Background

Bloodstream infection (BSI) is one of the infectious syndrome with the highest health burden and is associated with important costs to the healthcare system, [1,2,3,4]and in the last years the incidence of Enterobacterales BSI has been increasing [5,6,7,8,9,10,11,12,13].

Studies estimate that overall mortality in K. pneumoniae BSI can range from 20 to 54%, being particularly high in KPC-producing K. pneumoniae infections [12, 14,15,16,17,18,19,20]. Enterobacter cloacae BSI mortality seems to be lower than K. pneumoniae, ranging from 19 to 33%, [14, 21,22,23,24] with a risk of recurrence between 11 to 21%, higher in cases with extended-spectrum beta-lactamase (ESBL) production [22, 24,25,26]. Most of the studies report K. aerogenes outcomes aggregated in Enterobacter genus. Though, K. aerogenes BSI mortality appears to be lower than in Enterobacter spp. infections, between 10 and 21% [21, 27].

Since 1971, taxonomic studies have suggested that Enterobacter aerogenes should be classified into Klebsiella genus, due to its phenotypic and genotypic similarities to K. pneumoniae [28,29,30,31]. Recently, Tindall et al. has proposed to officially change its name to K. aerogenes and Clinical and Laboratory Standards Institute (CLSI) and European Committee on Antimicrobial Susceptibility Testing (EUCAST) have implemented it in their official documents [32, 33]. Genomic data also suggest that K. aerogenes could be more virulent than E. cloacae, since it possess virulence-encoding genes that were also identified in K. pneumoniae, but not described in E. cloacae [34,35,36]. If these similarities between K. aerogenes and K. pneumoniae translate into similar clinical outcomes is still to be determined. Some studies had tried to compare outcomes in K. aerogenes and E. cloacae BSI, but their findings have been contradictory and inconclusive [37,38,39,40]. We didn’t find any study comparing clinical outcomes in K. aerogenes and K. pneumoniae BSI.

This study aims to assess the clinical outcomes in patients with BSI caused by K. aerogenes, K. pneumoniae and E. cloacae, by exploring the association between a clinical composite outcome (mortality or recurrence) and these pathogens.

Methods

Study design

A secondary data analysis, nested in the PRO-BAC observational cohort study was performed. PRO-BAC was a Spanish multicentre study that prospectively included episodes of microbiological confirmed bacteraemia in hospitalised patients between 1st October 2016 and 31st March 2017 (Clinicaltrials.gov NCT03148769; Register date 05/04/2017) [41]. Screening was done by daily review of blood culture results; individuals without systemic signs or symptoms of infection and subsequent episodes caused by the same microorganism within three months of the primary episode were also excluded [41]. Microorganism identification and susceptibility testing was performed according to the local laboratory guidelines that were aligned with the 2016 EUCAST recommendations or the CLSI guidelines. Clinical outcomes such as fever ≥ 72 h, complications secondary to treatment and metastatic infections were recorded at 30-days follow-up after bacteraemia diagnosis or hospital discharge. Mortality and BSI recurrence were assessed at 30-days, with post-discharge follow-up done by reviewing medical records and national death registry [41].

Inclusion criteria for this study was adult individuals (≥ 18 years old) with BSI caused by K. aerogenes, K. pneumoniae or E. cloacae, each a different exposure group. Polymicrobial bacteraemia episodes were excluded. Primary outcome was a composite clinical outcome that includes all-cause mortality or recurrence within 30 days after BSI diagnosis. BSI recurrence was defined as bacteraemia with the same microorganism after negative blood cultures or clinical improvement and completion of active antimicrobial therapy [42]. Secondary outcomes were persistent fever, persistent BSI, and device infection secondary to BSI. Persistent fever was defined as fever ≥ 72 h in patients treated with an in vitro active antimicrobial drug [42]. Persistent BSI was defined as positive blood culture with the same microorganism after 72 h of active antibiotic therapy [42]. Secondary infection included endovascular or orthopaedic devices infection secondary to BSI [42].

Setting of BSI acquisition was considered nosocomial if BSI occurred more than 48 h after hospital admission. Episodes were considered HAI if any of the following was present: intravenous therapy, wound care or specialized nursing care at home, dialysis, radiotherapy or chemotherapy, two or more attendances to specialized outpatient clinic, residency in a nursing home or long term care facility in the 30 days before BSI; major surgery 30 days before BSI (90 days if implant); or hospitalization for more than 2 days in acute or chronic care hospital in the 90 days before BSI. Episodes that don’t fulfil these criteria were considered community acquired.

