Second infections independently increase mortality in hospitalized patients With cirrhosis: the north american consortium for the study of end-stage liver disease (NACSELD) experience


  • Potential conflict of interest: Nothing to report.

  • Partly supported by National Institutes of Health (NIH) grant NIDDK RO1DK087913 and UL1RR031990 from the National Center for Research Resources.


Bacterial infections are an important cause of mortality in cirrhosis, but there is a paucity of multicenter studies. The aim was to define factors predisposing to infection-related mortality in hospitalized patients with cirrhosis. A prospective, cohort study of patients with cirrhosis with infections was performed at eight North American tertiary-care hepatology centers. Data were collected on admission vitals, disease severity (model for endstage liver disease [MELD] and sequential organ failure [SOFA] scores), first infection site, type (community-acquired, healthcare-associated [HCA] or nosocomial), and second infection occurrence during hospitalization. The outcome was mortality within 30 days. A multivariate logistic regression model predicting mortality was created. 207 patients (55 years, 60% men, MELD 20) were included. Most first infections were HCA (71%), then nosocomial (15%) and community-acquired (14%). Urinary tract infections (52%), spontaneous bacterial peritonitis (SBP, 23%) and spontaneous bacteremia (21%) formed the majority of the first infections. Second infections were seen in 50 (24%) patients and were largely preventable: respiratory, including aspiration (28%), urinary, including catheter-related (26%), fungal (14%), and Clostridium difficile (12%) infections. Forty-nine patients (23.6%) who died within 30 days had higher admission MELD (25 versus 18, P < 0.0001), lower serum albumin (2.4 g/dL versus 2.8 g/dL, P = 0.002), and second infections (49% versus 16%, P < 0.0001) but equivalent SOFA scores (9.2 versus 9.9, P = 0.86). The case fatality rate was highest for C. difficile (40%), respiratory (37.5%), and spontaneous bacteremia (37%), and lowest for SBP (17%) and urinary infections (15%). The model for mortality included admission MELD (odds ratio [OR]: 1.12), heart rate (OR: 1.03) albumin (OR: 0.5), and second infection (OR: 4.42) as significant variables. Conclusion: Potentially preventable second infections are predictors of mortality independent of liver disease severity in this multicenter cirrhosis cohort. (HEPATOLOGY 2012;56:2328–2335)

One-third of all patients hospitalized with cirrhosis have at least one infection during hospitalization,1-5 which increases length of hospital stay, cost, and mortality. No published data exist on the impact of a second in-hospital infection in this population.6-8 Only one-third of these infections are community-acquired; two-thirds are healthcare-associated (HCA) or nosocomial in origin.1, 3 Previous studies indicate that one-third of HCA and nosocomial infections are potentially preventable.9, 10 As a result of the high prevalence of HCA and nosocomial infections in cirrhosis, first-line antibiotic therapy traditionally used against community-acquired infections is often ineffective.3 Because much of the data are from single-center European studies and there remains a need for a multicenter study to broadly understand the nature and impact of second infections, and assess resistance patterns in hospitalized patients with cirrhosis, to provide generalizable findings and also prognostic information. In addition, it is likely that the resistance pattern and antibiotic usage practices in North America are different from the European experience. A consortium of tertiary-care hepatology centers (The North American Consortium for End-Stage Liver Disease [NACSELD]) prospectively collects data on hospitalized patients with cirrhosis who have or develop infections. The overarching purpose of NACSELD is to describe the causes, the pathogenic organisms identified, and the antibiotic resistance patterns emerging in infected and hospitalized North American patients with cirrhosis. The long-term goal of the consortium is to develop interventions that decrease infections, cost of care, and ultimately mortality. The hypothesis was that second infections occur in cirrhosis and are associated with mortality independent of the liver disease severity. The objective of this study was to define the factors predisposing to bacterial infection-related mortality in the NACSELD cohort.

