Background

Carbapenem-resistant gram-negative bacteria (CRGNB) infections have emerged as a global public health problem. These infections, including carbapenem-resistant Enterobacterales (CRE), carbapenem-resistant Acinetobacter baumannii, and carbapenem-resistant Pseudomonas aeruginosa, are associated with limited effective treatments and poor outcomes [1, 2]. Among these, bloodstream infections (BSIs) have the highest mortality rate [2].

End-stage renal disease (ESRD) in intensive care units (ICUs) is not uncommon [3]. Patients with ESRD undergoing renal replacement therapy (RRT) face an increased risk of BSI [4] due to impaired innate and adaptive immune responses [5] and the requirement for vascular access, leading to the breakdown of skin barriers and exposure to external and skin pathogens [6]. Moreover, patients with ESRD experience increased intravenous antibiotic exposure [7, 8], which can lead to the risk of inappropriate antibiotic use [7] and a higher prevalence of multidrug-resistant microorganisms (MDROs) [9]. More data are available regarding elevated infection rates involving MDROs in the ESRD population, including methicillin-resistant Staphylococcus aureus (MRSA), vancomycin-resistant Enterococci [10, 11], extended-spectrum beta-lactamase producing GNB [12], and CRE [13].

Previous research has explored the characteristics and outcomes of CRGNB-BSI [2, 14,15,16], but literature on the outcomes of CRGNB infections in patients with ESRD is scarce [17]. Eilertson et al. reported the outcomes of CRE infection in patients with ESRD, showing a higher likelihood of death and increased in-hospital mortality compared to those with normal renal function [13]. However, their study included only 21 patients with BSI, and the outcomes of critically ill patients were not evaluated. Another prospective study assessing the outcomes of 137 patients with CRGNB-BSI included only 47.4% patients with ESRD [18]. No study has explored the clinical outcomes in critically ill patients with ESRD diagnosed with CRGNB-BSI.

This multicenter retrospective study aimed to investigate the clinical and microbiological characteristics, antimicrobial strategies, and clinical outcomes of ESRD patients with CRGNB-BSI in the ICU.

Methods

Patients and study design

This multicenter retrospective observational study was conducted in the ICUs of five tertiary referral hospitals in Taiwan from January 2015 to December 2019. All consecutive patients with ESRD undergoing RRT at the time of hospital admission and admitted to the ICUs with blood culture yielding CRGNB were included. Patients younger than 20 years old, those with acute renal failure receiving RRT, those not receiving any antimicrobial therapy with curative intent, or those with insufficient data for analysis were excluded. This study conformed to the principles of the Declaration of Helsinki and adhered to the strengthening the reporting of observational studies in epidemiology (STROBE) guidelines. The institutional review boards (IRBs) of all participating hospitals approved the study protocol (IRB numbers: 2020–11-006AC, KMUHIRB-E(I)-20,180,141, CE18100A, A202005146_TSGHIRB, and CMUH110-REC1-139). The requirement for informed consent was waived by the IRB of each participating institution due to the retrospective nature of this study and the absence of personally identifiable information collection.

Definitions and data collection

BSI was defined as positive blood cultures in patients exhibiting systemic signs of infection [19]. The sources of BSI were identified based on clinical evaluations and microbiological findings. These infections were categorized into pneumonia, urinary tract infection (UTI), soft tissue infection, and intra-abdominal infection. For patients with a central vascular catheter and no clear source of BSI, the infection was considered catheter-related. Primary bacteremia was defined as the absence of an identifiable source of infection that cultured the same organism(s) as found in the blood and the absence of a central vascular catheter. Further details on these definitions can be found in the Supplementary materials. CRGNB were defined as A. baumannii, P. aeruginosa, or Enterobacterales resistant to any of the carbapenems, including meropenem, imipenem, or ertapenem, according to Clinical and Laboratory Standards Institute guidelines. ICU-acquired BSI was defined when the onset occurred more than 48 h following ICU admission [20]. Early onset BSI was defined as BSI developing within 7 days of hospital admission [21].

Data on patient characteristics, comorbidities, ICU type, sources of BSI, disease severity, microbiological profile, antimicrobial therapy, and outcomes were retrospectively collected from electronic medical records.

The acute physiologic assessment and chronic health evaluation (APACHE) II score on the day of ICU admission and the Sequential Organ Failure Assessment (SOFA) score on the BSI onset date were used to assess disease severity.