Date of blood sample collection was considered the start of follow-up. In individuals with a primary outcome event, end of follow-up was considered when one of the events from the composite outcome was reached (death or recurrence), whatever happens first. In individuals without primary outcome, the date of last assessment was considered as the end of follow-up. In case of missing data in this variable, the date of hospital discharge was considered. Follow-up was right censored at 30 days after blood culture collection, according to the original study protocol. Loss to follow-up was considered when there was neither the date of last assessment nor date of hospital discharge.

Statistical analysis

Frequencies and proportions were used to describe the cohort characteristics and reported through tabulation of categorical variables by the exposure variable (K. aerogenes, E. cloacae and K. pneumoniae), including missing data. Median and interquartile range were calculated to describe non-normally distributed data and stratified by the microorganism group. All records in the final sample size for data analysis were included in the univariate and multivariate analysis, regardless of missing data in covariates.

To assess the impact of potential bias caused by excluding individuals with missing data on follow-up time, patients included in the final analysis were compared with the patients excluded. Differences in these groups were inspected and assessed using Pearson chi-squared test for categorical variables and Wilcoxon rank-sum test for continuous skewed variables (Additional file 1: Table S1).

Univariate analysis used two models: model 1 compares outcomes in K. aerogenes and E. cloacae groups with a baseline group of K. pneumoniae; model 2 compares outcomes in K. aerogenes and K. pneumoniae groups with a baseline group of E. cloacae. Rate of the outcome per 1000 persons-day and the 95% confidence interval for the exposure variable and other independent variables was calculated. Rate ratio and the 95% confidence interval of the association between the composite clinical outcome and the different microorganism’s groups and other independent variables was calculated using a multilevel mixed-effect Poisson regression model considering hospital as the first level, assuming clustering of observations within hospital. Wald test was used to test for association.

A direct acyclic diagram (DAG) was built to describe the conceptual framework between potential confounders and other independent variables in the association between exposure (K. aerogenes, K. pneumoniae or E. cloacae BSI) and outcome (death or recurrence), based on literature review (Additional file 1: Fig. S1). Possible confounders identified in the conceptual framework were BSI source, hospital ward, comorbidities, age, and antibiotic therapy. DAG was useful to exclude other independent variables from the multivariate model that can be in the causal pathway (severity) or ancestors of the exposure and/or outcome (BSI acquisition, antibiotic resistance, devices, and procedures). Disease severity is a mediator of the association between the exposure (microorganisms) and the outcome (death or recurrence), which may bias the results if included in the analysis. Ancestors are variables that are indirectly associated with the exposure or outcome through confounders; their inclusion in the model is redundant and can compromise model stability when missing data is an issue.

Multilevel mixed-effect Poisson regression model was used to calculate the crude and adjusted rate ratio for each possible confounder. Multicollinearity was checked by using variance inflation factor and variables with cut-off higher than 10 were excluded. The final multivariate model included all the variables without multicollinearity defined a priori as confounders (age and sex) and identified through the conceptual framework (BSI source, hospital ward, Charlson score and active antibiotic therapy).

Results

Descriptive analysis

From the 664 records in the PRO-BAC dataset with BSI caused by the microorganisms of interest, 132 were excluded due to exclusion criteria (131 polymicrobial BSI and 1 age < 18-year-old) and 89 were excluded due to missing data on follow-up time. The final sample size for analysis included 443 BSI episodes: 29 K. aerogenes, 77 E. cloacae and 337 K. pneumoniae (Additional file 1: Fig. S2).

The median follow-up time was 30 days (IQR 11, 71) and similar between exposure groups (K. pneumoniae 31 days vs E. cloacae 26 days vs. K. aerogenes 30 days) (Table 1).