Materials and Methods

After Institutional Board review and approval of the protocol at the clinical sites, patients were prospectively enrolled following written informed consent and data were collected. Data were managed using REDCap (Research Electronic Data Capture) tools hosted at Virginia Commonwealth University.11 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.

Consecutive patients who gave informed consent were included from all sites. However, the variation in the patient number enrolled between sites was due to differing dates of study initiation. We included patients with cirrhosis who were admitted with or developed an infection during their hospitalization. Cirrhosis was defined by either (1) biopsy or (2) clinical evidence of decompensation or varices or (3) radiological evidence of liver nodularity and intraabdominal varices in a patient with chronic liver disease. Data collected included site, patient demographics, comorbid conditions including diabetes mellitus, complications of cirrhosis, use of medications such as proton pump inhibitors (PPI), antibiotics, and beta-blockers, and history of hospitalizations and antibiotic use in the previous 6 months. First-infection details recorded included site of infection, organism isolated and susceptibility, and whether nosocomial, HCA, or community-acquired.12 Infections were defined as follows13: (i) spontaneous bacteremia: positive blood cultures without a source of infection; (ii) spontaneous bacterial peritonitis (SBP): ascitic fluid polymorphonuclear cells >250/μL with/without a positive fluid culture; (iii) lower respiratory tract infections: new pulmonary infiltrate in the presence of: (a) at least one respiratory symptom (cough, sputum production, dyspnea, pleuritic pain) with (b) at least one finding on auscultation (rales or crepitation) or one sign of infection (core body temperature >38°C or <36°C, shivering or leukocyte count >10,000/mm3 or <4,000/mm3) in the absence of antibiotics; (iv) Clostridium difficile: diarrhea with a positive C. difficile assay; (v) bacterial enterocolitis: diarrhea or dysentery with a positive stool culture for Salmonella, Shigella, Yersinia, Campylobacter, or pathogenic E. coli; (vi) skin infection: fever with cellulitis; (vii) urinary tract infection (UTI): urine WBC >15/high-power field with either positive urine gram stain or culture in a symptomatic patient; (viii) intraabdominal infections: diverticulitis, appendicitis, cholangitis, etc.; (ix) secondary bacterial peritonitis: >250 polymorphonuclear cells/μL of ascitic fluid in the presence of an intraabdominal source of peritonitis and multiple organisms cultured from ascitic fluid. We excluded patients with an unclear diagnosis of cirrhosis; patients admitted electively for surgery, interventional radiology, or other scheduled procedures; and immune-compromised patients with human immunodeficiency virus (HIV infection, prior transplant, or disseminated malignancies.

First infections were defined as community-acquired if they were diagnosed within 48 hours of admission without hospitalizations in the previous 6 months; HCA if they were diagnosed within 48 hours of admission in patients with hospitalization for at least 2 days in the previous 6 months; and nosocomial if the infection was diagnosed beyond 48 hours of admission. Second infections were defined as an infection separate from and following the first infection during the same hospitalization. These second infections would be nosocomial in origin.

Other data recorded included admission vital signs (first set that was recorded), sequential organ failure (SOFA) score, hospital course, antibiotics given, in-hospital complications of cirrhosis, and procedures performed during the hospitalization. The details obtained on second infections included the patient location at diagnosis, the site of infection, if procedures were temporally related, and the antibiotics used before and following the diagnosis of second infection. The outcomes recorded were death during admission or within 30 days of the first infection diagnosis; discharge to a facility, hospice, or home; and liver transplantation. Standard laboratory tests were performed at each facility as part of the standard of care. The variables studied were complete blood counts, serum sodium, model for endstage liver disease (MELD) score variables, and serum albumin.

All data was collected prospectively and entered by the site principal investigators and coordinator. The site principal investigators were responsible for accuracy of data entry of the symptoms, adjudication of nosocomial, HCA, community-acquired infections, and other clinical aspects. They were also responsible for auditing their site's data against the patients' medical records to ensure correct entry.

