Prudent Use of Blood Cultures for Hospitalized Patients With Cirrhosis

Background No reliable risk stratification method is available to guide the extent of infectious work-up among hospitalized patients with cirrhosis. Therefore, we aimed to create a risk stratification method for obtaining blood cultures from hospitalized patients with cirrhosis. Methods This was a retrospective cohort study using the Healthcare Cost and Utilization Project - National Readmission Database 2019. Adult patients who were not immunocompromised comprised the final cohort. The primary outcome was the incidence of bacteremia among hospitalized patients with cirrhosis. Secondary outcomes included length of hospital stay, inpatient mortality, and 30-day readmission rate among cirrhosis patients with and without bacteremia. After propensity score matching, the χ2 test was used to assess the primary outcome and inpatient mortality. The Wilcoxon signed-rank test was used to compare the length of hospital stay. Readmission rates were compared via survival analysis. Concomitant bacterial infection, cirrhosis causes, and complications were assessed as potential risk factors for bacteremia using binomial regression. Results The risk ratio (RR) of bacteremia was 1.66 (95% confidence interval (CI): 1.55-1.78) among patients with cirrhosis compared to those without cirrhosis. A concomitant bacterial infection was found to have a strong association with bacteremia in patients with cirrhosis (RR: 3.3, 95% CI: 3.03-3.59). Among cirrhosis patients without concomitant bacterial infection, the incidence of bacteremia was 0.76% (<1%). Among the causes of cirrhosis, primary sclerosing cholangitis was found to have a strong association with bacteremia (RR: 3.88, 95% CI: 2.3-6.04, P < 0.001). Patients with cirrhosis who had bacteremia were hospitalized three days longer than those without bacteremia. There was no difference in inpatient mortality or 30-day readmission rates between cirrhotic patients with and without bacteremia. Conclusion This study suggests that, in the absence of another concomitant bacterial infection and primary sclerosing cholangitis, we can avoid unnecessary blood cultures among immunocompetent patients with cirrhosis. However, given some inherent limitations associated with the database (such as the unavailability of vitals or laboratory values), additional studies are needed to validate its findings.


Introduction
Liver cirrhosis is the most common chronic liver disease, with a significant mortality burden [1].In 2021, chronic liver disease and cirrhosis were ranked ninth by the Centers for Disease Control in terms of mortality in the United States.Among hospitalized patients with cirrhosis, the incidence of infection has been reported to be approximately 25-35%, with bacteria being the most common pathogen [2].Bacterial infections in patients with cirrhosis are associated with higher morbidity and mortality when compared to patients without cirrhosis [3].Therefore, infectious work-up and timely management of infections are of paramount importance among hospitalized patients with cirrhosis.
Common bacterial infections among patients with cirrhosis include urinary tract infection (UTI), spontaneous bacterial peritonitis (SBP), community-acquired pneumonia, and bacteremia [4].Bacteremia has been associated with numerous concomitant infections, including SBP, UTI, community-acquired pneumonia, and others [2,3].Among bacterial infections, bacteremia has been known to be associated with the worst morbidity, including increased length of stay, excessive cost, and increased mortality [5].

Selection criteria
The inclusion criteria included all patients aged 18 years and older.Immunosuppression independently increases the risk of infection, including bacteremia; therefore, those patients who had a condition that warranted immunosuppression (for instance, renal transplant status) or in whom the condition itself suggested immunosuppression (for instance, neutropenia) or who were using medications that can cause immunosuppression (such as biologic therapy) were excluded from this study.A detailed list of conditions or therapies that suggest or cause immunosuppression and were used to exclude patients is given in the Appendix.
Since autoimmune hepatitis, primary biliary cholangitis, and primary sclerosing cholangitis can cause cirrhosis, patients with these conditions were not excluded.Binomial regression was used to assess the association between bacteremia and the cause of cirrhosis, including autoimmune hepatitis, primary biliary cholangitis, and primary sclerosing cholangitis, as further discussed below.

Groups
Once the final cohort was selected, patients were divided into two groups: patients with cirrhosis and patients without cirrhosis.

Outcomes
The primary outcome was the incidence of bacteremia among hospitalized patients with cirrhosis when compared to patients without cirrhosis.The secondary outcomes included: (1) length of hospital stay of cirrhosis patients with bacteremia and without bacteremia; (2) incidence of inpatient mortality among cirrhosis patients with bacteremia and without bacteremia; and (3) 30-day readmission rates among cirrhosis patients with bacteremia and without bacteremia who were admitted during the first 11 months and were discharged alive.If a patient was admitted during the 12th month, the 30-day readmission outcome could not be calculated for such a patient in the HCUP-NRD 2019.