Table 1 Sample characteristics: descriptive analysis according to microorganism group
Full size table

Comparing between groups, patients with E. cloacae BSI had more frequent nosocomial infections (K. pneumoniae 46.0%, n = 155 vs. E. cloacae 71.4%, n = 55 vs. K. aerogenes 41.4%, n = 12), less common BSI secondary to urinary source (K. pneumoniae 41.8%, n = 141 vs. E. cloacae 18.2%, n = 14 vs. K. aerogenes 37.9%, n = 11) and the length of hospital stay was longer (K. pneumoniae 14 days vs. E. cloacae 21 days vs. K. aerogenes 11 days); on the other hand, in K. pneumoniae group, antibiotic resistance was more common (ESBL: K. pneumoniae 23.4%, n = 79 vs. E. cloacae 2.6%, n = 2 vs. K. aerogenes 0%, n = 0; carbapenemase production: K. pneumoniae 8.8%, n = 33 vs. E. cloacae 0%, n = 0 vs. K. aerogenes 3.4%, n = 1) but active empiric antibiotic therapy more frequent (K. pneumoniae 74.2%, n = 250 vs. E. cloacae 61.0%, n = 47 vs. K. aerogenes 65.5%, n = 19) (Table 1).

The primary composite outcome was less frequent in K. aerogenes group (K. pneumoniae 19.0%, n = 64 vs. E. cloacae 20.8%, n = 16 vs. K. aerogenes 6.9%, n = 2) reflecting the findings of all-cause mortality (K. pneumoniae 18.4%, n = 62 vs. E. cloacae 19.5%, n = 15 vs. K. aerogenes 6.9%, n = 2) and BSI recurrence (K. pneumoniae 4.5%, n = 15 vs. E. cloacae 10.4%, n = 8 vs. K. aerogenes 0%, n = 0). Persistent BSI was also less frequent in K. aerogenes group, while other secondary outcomes were more common (Table 1).

Variables with more than 10% of missing data were ESBL production (26.4%, n = 117), carbapenemase production (33.2%, n = 147), active empiric antibiotic therapy (12.9%, n = 57) and time to start antibiotic (11.1%, n = 49) (Additional file 1: Table S1). Missing data were similar between groups, except in active empiric antibiotic therapy, in which the proportion of missing data was higher in the E. cloacae group (Table 1).

Survival analysis

In total, 436 episodes were included in the survival analysis, corresponding to 75 events for a total time at risk of 9360 days (8 events per 1000 person-days). Seven BSI episodes (six K. pneumoniae and one E. cloacae) were excluded due to death at day 0.

The Kaplan–Meier curve shows a similar survival in E. cloacae and K. pneumoniae groups until day 17 (around 87.5% survival). After day 17 the survival for E. cloacae group is slightly lower than in K. pneumoniae group, reaching around 75% in both groups at day 30. The K. aerogenes survival curve is difficult to interpret because only two events occurred in the early follow-up period (Fig. 1). There was no evidence of difference between groups in the log-rank test (p-value 0.337).

Fig. 1
figure 1

Kaplan–Meier curve: cumulative survival by microorganism

Full size image

Univariate analysis

Patients with K. aerogenes BSI were 58% and 65% less likely to die or have recurrence, when compared with K. pneumoniae (rate ratio (RR) 0.42; 95% CI 0.10 to 1.71) and E. cloacae (RR 0.35; 95% CI 0.08 to 1.55), respectively, without statistical evidence for an association between microorganisms and outcome (p-value 0.379) (Table 2). Individuals with E. cloacae BSI were 18% more likely of having the outcome of interest (RR 1.18; 95% CI 0.67 to 2.10), when compared to K. pneumoniae group, without statistical evidence of an association (p-value 0.379) (Table 2).

Table 2 Univariate analysis: association between the microorganism and other independent variables with composite outcome (death or recurrence)
Full size table

Men were 31% less likely to develop the composite outcome, when compared to women, without evidence of an association (RR 0.69; 95% CI 0.43 to 1.09; p-value 0.110) (Table 2). Compared to patients aged 18 to 59 years, individuals aged 70 to 79 years and over 80 years were 2.21 times (RR 2.21; 95% CI 1.12 to 4.36) and 2.14 times (RR 2.14; 95% CI 1.02 to 4.46) more likely to die or have recurrence, respectively, with evidence for this association (p-value 0.039) (Table 2).