Statistical Analysis.

Data were analyzed using SAS v. 9.2. Patients who died during index hospitalization or within 30 days of the first infection diagnosis were compared with patients who survived >30 days after the infection. We also compared the demographics, severity of liver disease, and infection-related variables between those with community-acquired, HCA, and nosocomial first infections. The clinical characteristics of patients who developed a second infection were compared with those who remained free of second infections. Results are expressed as mean (standard deviation [SD]) unless specified otherwise. When comparing groups, a chi-square or Fisher's exact test was used for categorical variables, whereas for continuous variables t-tests were used. The determinants of mortality were calculated using a logistic regression model. Univariate analysis was performed to determine predictors of death. The variables analyzed were age; gender; etiology of cirrhosis; diabetes; admission MELD, albumin, sodium; admission mean arterial blood pressure, heart rate, SOFA score, use of PPIs, antibiotics such as rifaximin or others used for SBP prophylaxis, and nonselective beta-blockers; organism Gram stain to generate a model with variables available at baseline. We also examined another model with additional variables of length-of-stay and second infection that were evident during hospitalization. All variables with P < 0.25 on univariate analysis were included in the stepwise logistic regression analysis. Backwards elimination was performed to arrive at a parsimonious model where all variables included in the multivariate logistic regression model were significant at the 0.25 level or less. 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 and the fit of this model was assessed using the Hosmer and Lemeshow goodness of fit test.


HCA, healthcare-associated; MELD: model for endstage liver disease; NACSELD, North American Consortium for the Study of End-Stage Liver Disease; SBP, spontaneous bacterial peritonitis; SIRS, systemic inflammatory response syndrome; SOFA, sequential organ failure; UTI, urinary tract infection.


A total of 207 patients with cirrhosis were included in the analysis from December 2010 to December 2011. The demographics and clinical details are displayed in Table 1. Patients were equally distributed across North America: the five leading enrollment sites were Virginia Commonwealth University and VA Medical Center, Richmond (31%); Baylor University Medical Center, Dallas (24%); University of Pennsylvania, Philadelphia (16%); Mayo Clinic, Rochester (12%); and University of Toronto (10%).

Table 1. Baseline Variables of the Included Population
Variable, Mean (%)Patient Information (n = 207)
Gender (men/women)124/ 83
 Caucasian171 (82.6%)
 African-American27 (13.0%)
 Asian6 (2.9%)
Other3 (1.4%)
Ethnicity (Hispanic)11 (5.3%)
Etiology of cirrhosis 
 Hepatitis C alone52 (25%)
 Alcohol alone63 (31%)
 Hepatitis C and alcohol29 (14%)
 Cryptogenic or NASH31 (15%)
 Other32 (15%)
Diabetes63 (30%)
Median hospitalizations within 6 months (range)2 (0-32)
Median infections within 6 months (range)1 (0-8)
Medications at time of admission 
Beta-blockers88 (44%)
Proton pump inhibitors111 (55%)
SBP prophylaxis45 (22%)
Rifaximin72 (36%)
Admission Mean Value (SD) 
Age55 (9)
Mean arterial blood pressure (mmHg)82.7 (15.8)
Mean heart rate (beats/minute)88.1 (16.2)
Mean SOFA score13.8 (33.9)
Serum albumin2.7 (0.7)
Serum sodium132.3 (6.3)
Serum creatinine1.5 (1.1)
Serum total bilirubin6.2 (6.3)
INR1.7 (0.7)
MELD score20 (8)

In the 6 months preceding enrollment, 85% of patients had been hospitalized at least once and 51% experienced an infection requiring an emergency room visit or hospitalization. In patients with prior infections, 36% received beta-lactam antibiotics. Most patients had complications prior to their infection: hepatic encephalopathy (63%), ascites (79%), hyponatremia (61%), and variceal hemorrhage (28%). Medications at the time of admission are listed in Table 1. SBP prophylaxis was with fluoro-quinolones (75%) or trimethoprim-sulfamethoxazole (25%).