Intravenous antibiotics for CRGNB-BSI treatment, administered for at least 2 days within 7 days of the BSI onset date, were recorded. To mitigate time-window bias related to delayed intravenous antibiotics initiation, only antibiotics initiated within 3 days of the BSI onset date were included in the outcome analysis. Combination therapy was defined as the concurrent use of at least two intravenous antibiotics with an overlap of at least 2 days. The loading dose of intravenous colistin was as defined in associated studies [22]. Throughout the study period, novel β-lactam/β-lactamase inhibitors, including ceftazidime–avibactam and ceftolozane–tazobactam, were not available in Taiwan.

Outcome evaluation

The primary outcome was day-28 all-cause mortality following the onset of CRGNB-BSI. Survival analysis was conducted for both the overall study population and the cohort excluding patients who died on the BSI onset date (time-window bias-adjusted cohort). Other outcome measures included clinical failure and microbiological eradication at days 7, 14, and 28 following BSI onset, as well as in-hospital mortality.

Treatment outcomes were classified as follows: cure (negative blood culture for three consecutive days with no further requirement for antimicrobial therapy), improvement (partial resolution of signs and symptoms with the requirement for antimicrobial therapy or another course of antibiotic prescription for newly developed infections), or failure (persistence of signs and symptoms, antibiotic regimen escalation, or death). Microbiological response to treatment was categorized as either eradication (absence of the causative pathogen in blood specimens), failure (persistence of the causative pathogen in at least one blood specimen), or undetermined (culture of blood specimens not available during follow-up).

Statistical analysis

Statistical Package for the Social Sciences version 25 (SPSS Inc., Chicago, IL, USA) was used for all statistical analyses. For ordinal and continuous data, medians and interquartile ranges (IQRs) were reported, and the Mann–Whitney U test was used to assess group differences. Categorical variables were presented as numbers and percentages, and analysis was performed using the chi-square test. The Kaplan–Meier method and log-rank tests were employed for assessing day-28 mortality. Prognostic factors for day-28 mortality and treatment outcomes were analyzed using COX regression and logistic regression. In the multivariate model for day-28 mortality, significant variables (p < 0.05) from the univariate analysis and clinically important variables, such as age and antimicrobial regimens, were included. The strength of the association was presented as the hazards ratios (HR) and odds ratio (OR) along with their 95% confidence intervals (CI). Subgroup analyses stratified by disease severity, infection focus, and pathogens were performed to evaluate the efficacy of antimicrobial therapy. However, subgroups with insufficient sample sizes were not analyzed. For practical purposes, continuous variables with a maximal Youden’s index were dichotomized before subgroup analysis. All tests in this study were two-sided, and p < 0.05 was considered statistically significant.

Results

Demographic characteristics

Of the 149 critically ill patients with ESRD diagnosed with CRGNB-BSI (Fig. S1), 84 (56.4%) were male, with a median age of 69.9 years. Diabetes mellitus (44.3%), liver disease (19.5%), and malignancy (16.8%) were the most common comorbidities. More than 40% of the patients had antibiotic exposure in the 14 days preceding BSI onset. The median hospital and ICU length-of-stay (LOS) before BSI onset was 24 (IQR, 12.5–19.5) and 13.5 (IQR, 6–20) days, respectively. Approximately 70% of the patients were in the medical ICU, and 127 patients (85.2%) developed ICU-acquired CRGNB-BSI. The median APACHE II and SOFA scores were 23 (IQR, 19.3–29) and 11 (IQR, 8–15.5), respectively. Most patients (92.6%) received invasive mechanical ventilation, and 49 patients (32.9%) presented with shock (Table 1).

Table 1 Demographic data and laboratory findings of critically ill patients with end-stage renal disease and carbapenem-resistant gram-negative bacteria blood stream infection
Full size table

Microbiological profile and antimicrobial therapy

Catheter-related infections (47.7%) and pneumonia-related infections (32.2%) were the most common CRGNB-BSI etiologies in patients with ESRD (Fig. 1). The most common isolated pathogen was A. baumannii (49.0%), followed by Klebsiella pneumoniae (31.5%) and P. aeruginosa (8.7%) (Table 1).

Fig. 1
figure 1

Microbiological profile stratified by different infection foci. Two patients had primary bacteremia, and the attributed pathogens in both cases were Acinetobacter baumannii.

Full size image

Stratified by infection focus, the most common pathogen in catheter-related, pneumonia-related, and soft tissue infections was A. baumannii, whereas the most common pathogen in UTIs was K. pneumoniae (Fig. 1).