There was evidence that participants with a Charlson comorbidity index ≥ 5 were 3.56 times more likely of dying or having recurrence, when compared to index < 5 (RR 3.56; 95% CI 2.11 to 6.01; p-value 0.001) (Table 2). Compared with individuals admitted to medical wards, patients in surgical departments were 47% less likely to die or recurrence (RR 0.53; 95% 0.27 to 1.06), without statistical evidence of an association (p-value 0.198) (Table 2). Individuals with catheter associated BSI were 49% less likely to have the composite outcome (RR 0.51; 95% CI 0.19 to 1.38), while respiratory tract were 35% more likely (RR 1.35; 95% CI 0.56 to 3.24), when compared to abdominal source, without statistical evidence of an association (p-value 0.598) (Table 2). There was no association between active empiric antibiotic and the outcome under study (RR 0.94; 95% CI 0.49 to 1.81, p-value 0.863) (Table 2).

Multivariate analysis

In the adjusted analysis the association between microorganisms and clinical outcome was slightly decreased towards the null, when compared to the crude analysis. Individuals with K. aerogenes were 56% and 52% less likely to die or having recurrence, when compared to K. pneumoniae (RR 0.44; 95% CI 0.11 to 1.84; p-value 0.262) and E. cloacae (RR 0.48; 95% CI 0.10 to 2.30; p-value 0.361), respectively, without statistical evidence of an association (Table 3).

Table 3 Multivariate analysis adjusted for age, sex, Charlson score, department, BSI source, and active empiric antibiotic
Full size table

Men were 24% less likely to develop the composite outcome, when compared to women (RR 0.76; 95% CI 0.44 to 1.31; p-value 0.330) (Table 3). Compared to the age group 18 to 59 years, individuals between 60 to 69 years were 27% less likely to die or have recurrence (RR 0.73; 95% CI 0.30 to 1.78; p-value 0.483), while individuals between 70 to 79 years old were 32% more likely (RR 1.32; 95% CI 0.59 to 2.93; p-value 0.499) (Table 3).

Charlson comorbidity index equal or above 5 was associated with a 3.87-fold increase in the risk of death or recurrence, when compared to patients with a lower index (RR 3.87; 95% CI 1.94 to 7.69; p < 0.001) (Table 3). Patients admitted to surgical had 34% less risk of worse outcome, when compared to individuals in the medical ward (RR 0.66; 95% CI 0.29 to 1.52; p-value 0.331) (Table 3). The risk of death or recurrence was 39% higher in BSI secondary to respiratory tract infection, when compared to abdominal source (RR 1.39; 95% CI 0.50 to 3.82; p-value 0.527) (Table 3). Patients treated with active empiric therapy seems to have a lower risk of death or recurrence, without statistical evidence of an association (RR 0.84; 95% CI 0.41 to 1.72; p-value 0.634) (Table 3).

Discussion

In this cohort study, individuals with K. aerogenes BSI were 58% and 65% less likely to die or have recurrence, when compared to K. pneumoniae and E. cloacae, respectively, although there was no evidence for an association. After adjusting for possible confounders, the rate ratio decreased towards the null effect. Despite the structural and genetic similarities between K. aerogenes and K. pneumoniae, this study demonstrates that K. aerogenes infection might have better outcomes.

In the survival analysis E. cloacae and K. pneumoniae groups have similar survival proportion. In fact, in the univariate analysis, the difference between these pathogens was very small; patients with E. cloacae BSI were 18% more likely of dying or having recurrence, when compared to K. pneumoniae group, without statistical evidence of an association. In our study, patients with E. cloacae BSI had more frequent nosocomial infections and catheter-associated BSI, when compared to other groups, which might be indicators of worse prognosis in these patients.