First Infections.

The most common first infection was UTI (52 or 25%), followed by SBP (47 or 23%), spontaneous bacteremia (43 or 21%), skin (27 or 13%), and lower respiratory infections (16 or 8%) (Table 2). C. difficile was seen in 10 (5%) patients and other infections made up the remainder (Fig. 1). Twenty-eight patients (14%) had community-acquired, 147 (71%) had HCA, and 32 (15%) had nosocomial infections. The nosocomial infections were predominantly UTI and C. difficile followed by relatively smaller proportions of cases of SBP, spontaneous bacteremia, respiratory, and skin infections. Of the infections diagnosed, there was an increased likelihood of UTI (P = 0.005) and C. difficile (P = 0.001) as the first nosocomial infection, whereas skin infections (P = 0.035) or SBP (P = 0.01) were less likely to be diagnosed as nosocomial infections.

Figure 1.

Distribution of percentage of individual first infections that were community-acquired, healthcare-associated, or nosocomial infections. A significantly higher percentage of the nosocomial first infections were UTI and Clostridium difficile, whereas a significantly lower percentage was due to skin infections and SBP. CA: community-acquired, HCA: healthcare-associated, Noso: nosocomial.

Table 2. Distribution and Characteristics of the First Infection
Site of InfectionCommunity- AcquiredHealthcare- AssociatedNosocomialTotal First Infections
(n = 28)(n = 147)(n = 32)
UTI5331452 (25%)
SBP837247 (23%)
Spontaneous bacteremia635243 (21%)
Respiratory210416 (8%)
Skin/soft-tissue422127 (13%)
C. difficile04610 (5%)
Others36312 (5%)
Type of Organism 
 Gram-positive8511372 (35%)
 Gram-negative646759 (28%)
 Fungus0347 (3%)
 Others1405 (2%)
 Negative culture1343864 (31%)

Organisms Isolated with First Infections.

An organism was isolated in 69% of infections with gram-positive bacteria accounting for 72 (35%) and gram-negative bacteria accounting for 59 (29%) isolates. The remainder included fungi in seven (3%) and others in five (2%) isolates. Gram-positive organisms were Streptococci (n = 19.26%), methicillin-sensitive Staphylococci (n = 12.17%), methicillin-resistant Staphylococcus aureus (n = 7,10%), Enterococcus (n = 19.26%: vancomycin-resistant Enterococcus [VRE] 47% and vancomycin-sensitive 53%), C. difficile (n = 10.14%), and gram-positive bacilli (n = 6,8%). Among gram-negative infections, the isolates were predominantly E. coli (n = 29, 49%-21% with extended beta-lactamase resistance [ESBL] and 21% with fluoroquinolone resistance). The remaining were Klebsiella (n = 14, 24%-43% with fluoroquinolone resistance), followed by Enterobacter cloacae (n = 4,7%, 50% with ESBL), Acinetobacter (n = 3, all meropenem-sensitive), Serratia (n = 2), Proteus (n = 2), Citrobacter (n = 2), Gardnerella (n = 1), Pasteurella (n = 1), and Salmonella (n = 1). Therefore, among first infections VRE was seen in 10 isolates, MRSA in 7, fluoroquinolone resistance in 18, and ESBL in six isolates. All fungal infections were Candida-related (three were UTI, two were soft-tissue, one pneumonia and one septicemia).

Hospital Course.

Transfer to the intensive care unit (ICU) was required for 72 (35%) patients. Twenty percent (n = 41) of patients developed septic shock. The mean length-of-stay (LOS) of the entire cohort was 11.9 ± 13.5 days. Complications that developed included hepatic encephalopathy (61%), gastrointestinal bleeding (22%), and hepatorenal syndrome (25%); renal replacement therapy was required in 17% of the population. Almost half (47%) of patients required central venous line placement, 38% underwent a large-volume paracentesis or thoracentesis, 19% needed mechanical ventilation, and only 5% were initiated on parenteral nutrition.

