sodium, serum albumin, MAP, and heart rate as variables (Table 4A). We found that MELD score and nosocomial first infections were risks for ACLF, whereas a high MAP was protective and SBP had a lower risk for I-ACLF than other infections.
Table 4. Multivariate Logistic Regression Was Performed to Determine the Following Outcomes: (A) I-ACLF (Defined as Two Or More Organ Failures), (B) 30-Day Survival
|Nosocomial first infection||0.7968||0.3047||0.0089||2.22 (1.22, 4.03)|
|Admission MELD||0.1005||0.0170||< 0.0001||1.11 (1.07, 1.14)|
|SBP as first infection||−0.6051||0.2967||0.0414||0.55 (0.31, 0.98)|
|Mean arterial pressure||−0.0330||0.0096||0.0005||0.97 (0.95, 0.99)|
|(B) 30-Day Survival|
|I-ACLF||−1.836||0.3325||< 0.0001||0.16 (0.08, 0.30)|
|Second infection||−0.8970||0.3160||0.0045||0.41 (0.22, 0.76)|
|Admission MELD||−0.0579||0.0217||0.0078||0.94 (0.90, 0.98)|
|Admission WBC||−0.4770||0.1979||0.0160||0.62 (0.42, 0.91)|
|Admission serum albumin||0.6201||0.2229||0.0054||1.85 (1.20, 2.85)|
A regression analysis was also performed to predict 30-day survival using the following variables (Table 4B): I-ACLF, presence of second Infection; nosocomial first infection; admission MELD; age; gender (male); SBP versus other infections; gram-positive organism versus other organisms for the first infection; alcoholic versus nonalcoholic cirrhosis etiology; and admission values WBC, serum sodium, serum albumin, MAP, and heart rate as variables. Higher serum albumin was predictive of survival while I-ACLF, second infections, MELD score, and higher admission WBC counts were predictive of mortality. In order to assess “liver-specific” variables we refit these models by replacing the MELD score with its individual components. For I-ACLF, the only significant predictor of the MELD components was creatinine (P < 0.0001), not bilirubin (P = 0.08) or International Normalized Ratio (INR) (P = 0.16). For the survival model, none of the individual components were significant predictors.
Patients with cirrhosis form a disproportionately large percentage of patients hospitalized with infections. These infections are often a preterminal event and as such use a tremendous amount of resources. The development of simple triage criteria for infected cirrhosis patients who are likely to survive to discharge or liver transplant is therefore imperative for appropriate counseling of patients and family members to offer realistic expectations of outcome for patients and set appropriate goals of care accordingly. Defining a group of patients at high risk for 30-day mortality, that is, I-ACLF, using simple clinical criteria in infected cirrhosis patients is critical to bedside decision-making to accurately identify potential survivors for cost-effective healthcare resources utilization.
Our analysis defined I-ACLF based on data in decompensated infected cirrhosis patients using a large, prospective, multicenter cohort. Infected cirrhosis patients represent an ideal cohort for analysis of ACLF given that infection is a common cause for hospitalization and acute decompensation. The characteristics of infections in cirrhosis have evolved rapidly over the years and with the increase in hospitalizations, use of proton pump inhibitors, and antibiotic prophylaxis, the profile of the infectious agents has also changed dramatically over time. Whereas previously the majority of infections were caused by gram-negative bacteria, gram-positive bacteria are increasing in incidence. There is also a tremendous increase in non-SBP infections, such as UTI, respiratory, or skin/soft-tissue infections, which is again a profile change from prior studies. The current results are a reflection of evolving epidemiological patterns with the majority of infections being non-SBP, gram-positive or with no isolated organism, and an increasing proportion being nosocomial infections. These results are derived from diverse sites throughout North America and therefore likely reflect overall trends for infections occurring in cirrhosis patients. Therefore, the outcome and survival assumptions, including the ACLF definition in the setting of an infection, that were observed and made several years ago are no longer applicable and need to be reconsidered.