Overall, 71 patients (47.7%) received colistin, 23 of whom received a loading dose. Carbapenem, tigecycline, and aminoglycoside were administered to 71 (47.7%), 33 (22.1%), and 20 (13.4%) patients, respectively. Following adjustment for time-window bias in antimicrobial prescription, 40 patients (26.8%) received a combination of anti-CRGNB antibiotics, whereas 12 patients (8.1%) were treated with antibiotics with in vitro susceptibility (Table 1).

Primary outcome and prognostic factors for day-28 mortality

The day-28 mortality rate from CRGNB-BSI onset was 52.3%, with a 73.2% in-hospital mortality rate. Median ICU LOS since BSI onset was 11.5 days (IQR, 5–24.8 days), and hospital LOS was 42.5 days (IQR, 25–63.8 days) (Table 1).

The analysis of day-28 mortality in different CRGNB-BSI subgroups is presented in Fig. 2. Pneumonia-related CRGNB-BSI exhibited significantly higher day-28 mortality rates than non-pneumonia-related CRGNB-BSI. Catheter-related and early onset CRGNB-BSIs exhibited a favorable trend in survival outcomes, although without statistical significance. No significant difference in day-28 mortality was observed among various pathogens.

Fig. 2
figure 2

Kaplan–Meier curve of day-28 survival among different subgroups. AB Acinetobacter baumannii, BSI bloodstream infection, CR carbapenem resistant, CRE carbapenem-resistant Enterobacterales, KP Klebsiella pneumoniae, LOS length-of-stay, PA Pseudomonas aeruginosa

Full size image

Multivariate analysis revealed that day-28 mortality was significantly associated with a higher SOFA score (adjusted HR [aHR], 1.28; 95% CI 1.18–1.38) and shock status (aHR, 2.24; 95% CI 1.18–4.26). In the time-window bias-adjusted cohort, the prognostic effect of the SOFA score (aHR, 1.25; 95% CI 1.17–1.35) and shock status (aHR, 2.12; 95% CI 1.14–3.94) remained consistent (Table 2). Patients receiving colistin-based and aminoglycoside-based regimens showed a favorable trend toward day-28 survival.

Table 2 Univariate and multivariate analyses of clinical variables associated with day-28 mortality
Full size table

To illustrate the subgroups that received colistin versus non-colistin, a forest plot was generated. Among these subgroups, patients with a SOFA score of ≥ 13 or those experiencing shock demonstrated a lower risk of day-28 mortality with colistin-based therapy compared to those with non-colistin-based therapy (aHR, 0.34; 95% CI 0.16–0.72 and aHR, 0.28; 95% CI 0.09–0.85, respectively) (Fig. 3). Additionally, colistin-based therapy was associated with better day-28 outcome in patients with non-pneumonia-related BSI (aHR, 0.38; 95% CI 0.18–0.80) and those with A. baumannii infection (aHR, 0.40; 95% CI 0.16–0.87).

Fig. 3
figure 3

Forest plot of adjusted hazards ratios for day-28 mortality in different subgroups, comparing colistin-based treatment versus non-colistin-based treatment. Adjusted factors include age, APACHE II score, SOFA score, shock status. AB: Acinetobacter baumannii, aHR: adjusted hazards ratio, CI: confidence interval, SOFA: Sequential Organ Failure Assessment

Full size image

Secondary outcomes

The clinical and microbiological outcomes among different subgroups are illustrated in Table S1. Pneumonia-related CRGNB-BSI demonstrated the highest clinical failure rate (81.3%) and the lowest microbiological eradication rate (27.1%) at day 28. In contrast, UTI-related CRGNB-BSI showed a lower clinical failure rate (37.5%), and catheter-related CRGNB-BSI demonstrated the highest microbiological eradication rate (54.2%) at day 28. Moreover, comorbidities including liver disease and malignancy, along with higher disease severity, were associated with higher clinical failure rates and lower microbiologic eradication rates.

Logistic regression was employed to assess the clinical and microbiological outcomes associated with different antibiotics (Table 3). In the original cohort, colistin-based therapy was associated with a lower risk of clinical failure at day 28 (aOR, 0.20; 95% CI 0.05–0.83) and higher microbiological eradication at days 7, 14, and 28 (aOR, 3.92; 95% CI 1.03–14.91; aOR, 4.70; 95% CI 1.23–17.96; and aOR, 4.19; 95% CI 1.09–16.04, respectively). The benefit of colistin-based therapy remained observable in the time-window bias-adjusted cohort.

Table 3 Logistic regression for assessing the clinical and microbiological outcomes associated with different antibiotics
Full size table

On days 7, 14, and 28, aminoglycoside-based therapy demonstrated a reduced risk of clinical failure. Moreover, the observed benefit on day 28 persisted in the time-window bias-adjusted cohort. Conversely, carbapenem-based therapy was associated with lower microbiologic eradication on day 7 and an increasing trend in clinical failure and in-hospital mortality (Table 3).