Second Infections.

A total of 50 (24.1%) patients developed a second infection during hospitalization, a median of 5 days after the first infection (Table 3; Fig. 2). Contracting second infections was equally likely whether the patients were on the medical ward (n = 28, 56%) or in the ICU (n = 22, 44%), P = 0.12. The majority of second infections were respiratory (n = 14, 28%), UTI (n = 13, 26%), and C. difficile (n = 6,12%). Respiratory second infections were commonly associated with aspiration (n = 6) and mechanical ventilation (n = 4) and only two organisms were isolated (Klebsiella and an unspeciated gram-negative bacillus). In contrast, all UTIs were culture-positive and related to bladder catheterization in six cases. The leading causes of UTI were Enterococcus (VRE n = 2 and vancomycin-sensitive, n = 2), Candida (n = 4), and gram-negative organisms (E. coli [n = 1], Enterobacter [n = 1], Klebsiella [n = 1], Pseudomonas [n = 1] and an unspeciated gram-negative bacillus [n = 1]). The overall distribution of organisms isolated was equally between gram-positive and -negative organisms. Gram-positive organisms were seen in 19 infections (seven VRE, two vancomycin-sensitive, three MRSA, one Streptococcus viridans, and six C. difficile), gram-negatives in 10 (three E coli, two Klebsiella, one each of Pseudomonas, Acinetobacter, Enterobacter; the remaining were unspeciated gram negative bacilli), Candida in seven; whereas the remaining 14 did not have any organism isolated. The seven patients with a fungus as a cause of their second infections included four with UTI and three with esophagitis. There was no difference in the distribution of gender, race, ethnicity, or etiology of cirrhosis in those who developed a second infection compared to those who did not. In addition, the site of prior infection did not predict the development of a second infection. However, second infections were more frequently seen in patients in whom the first infection was nosocomial (n = 16, 50%) compared to HCA (n = 30, 20%), or community-acquired first infections (n = 4, 14%, P = 0.004). Further, second infections more often occurred in patients in the ICU (43% versus 15%, P < 0.0001), patients with shock (47% versus 20%, P < 0.0001), those requiring renal replacement therapy (46% versus 21%, P = 0.003), and those with hepatic encephalopathy (30% versus 17%, P = 0.024). Second infection rates were also significantly higher in patients who had central lines (37% versus 14%, P < 0.0001) and required mechanical ventilation (46% versus 19%, P < 0.0001). There was no significant increase in the rate of second infections in patients who underwent large volume paracentesis, had parenteral nutrition, or those who experienced gastrointestinal bleeding during their hospitalization. Of note, the infection rates due to VRE (14% versus 3.5%, P < 0.003), C. difficile (12% versus 5%, P = 0.003), and fungi (14% versus 3.5%, P < 0.0001) were higher among patients with second infection when compared to first infections.

Figure 2.

Percentage of infections (first and second) according to body site. The percentage distribution of first and second infections is shown. Blue bars represent the percentage of the first infections by site, whereas red bars represent the second infections. There is a significantly higher percentage prevalence of respiratory (P < 0.0001), fungal (P = 0.003), and C. difficile (P = 0.05) in the second infections compared to the first, whereas the opposite was true for SBP (P = 0.045), skin infections (P = 0.02), and spontaneous bacteremia (P = 0.036). UTI percentage prevalence was similar in the first and second infections (P = 0.89). Resp: lower respiratory tract infections, UTI: urinary tract infections, Bact: spontaneous bacteremia.