Hospitalizations
If a patient had more than one hospitalization recorded in the HCUP-NRD 2019, only the first hospitalization was assessed for all outcomes, except the 30-day readmission outcome.For the readmission outcome, only the time to first readmission was assessed (if there was a readmission encounter/hospitalization).

Statistical analysis
In the study population, patients who had a diagnosis of cirrhosis were identified.Patients with cirrhosis were then matched with patients without cirrhosis on age, sex, and baseline comorbidities, including congestive heart failure, ischemic heart disease, hypertension, hyperlipidemia, diabetes mellitus, chronic kidney disease/end-stage renal disease, chronic obstructive pulmonary disease/emphysema, and asthma.
Propensity score matching (PSM) with the nearest neighbor method at a 1:1 ratio and without replacement was used for matching.The caliper was set at 0.1 for PSM.
After PSM was complete, the χ2 test was used to compare the incidence of bacteremia between patients with and without a diagnosis of cirrhosis (primary outcome).Both the incidence rate and risk ratio (RR) were calculated for bacteremia.Among patients with cirrhosis, infections other than bacteremia, upper GI bleeding, ascites, portal hypertension, encephalopathy, hepatocellular carcinoma, hepatorenal syndrome, and hepatopulmonary syndrome were subsequently evaluated as potential risk factors for bacteremia using binomial regression.The association between the cause of cirrhosis and bacteremia was also assessed via binomial regression among patients with cirrhosis.All binomial regression assumptions were assessed, including assumptions of no multicollinearity.To ensure model adequacy, the Hosmer-Lemeshow goodnessof-fit test was also performed.
As the duration of hospital stay did not have a normal distribution, the Wilcoxon signed-rank test was used to compare the duration of hospital stay between cirrhosis patients with and without bacteremia.Inpatient mortality between cirrhosis patients with and without bacteremia was compared via the χ2 test, and the mortality rate and RR were calculated.
Once patients were selected for the 30-day readmission outcome, PSM was used first to match patients with and without bacteremia based on age, sex, baseline comorbidities, and complications of cirrhosis (including upper GI bleeding, ascites, portal hypertension, encephalopathy, hepatocellular carcinoma, hepatorenal syndrome, and hepatopulmonary syndrome).After PSM was complete, survival analysis was performed to compare the readmission rates between cirrhotic patients with and without bacteremia.For survival analysis, a Kaplan-Meier survival curve was constructed, and a Cox proportional hazards model was used to calculate the hazard ratio.The proportional hazards assumption was assessed using the Schoenfeld proportionality test.
The alpha criterion was set at 0.05 for all statistical tests.Weighted analysis was not performed because this study focused on comparative analysis between two groups rather than calculating the national estimates.R software was used for data extraction, data cleaning, and analysis.

Results
There are more than 12 million unique adult patients in the HCUP-NRD 2019.Of those patients, 11.107 million patients were included in the final cohort, as illustrated in Figure 1.

Primary outcome
After PSM was complete for patients with and without cirrhosis, the incidence of bacteremia was compared between the two groups, which was the primary outcome of this study.The summary of the primary outcome is presented in Table 2.

Exploratory analysis results
Out of 176200 patients with cirrhosis who were included for analysis after the PSM, 59% (n = 103840) had at least one important complication secondary to cirrhosis, including ascites, SBP, upper GI bleeding, encephalopathy, hepatorenal syndrome, hepatopulmonary syndrome, hepatocellular carcinoma, or portal hypertension.Besides the presence of infection other than bacteremia, these aforementioned complications were included in the binomial regression to assess their association with bacteremia.In binomial regression, the presence of infection other than bacteremia had a strong association with bacteremia among patients with cirrhosis, with an RR of 3.3 (95% confidence interval (CI): 3.03-3.59,P < 0.001).A summary of this binomial regression is presented in Table 3.The binomial regression model met all assumptions, including the assumption of no multicollinearity, and the model appeared to fit the data adequately (Hosmer-Lemeshow goodness-of-fit test, P > 0.05).Based on the results of the binomial regression, further exploratory analysis was performed to assess the incidence of bacteremia among patients with cirrhosis who had no other infection.The incidence of bacteremia among patients with cirrhosis in the absence of another identifiable infection (such as UTI or SBP) was only 0.76% (n = 972).
Alcohol use was found to be the most common cause of cirrhosis (36.59%, n = 64465), followed by chronic hepatitis C infection (9.93%, n = 17492) and the simultaneous presence of alcohol use and chronic hepatitis C infection (7.6%, n = 13387).The exact etiology of cirrhosis could not be identified in 26.97% (n = 47522) of patients with cirrhosis.A summary of the causes of cirrhosis is provided in Table 4.Among the causes of cirrhosis, primary sclerosing cholangitis was found to have a strong association with bacteremia, with an RR of 3.88 (95% CI: 2.3-6.04,P < 0.001).Chronic hepatitis C infection was also found to have a statistically significant association with bacteremia, but the strength of association was weak, which was likely related to the large power because of the large sample size (RR of 1.19 with a 95% CI of 1.08-1.32).