Previous studies have found a trend in the opposite direction, with higher mortality in K. aerogenes when compared to E. cloacae BSI, without statistical evidence of difference between groups. [37, 40] A multicentre case–control study conducted in five Spanish hospitals have assessed all-cause mortality at 30-day follow-up in cases with K. aerogenes or E. cloacae BSI with a control group of individuals without BSI matched by age, sex, and hospital area. [37] The authors reported no difference in 30-day mortality between these two pathogens and E. cloacae and K. aerogenes groups were similar in terms of place of acquisition, BSI source, severity, and antibiotic resistance. [37] The higher mortality in this study can be due to higher proportion of nosocomial K. aerogenes infections. [37] A single centre cohort study in the USA compared in-hospital all-cause mortality in patients with K. aerogenes and E. cloacae BSI. [40] When assessing a composite outcome of death before discharge, recurrence and/or complications, K. aerogenes group had a worse clinical outcome, when compared to E. cloacae [40]. The worst clinical outcome in the K. aerogenes group can be due to higher proportion of acute kidney injury in these patients, an outcome that was not included in the PRO-BAC analysis [40]. Proportion of recurrence was very low in both groups, when compared to PRO-BAC results [40]. The single centre design and the long period of enrolment can justify this difference, for example through spread of more virulent clones in the USA centre, differences in healthcare services through time or higher severity of patients admitted to this institution. This study also has the limitations of not considering the differences in length of hospital stay and it is not clear if these events have occurred before or after 30-days follow-up. An older cohort study from 2010, compared mortality in individuals admitted to a single centre in Seoul with K. aerogenes and E. cloacae BSI [39]. In this Korean study, there was statistical evidence of a higher mortality in the K. aerogenes group at 7-, 14- and 21-days follow-up [39]. Even though the proportion of septic shock is similar in both studies, the differences in results can be due to a higher proportion of ESBL production and nosocomial infections in the K. aerogenes group in the Korean study, when compared to PRO-BAC dataset [39]. The single centre design is a limitation of this study that could have influenced the higher proportion of ESBL-producing bacteria and nosocomial infections, more common in centres with inefficient infection prevention and control measures [43, 44]. Additionally, the analysis was not adjusted for confounding and K. aerogenes group was more frequently treated with inappropriate therapy, which could have biased the results [39].

On the other hand, findings from a more recent Korean case–control study are similar to the PRO-BAC analysis; there was statistical evidence of a higher mortality in E. cloacae group, when compared to K. aerogenes in the propensity score matched analysis [38]. Compared to PRO-BAC cohort, patients in Korean study have more frequent healthcare-associated infections and resistance to 3rd generation cephalosporins, while a higher proportion was treated with appropriate empiric therapy [38]. Even though the single centre design could be a limitation to results generalizability, the propensity score matched analysis reduces bias due to confounding. We didn’t find any study comparing clinical outcomes in K. aerogenes with K. pneumoniae BSI.

Our study uses a composite outcome that includes mortality or BSI recurrence at 30-day follow-up which may be consider an advantage, allowing to reflect a wider spectrum of clinical outcomes and increase the statistical study power, by allowing a higher number of outcome events per exposure group [45, 46]. The use of conceptual framework to decide which potential confounders should be included in the adjusted analysis improved the reliability of results, by avoiding over adjusting, increases the transparency of the analysis and supports the interpretation of the study results [47, 48].

The main limitation of our study is the small sample size and the low number of events in the three pathogens groups (particularly in the K. aerogenes group). The outcome assessment was done by review of medical records and national death registry, so we don’t expect that misclassification of the outcome have been an important cause of bias. Additionally, seven observations were excluded from the analysis, due to death at day 0, which could represent severe ill patients infected by a more virulent pathogen. However, most of them were infected by K. pneumoniae, which might have biased the results towards the null. Another relevant weakness was the exclusion of a high number of BSI episodes that didn’t have the date of entry or exit in the study. When comparing patients included and excluded from the analysis, the groups are very similar, only different in terms of antibiotic resistance and comorbidities (Additional file 1: Table S1). The impact of these differences is difficult to predict; patients excluded could be less severe due to lower proportion of comorbidities or more severe due to higher presence of carbapenemase.

Reporting the results using rate per 1000 person-days provides an adjusted analysis for the time at risk. The fact that the analysis didn’t take into consideration the possibility of competing events probably hasn’t affected the results because the overall mortality in PRO-BAC dataset is lower than in previous studies.

Due to small sample size this study results might not be generalizable. Nevertheless, the high proportion of death and recurrence in general in K. aerogenes, E. cloacae and K. pneumoniae BSI, should raising awareness about the burden of these pathogens and the importance of appropriate therapy. The study findings stress the need to develop and implement rapid diagnostic tools in BSI, which could help in early prescription of appropriate target therapy. Other strategies, for example, antimicrobial stewardship programs, can also be important to support clinicians when deciding empirical therapy, taking into consideration local epidemiological data.

Conclusions

This study suggests that BSI caused by K. aerogenes may have a better prognosis than E. cloacae or K. pneumonia; and that E. cloacae BSI may have a similar prognosis as K. pneumoniae.

New studies with bigger sample size and focused on other populations (e.g., paediatrics) will be useful to clarify the contradictory findings described in literature. It would be interesting to have more studies using composite outcomes, since they represent a wider variety of possible outcomes, reflecting the natural history of this disease.