Table 3. Comparison Between Patients Who Did or Did Not Develop a Second Infection
 Second Infection (n = 50)No Second Infection (n = 157)P-value
Age54.8 (8.7)55.4 (9.3)0.69
Admission MELD22.1 (8.1)19.6 (7.6)0.06
Admission albumin2.6 (0.6)2.7 (0.7)0.14
Admission serum sodium132.6 (6.3)132.1 (6.4)0.66
SOFA score14 (32)14 (37)0.99
On beta-blockers17 (35%)71 (45%)0.12
On PPIs28 (56%)83 (53%)0.83
On rifaximin18 (36%)54 (34%)0.98
On SBP prophylaxis10 (20%)26 (17%)0.48
Median infections in the past 6 months110.31
Nosocomial first infection16 (32%)16 (10%)<0.0001
Length-of-stay (days)16.1 (12.7)10.6 (13.5)0.013
First Infection Organism   
 Gram-positive18 (36%)54 (34%)0.90
 Gram-negative18 (36%)41 (26%)0.21
 Fungus4 (8%)3 (2%)0.36
 No organism/others10 (20%)59 (38%)0.10


Forty-nine patients (23.6%) died during hospitalization or within 30 days of discharge. The first infection case fatality rate was the highest for patients with C. difficile (40%, 4/10), followed by respiratory infections (37.5%, 6/16), and spontaneous bacteremia (37%, 16/43). The case fatality rate was lower for skin infections (22%, 6/27), UTIs (15%, 8/52), and SBP (17%, 8/47). Patients with higher MELD scores had higher death rates (Table 4). No differences in the first infection bacterial isolates were identified between survivors and those who died (gram-positive organisms in 19% of survivors versus 17% of those who died, P = 0.80). Of note, there was no effect of MRSA (0% versus 23%, P = 0.35), VRE (30% versus 22%, P = 0.70), or fluoro-quinolone resistance (15% versus 23%, P = 0.73) on mortality related to the first infection. However, the occurrence of a second infection (regardless of the origin of the first infection) was highly predictive of mortality (P < 0.00001) (Table 4).

Table 4. Comparison Between Patients Who Died or Survived
(n = 49)(n = 158)
Age54.9 (8.6)55.3 (9.5)0.77
Admission MELD25.4 (7.5)18.5 (7.2)<0.0001
Admission albumin2.4 (0.6)2.8 (0.7)0.002
Admission serum sodium132.5 (6.4)131.3 (6.1)0.23
SOFA score9.2 (5.4)9.0 (7.7)0.86
On beta-blockers17 (35%)71 (45%)0.15
On PPIs24 (49%)87 (55%)0.34
On rifaximin21 (43%)51 (32%)0.23
On SBP prophylaxis10 (20%)35 (22%)0.81
Nosocomial first infection10 (20%)22 (14%)0.28
Length-of-stay16.1 (12.7)10.6 (13.5)0.013
Events During Admission   
Second infection24 (49%)26 (16%)<0.0001
Intensive care unit37 (76%)35 (22%)<0.0001
Hepatic encephalopathy44 (90%)90 (57%)<0.0001
Variceal bleeding15 (31%)33 (31%)0.08
Mechanical ventilation31 (63%)10 (6%)<0.0001
Hepato-renal syndrome23 (47%)31 (20%)<0.0001
Renal replacement therapy14 (29%)23 (15%)0.019
Central line39 (80%)63 (40%)<0.0001
Parenteral nutrition5 (10%)8 (5%)0.08
Large volume paracentesis21 (43%)62 (39%)0.38

Using the regression model at baseline, the significant predictors of mortality from admission were MELD score (odds ratio [OR] = 1.123, 95% confidence interval [CI] 1.067-1.1182, P < 0.0001), admission heart rate (OR = 1.028, 95% CI 1.006-1.052, P = 0.014), and admission serum albumin (OR = 0.490, 95% CI 0.268-0.896, P = 0.021). The R-square value was 0.1787, the −2log likelihood was 184.782, and the Goodness of fit test (χ2 = 10.7192. 8 degrees of freedom [d.f.], P = 0.2181) indicated a good fit.