TABLE 4: Causes of cirrhosis
In 43.63% (n = 972) of cirrhosis patients with bacteremia, another concomitant infection could not be identified in this study.Among the concomitant infections that could be identified along with bacteremia, UTI was the most common infection (12.16%, n = 271), followed by cellulitis/soft tissue infections (9.07%, n = 202) and bacterial pneumonia (8.75%, n = 195).
In 42.41% (n = 945) of cirrhosis patients with bacteremia, the exact bacterial agent responsible for bacteremia could not be identified.Among the remaining 57.59% (n = 1283) cirrhosis patients with bacteremia, the exact bacterial agent could be identified.Among those 1283 cirrhosis patients with bacteremia, in whom the exact bacterial agent could be identified, Staphylococcus aureus (S. aureus) was responsible for bacteremia in 28.60% (n = 367) of patients, followed by all groups of Streptococcus (24.24%, n = 311) and Escherichia coli (E.coli) (23.23%, n = 298).

Secondary outcomes
Cirrhosis patients with bacteremia were hospitalized for a longer duration than patients without bacteremia, with a median hospital length of stay of seven days compared to four days, respectively (median difference: three days, 95% CI: 2.99-3.00,P < 0.001).The inpatient mortality rates for cirrhosis patients with and without bacteremia were 6.78% (n = 151) and 7.17% (n = 12469), respectively.There was no significant difference in mortality between these two groups, with an RR of 0.95 (95% CI: 0.81-1.10,P = 0.48).
Before determining readmission outcomes for patients with cirrhosis who were discharged alive, patients with cirrhosis who had bacteremia and those who did not were matched as well.A summary of this PSM is provided in Table 5.After PSM, survival analyses were performed.As shown in Figure 2, the Kaplan-Meier survival plot does not show any significant deviation in readmission rates between cirrhosis patients with and without bacteremia.

FIGURE 2: Kaplan-Meier survival curve for 30-day readmission outcome between cirrhosis patients with and without bacteremia
The hazard ratio for the 30-day readmission rate between cirrhosis patients with and without bacteremia was also 1.11 (95% CI: 0.95-1.29,P = 0.200).Hence, there was no significant difference between cirrhosis patients with and without bacteremia in regard to the 30-day readmission rate.The proportional hazards assumption was met (Schoenfeld proportionality test, P > 0.05).