We analyzed determinants of survival after prospective data collection. The definitions of organ failure used were simple, objective, and readily accessible to both clinicians and researchers. Simple definitions are important because complex and unwieldy instruments such as APACHE (Acute Physiology and Chronic Health Evaluation) rarely gain traction beyond critical care specialists. The organ failure definitions in this study were used to define ACLF based on their effect on survival and restricted to patients with infections. We found that any one of these organ failures individually and most of them in combination were significantly associated with diminished survival. We did not include “liver-specific” laboratory values to define liver failure since changes in INR and bilirubin are difficult to translate into simple criteria and the multivariate analyses using INR and bilirubin individually in this group did not predict outcomes. This is likely because patients were markedly decompensated at baseline, with 96% of our patients having a Child score ≥7 on admission. Therefore, while the liver failure in these patients may play a role in susceptibility to infection in the first place, the nonhepatic organs may be important in determining overall survival. Our results are similar to those seen in a recent European study, which increases the validity of these results in determination of survival in infection-associated ACLF. However, the severity of organ failure in this study was defined using the “CLIF-SOFA” score, which is complex and not readily adaptable to clinical care. We also found a significant decrement in overall survival after two or more organ failures using a simpler and clinically usable definition. Thus, a similar message is being sent from both the European and NACSELD studies involving a large number of subjects in different regions of the world that, not surprisingly, the higher the number of organ failures, the lower the overall patient survival.[7, 11]
In prior studies, including in Europe and our dataset, acute kidney injury (AKI), defined by change in serum creatinine rather than a requirement for dialysis, predicts survival.[12, 13] We did not find the need for dialysis as an independent predictor of survival. Dialysis as a marker of prognosis must be interpreted carefully, as this intervention may only be offered to individuals thought likely to recover from infection and go on to liver transplant (thus not meeting criteria for dialysis may be a better indicator of overall prognosis than indicative of more severe kidney injury). In our study there were no protocol prespecified criteria for dialysis and the decision to initiate it was center-specific. Therefore, the need for dialysis, which is a hard endpoint, is not solely reflective of AKI, as in prior studies. HE had the next lowest prognostic implication. This is likely due to the fact that when infection is the precipitant of HE, treatment of the infection is associated with HE improvement. While not unexpected that ventilated patients who develop shock or require dialysis have the worst survival, it is notable that any one of these individual organ failures is associated with a relatively good survival compared with the combination in infected patients. Therefore, prognostication in this group of patients is nuanced and several factors ultimately decide the survival of the patients.
The determinants of ACLF development were as expected, a higher MELD score and lower MAP on admission. Interestingly nosocomial infections were significantly associated with development of ACLF, while SBP conferred a lower risk. These two aspects highlight changes that have occurred in the epidemiology of infections in this population. The role of nosocomial infections, which were found as first infections in 15.8% of our patients, is an important contributor to adverse outcomes. Causative organisms in nosocomial infections are more likely to be drug-resistant, are associated with procedures or hospital cross-contamination, and medication exposure (antibiotics, proton pump inhibitor use, or SBP prophylaxis) may play an important role in their occurrence.[2, 4, 9] The lower risk for poor outcomes conferred by SBP as the source of infection is a significant departure from prior studies in which SBP was associated with AKI and mortality.[14, 15] We hypothesize that the relative protection from kidney injury afforded patients by current guidelines that promulgate intravenous albumin use and aggressive screening and treatment for SBP even in asymptomatic patients may be responsible for more favorable prognosis. This highlights the potential power of evidenced-based guidelines for improving patient outcomes and the continued need to determine interventions that reduce incidence and lower mortality for non-SBP infections in cirrhosis. The ACLF definition derived from NACSELD data was predictive of survival when controlling for both severity of liver disease and characteristics of infection (nosocomial, healthcare-associated, or second infections during the index hospitalization). Second infections are particularly important to control for, as data from NACSELD have demonstrated they are an important predictor of mortality and they are more likely to occur in patients with greater length of stay. The association of mortality with ACLF despite controlling for significant confounders supports the inherent biological validity of this definition. Other determinants of survival were admission cirrhosis severity (MELD score and serum albumin) and admission WBC; a higher admission WBC count was associated with impaired survival, which is similar to the European trial findings. The finding that WBC is associated with survival may indicate that patients with cirrhosis who mount a leukocytosis have a higher proinflammatory milieu that results in impaired survival. Thus, a paradigm shift in infection epidemiology and outcomes has occurred; a greater proportion of gram-positive organisms are seen and non-SBP and nosocomial infections are increasingly occurring. SBP mortality has declined while nosocomial first infection-related complications have increased.
Of note, previous studies observed a high mortality in patients with cirrhosis requiring intensive care with respiratory failure. A predictive model for mortality based on MELD score and ventilator support has been developed in cirrhosis patients admitted to the ICU for a variety of reasons. In addition, shock has been a predictor of mortality. Our study uniquely confined to ACLF patients precipitated by infections has validated some of these previous observations. The study is limited by the use of these simple terms for organ failure, which means that subtler manifestations of organ failure would have been missed. Also, mechanical ventilation could be used for airway protection for severe HE rather than respiratory failure. However, despite these terms, and our 10% lost to follow-up rate, our results were similar to prior studies in terms of survival.
We conclude that a simple bedside definition of I-ACLF (defined two or more organ failures—shock, HE, ventilation, and renal replacement therapy) is an excellent prognostic tool in our large, prospective, multicenter study of infected hospitalized decompensated cirrhosis patients.