When the second infection and LOS were added, the model now consisted of admission MELD score (R = 1.124, 95% CI 1.064-1.1888, P < 0.0001), second infection (OR = 4.416, 95% CI 2.007-9.713, P = 0.0002), admission heart rate (OR = 1.028, 95% CI 1.004-1.053, P = 0.02), and admission serum albumin (OR = 0.503, 95% CI 0.262-0.969, P = 0.04). The goodness of fit test (Hosmer and Lemeshow) was χ2 = 4.708, 8 d.f., P = 0.7882, −2log likelihood of 169.888 indicating a good fit for the database. This model was significantly improved in its prediction capacity compared to the model solely based on baseline values (−2log likelihood 184.782 − 169.888 = 14.894, P < 0.0001). Therefore, the final regression equation to predict the percentage chance of death was calculated as the Mortality Score = −4.9689 + 0.1169 (Admission MELD) + 0.0279 (Admission Heart Rate) −0.6863 (Admission Serum Albumin) + 1.4852 (Second Infection, as yes or no). The interpretation of this mortality score is that the larger the value of this mortality score, the greater the estimated probability of death. Full results from this model are presented in Table 5.

Table 5. Final Predictive Model Results
ParameterEstimateStandard ErrorOROR 95% CIP-value
  1. OR: odds ratio; CI: confidence interval.

MELD0.11690.02811.124(1.064, 1.188)< 0.0001
Albumin−0.68630.33410.503(0.262, 0.969)0.0399
Heart rate0.02790.01201.028(1.004, 1.053)0.0202
Second infection1.48520.40224.416(2.007, 9.713)0.0002


In this multicenter North American cohort study, infections in patients with cirrhosis are associated with a high risk of mortality. Importantly, we found that despite the presence of guidelines for reduction of inpatient infections, nosocomial and potentially preventable second infections in patients with cirrhosis continue to occur, and are associated with a significantly high mortality that is independent of their liver disease severity.

HCA and nosocomial infections have been implicated as a major cause of morbidity and mortality in cirrhosis, with multidrug-resistant organisms leading to a worse prognosis.2, 3 These conclusions are drawn from single-center studies and, therefore, their results cannot be reliably extended to multicenter and international cohorts. We confirmed in a multicenter North American cohort that the majority of first infections were HCA in origin and were most likely to be SBP and UTI. However, in contrast to prior studies, we did not find a high case-fatality rate for either of these infections when compared to other first infections such as respiratory infections, spontaneous bacteremia, and C. difficile.14 This low fatality rate for SBP may be due to the effectiveness of guidelines recommending a low threshold for diagnostic paracentesis and prompt initiation of antibiotics and albumin therapy in affected patients.15 The higher risk of death in our patients with C. difficile is supported by the conclusions of a large national database where patients with cirrhosis with C. difficile had a significantly higher mortality compared to noncirrhosis patients with C. difficile and patients with cirrhosis without C. difficile.7 This could be reflective of the altered intestinal microbiome in patients with cirrhosis or deficiencies in their mucosal immunity.16-18

Surprisingly, we also did not find a difference in mortality rate between community-acquired, HCA, or nosocomial infections. This is despite the expanded definition of HCA infections (extending to 6 months prior to admission).19 We also did not find a difference in VRE, MRSA, and fluoroquinolone resistance between those with community-acquired, HCA, or nosocomial first infections. This finding may be due to the breadth of our sample or because of the generally high use of antibiotics in the community.

The most significant determinants of mortality were the severity of underlying liver disease indicated by the MELD score and the development of a second infection. Liver-specific severity scales such as MELD score have been compared to nonspecific scales such as SOFA to prognosticate outcome in patients with cirrhosis in the ICU.20, 21 Whereas the conclusions have been inconsistent in prior studies in determining superiority of one score over the other, our results showed that MELD was superior to SOFA score in the prediction of mortality. This may be because only 35% of our patients required ICU admission. Therefore, underlying liver disease severity judged by MELD score and albumin may be more relevant in the prognostication of hospitalized patients with cirrhosis.