Discussion
This study found the risk of bacteremia to be 1.66 times higher among patients with cirrhosis than among those without cirrhosis.The presence of an infection other than bacteremia (such as UTI, SBP, or pneumonia) was found to have a strong association with bacteremia among patients with cirrhosis, with an RR of 3.3 (95% CI: 3.03-3.59,P < 0.001).This association suggests that the higher incidence of bacteremia may be driven by the higher incidence of concomitant bacterial infections among patients with cirrhosis.Among cirrhosis patients without another concomitant infection, the incidence of bacteremia was similar to that among patients without cirrhosis (0.76% vs. 0.76%, respectively).
The most common cause of cirrhosis in this study was alcohol use.Among the causes of cirrhosis, primary sclerosing cholangitis was only found in 0.09% (n = 152) of patients, but it was found to have a strong association with bacteremia, with an RR of 3.88 (95% CI: 2.30-6.04,P < 0.001).Among cirrhosis patients with bacteremia, the most common concomitant infections were UTIs, followed by cellulitis and bacterial pneumonia.The most common bacterial agent responsible for bacteremia that could be identified in our study was S. aureus, followed by Streptococcus (all groups) and E. coli.Patients with cirrhosis who had bacteremia were hospitalized three days longer than those without bacteremia.There was no difference in inpatient mortality or 30-day readmission rates between cirrhosis patients with and without bacteremia in our study.
Patients with cirrhosis are vulnerable to bacterial infections [6][7][8].This is evident by the findings of our study as well, in which bacteremia was noted to have a higher incidence among patients with cirrhosis than among patients without cirrhosis.According to the published literature, bacterial infections account for at least 25% of hospital admissions among patients with cirrhosis and are considered an important factor in decompensation [9,10].Potential causes that lead to a higher incidence of bacterial infections among patients with cirrhosis include altered intestinal permeability, changes in gut microflora, and alterations in innate immunity and humoral immunity [9,10].Other described mechanisms include decreased hepatic complement production, impaired Kupffer cells, altered neutrophil chemotaxis, and downregulated HLA-DR expression on monocytes [11,12].
Bacterial infections in cirrhosis are associated with a high incidence of sepsis and related mortality [13].Other factors linked with hospital mortality in cirrhosis include hepatorenal syndrome, hepatic encephalopathy, advanced age, renal insufficiency, and admission to the intensive care unit [14].High white blood cell count, multi-organ failure, and initiation of dialysis for renal failure in liver failure also have a grave prognosis.Our study also found that patients with cirrhosis who had UTIs, cellulitis, pneumonia, SBP, or any other bacterial infection had a significantly increased risk of bacteremia.However, this study did not find any difference in inpatient mortality among cirrhosis patients with and without bacteremia.This may be the result of the broader use of empiric antibiotics for patients with cirrhosis in the recent past.Similarly, we (the authors) also believe that the broader use of empiric antibiotics among cirrhosis patients with GI bleeding may be the reason why those patients with GI bleeding had a low incidence of bacteremia in our study.However, we could not ascertain this fact because the HCUP-NRD does not provide data on medication administration.
Gram-negative bacteria, namely E. coli, Klebsiella, and Aeromonas have been the most reported organisms in culture-positive hospitalized patients with cirrhosis (>70%) [6,15].Gram-positive bacteria, predominantly Streptococcus and Staphylococcus, were reported in 21.2% of patients in a relevant study by Kuo et al. [15].Among the identified pathogens in our study, S. aureus and Streptococcus (all groups) represented the most common pathogens responsible for bacteremia in patients with cirrhosis.Collectively, our study findings do correlate with those of previous studies in which E. coli, Staphylococcus, and Streptococcus represent the bulk of causative agents responsible for bacteremia among patients with cirrhosis.
In the literature, common sources linked to bacteremia are SBP, UTI, pneumonia, and biliary tree infection [4,16].Multiple other studies have mentioned UTI as a predictor of other bacterial infections in patients with cirrhosis [17][18][19].In a study by Choudry et al. [20], Gram-positive bacteria were prevalent in patients with SBP, whereas Gram-negative organisms were isolated in UTIs and pneumonia.Although SBP has been widely reported to be one of the leading causes of bacteremia among patients with cirrhosis, this was not the case in our study.In our study, the most common infections associated with bacteremia in patients with cirrhosis were UTIs, followed by cellulitis and bacterial pneumonia.This is in part in agreement with the findings of previous studies, as this study agreed on UTI and pneumonia as important risk factors for bacteremia.
Because of the higher incidence of bacterial infections and associated worse outcomes, a broad infectious work-up is usually ordered for all patients with cirrhosis on admission, including blood cultures, even if the patient does not have evidence of any concomitant infection.Although it is of the utmost importance to perform appropriate infectious work-up, currently no reliable stratification method exists to guide the proper utilization of blood cultures among hospitalized patients with cirrhosis.Our study undertook this challenge to provide at least some guidance on the appropriate use of blood cultures for patients with cirrhosis.If there is no concomitant infection, the incidence of bacteremia is less than 1%, according to the findings of our study.Therefore, among patients with cirrhosis who do not have any objective evidence of concomitant bacterial infection (such as SBP, UTI, pneumonia, or cellulitis), who are not immunocompromised, in whom the cause of cirrhosis is not primary sclerosing cholangitis, and who do not have any other signs/symptoms to suggest an infection (such as fever), we can avoid unnecessary blood cultures.
There were limitations to this study.This study was based on the HCUP-NRD, which does not provide data on vitals, laboratory values, or medication administration.This limited our ability to determine the severity of cirrhosis (as we could not calculate the model for End-Stage Liver Disease or the Child-Pugh score).To counter this limitation, the International Classification of Diseases, 10th Revision codes were used to identify important complications of cirrhosis, including ascites, portal hypertension, encephalopathy, hepatocellular carcinoma, hepatorenal syndrome, and hepatopulmonary syndrome.Since these complications suggest decompensated cirrhosis, binomial regression was used to assess their association with bacteremia, as summarized in Table 3.We expected some variability in billing codes for the same diagnosis between different providers; for example, hypertension may be billed as I10 or I16.To counter this shortcoming, we used meticulously defined computable phenotypes, as detailed in the Appendix.The HCUP-NRD also does not provide data on out-of-hospital mortality.