Interestingly, we found the development of a second infection during hospitalization to significantly increase the risk of mortality independent of the MELD score. A mortality model including admission MELD, albumin, heart rate (despite the use of beta-blockers in 44% of patients) along with second infection strongly predicted mortality. The concept of “second infections” has been introduced because clinical practice led to the observations that hospitalized patients with cirrhosis often struggle with sequential unrelated infections that may impact survival. Ironically, patients with sepsis may develop a state of immunosuppression further exacerbating the risk of second infections.22 Second infections were often C. difficile, but pneumonias and UTIs were other common causes. As expected, we found a significantly higher prevalence of second infections in those who had a nosocomial first infection, and in those with extended lengths of hospitalization. Not surprisingly, the pathogens responsible for the second infections tended to be multidrug-resistant bacteria or fungi.3 Of note, second infections were associated with therapeutic maneuvers such as urinary tract catheterization, mechanical ventilation, and antibiotics (emergence of C. difficile). Judicious use of these interventions and adequate precautions for prevention of infection could potentially reduce this incidence.23 We also found that aspiration pneumonia was a significant cause of second infections. This can be potentially prevented by careful airway monitoring of patients with cirrhosis with altered mental status and those with upper gastrointestinal bleeding. The study, therefore, points us toward management strategies aimed at limiting the use of instrumentation and antibiotics and encouraging early mobilization and airway protection to prevent aspiration. Specific antibiotics might be initiated in patients at high risk for developing second infections. Many of these infections occurred on the medical wards; therefore, hypervigilant surveillance for second infections, especially fungal, should not be restricted to patients with cirrhosis admitted to the ICU. We strongly suggest that when patients with cirrhosis have a decline in their condition, especially when they are in the ICU or if they have prolonged hospitalization, a search should be made for a second infection, especially respiratory, urinary, fungal, or C. difficile. Therefore, despite guidelines being in place in most hospitals to prevent second infections, our multicenter experience has shown an unacceptably high rate of these potentially preventable infections.

Our study is the largest North American experience of infections in cirrhosis that has prospective data collection and involves multiple centers. This will increase the generalizability of the results within this population. The study is limited by the use of different practice patterns over the different centers and also by the lack of a control group without infections. Future studies with control groups without infections are warranted to improve prediction of infection development.

Through our large, multicenter prospective dataset we have demonstrated that second infections are independent predictors of mortality above and beyond the severity of liver disease as determined by the MELD score. Therefore, a better understanding of the altered immune mechanisms, which confer increased susceptibility to infections and the early prevention of these second infections, is key to reducing infection related mortality in cirrhosis.


The following centers and their staff are also part of NACSELD: University of Colorado, Denver, CO: Scott Biggins, MD, and Kiran Bambha, MD (co-PIs) with John Nguyen (coordinator); Mayo Clinic, Rochester, MN: Irakli Kalaoni (Clinical Research Fellow); University of Pennsylvania, Philadelphia: Aaron Blouin (coordinator); University of California, San Diego, CA: Heather Patton, MD (PI) with Natali Navarro (coordinator); Emory University, Atlanta GA: Elizabeth Ferry (coordinator); University of Toronto, Toronto, ON, Canada: Martha Orgill (coordinator); Beth Israel Deaconess Medical Center, Boston, MA: Raza Malik, MD (PI) and Parham Safaie (Clinical Research Fellow); Baylor University Medical Center, Dallas, TX: Arlen Waclawczyk (coordinator); University of Texas, Southwestern Medical Center, Dallas, TX: Bingru Xie. MD (Co-PI); University of Texas Medical Center, Houston, TX: Michael B. Fallon, MD (PI) with Teresa Ramirez (coordinator).