Conclusions
Among patients with cirrhosis who do not have any objective evidence of concomitant bacterial infection, who are not immunocompromised, in whom the cause of cirrhosis is not primary sclerosing cholangitis, and who do not have any other signs/symptoms to suggest an infection (such as fever), we can avoid unnecessary blood cultures.This can avoid unnecessary discomfort for the patients, reduce unnecessary costs, and reduce the length of the hospital stay.However, considering the inherent limitations of this study, providers should exercise caution, especially if a cirrhosis patient has severe disease (e.g., Child Class C or Model for End-Stage Liver Disease -Sodium Score >24).Further prospective studies are crucial for the validation of the conclusion of this study, and in those studies, it will be important to include data on the severity of liver disease as well as prior antibiotic use.

Immunosuppression status
If the patient had any of the ICD-10 codes listed in Table 7 during any hospitalization in 2019, then it was considered that the patient was likely immunosuppressed.For example, Mr. C was admitted to the hospital in March 2019 and July 2019.There was no diagnostic code for liver transplant status (as detailed in Table 7) in the March 2019 hospitalization, but the patient had a billing code of Z94.4 in July 2019.In this example, Mr. C was considered to have a history of liver transplants in 2019 (which requires immunosuppression) and, therefore, was excluded from the study.Since autoimmune hepatitis, primary biliary cholangitis, and primary sclerosing cholangitis can cause cirrhosis, these patients were not excluded.

Baseline comorbidities
Baseline comorbidities that have been included in this study are congestive heart failure, ischemic heart disease, hypertension, hyperlipidemia, diabetes mellitus, CKD/ESRD status, chronic obstructive pulmonary disease (COPD)/emphysema, and asthma.
The presence of each baseline comorbidity is defined by the presence of any ICD-10 code from

Causes of cirrhosis
The cause of cirrhosis was identified via the presence of any ICD-10 code from Table 9 during any hospitalization in 2019.

Bacteremia and causative agent of bacteremia
Just like complications related to cirrhosis, only the first hospitalization was assessed for the presence of any billing codes listed in For bacteremia, two billing diagnostic codes are used simultaneously.For instance, if a patient has Streptococcus (group A) bacteremia, the billing code of R78.81 (for bacteremia) is combined with B95.0 (to specify the type of bacteremia).

TABLE 3 : Summary of binomial regression for potential risk factors for bacteremia among patients with cirrhosis
* present CI: confidence interval; RR: risk ratio; GI: gastrointestinal

Table 8
during any hospitalization in 2019.For example, Ms. D was hospitalized in January 2019, May 2019, and then again in September 2019.Only during the May 2019 hospitalization did Ms. D have a billing code of I10.Ms.D didn't have any of the billing codes for hypertension (as listed in Table8) during her hospitalization in January 2019 and September 2019.In this case, Ms. D was considered to have hypertension because she had a billing code for hypertension at least during one hospitalization in 2019.

TABLE 8 : ICD-10 codes for baseline comorbidities
If the patient has many chronic comorbidities and is stable, sometimes clinicians don't bill for all comorbidities.In contrast to the definition of cirrhosis (as mentioned above), it was the reason why more sensitive but less specific criteria were used to define each baseline comorbidity.For instance, Ms. D has a history of cirrhosis, congestive heart failure, ischemic heart disease, CKD, hyperlipidemia, and hypertension.Ms. D was admitted for acute kidney injury on CKD each time in 2019.Ms. D's blood pressure was within normal range with only one anti-hypertensive medication during all of her three hospitalizations in 2019.A clinician didn't bill for hypertension in January 2019 and September 2019, but another clinician decided to bill for it in May 2019.This way, Ms. D ended up having a billing code for hypertension only in May 2019.In this scenario, Ms. D was considered to have hypertension for our study.

TABLE 12 : ICD-10 codes for bacteremia and causative agent of bacteremia
Table 12 to identify bacteremia, and if bacteremia is present, what is the potential causative agent?