Discussion

This study is the first multicenter investigation to explore the outcomes of CRGNB-BSI in ICU patients with ESRD and to identify prognostic factors for day-28 mortality. Critically ill patients with ESRD diagnosed with CRGNB-BSI demonstrated poor outcomes, with day-28 and in-hospital mortality rates of 52.3–73.2%, respectively. Prognostic factors for day-28 mortality included a higher SOFA score and shock status. Colistin-based therapy reduced mortality in the subgroups with higher disease severity and A. baumannii-related BSI.

It had been reported that patients undergoing RRT have a high incidence of CRKP infection, with dialysis identified as a factor associated with day-30 mortality (OR, 2.9; 95% CI 1.3–6.5) [23]. Khamis et al. reported a similarly high mortality rate associated with CRE-BSI in a cohort where ESRD patients accounted for 57% (96 out of 169) of the study participants [24]. Among critically ill patients in their study, 60.2% had ESRD on RRT, with an in-hospital mortality rate of 80% [24]. Our results align with these findings and highlight the increased risk of CRGNB-BSI–related mortality among ICU patients with ESRD.

In our study, independent predictors of day-28 mortality included a higher SOFA score and shock status in both the original and time-window bias-adjusted cohorts. Another retrospective cohort study evaluating CRGNB-BSI in patients with immunosuppression also reported that increased age and septic shock were independently associated with higher mortality [25]. Moreover, our results showed that ICU-acquired BSI demonstrated a trend towards statistical significance in predicting day-28 mortality (HR, 2.40; 95% CI 0.97–5.96), a risk that has been reported in a previous study [26]. Additionally, a recent study revealed a higher incidence of drug-resistant pathogens in ICU-acquired BSI [20], which may further worsen patient outcomes. Our results contribute to the current knowledge by highlighting the adverse outcomes of ICU-acquired CRGNB-BSI and underscoring the necessity for infection control measures in ESRD.

Although several novel antimicrobial agents have been developed, providing clinical benefits and being recommended in treatment guidelines [27, 28], these medications are not universally available. Due to this limitation, various antimicrobial regimens including older agents for CRGNB infection exist [27,28,29], which could be adopted on the basis of medical resources of different regions. In our study, colistin was the most prescribed antibiotic, consistent with previous literature indicating that colistin was more commonly administered in patients with CRGNB-BSI [13] and those with critical illness [30]. Furthermore, colistin was more often utilized in patients with ESRD on RRT compared to individuals with normal renal function [13], indicating heightened consideration for colistin-related renal toxicity by the treating physician and lesser concern for nephrotoxicity in the context of ESRD.

We found that Colistin-based therapy resulted in better day-28 survival in patients with shock and a SOFA score of 13 or more. Furthermore, colistin-based therapy was associated with higher microbiologic eradication rates at days 7, 14, and 28, and reduced the risk of clinical failure at day 28. These findings are supported by a prospective study that revealed an 83.6% bacteriological clearance rate with colistin-based therapy in patients with CRGNB-BSI [18]. Another study reported a 72.9% clinical cure rate in patients with chronic kidney disease and extensively drug-resistant P. aeruginosa infection who were treated with colistin [31]. However, these two studies had limited sample sizes of critically ill patients. Our results contribute to existing studies, supporting the observation of the clinical benefit of colistin-based therapy, especially in ICU patients with high disease severity [32]. Given the survival benefit in severe disease and less concern for nephrotoxicity in patients with ESRD, we suggest that colistin be empirically added to the treatment of patients with ESRD diagnosed with BSI and high disease severity if they are at risk of drug-resistant pathogens [15, 18, 33, 34].

We observed that a high proportion of patients did not receive combination treatment within three days of BSI onset date. This delay may be attributed to the inclusion of CRGNB-BSI in the current study, inappropriate empirical antimicrobial therapy for drug-resistant pathogens, and the time-consuming nature of conventional culture methods with antimicrobial susceptibility testing. A systematic review and meta-analysis reported that monotherapy (vs. combination therapy) for patients with CRE infections was associated with high overall mortality (OR, 2.19; 95% CI 1.00–4.80) [14]. Although empirical antimicrobial therapy is mainly determined by patients’ risks and local epidemiology [27, 35], emerging evidence shows that early detection of pathogens and antimicrobial resistance through molecular microbiology testing [36] and artificial intelligence/machine learning [37] aids in promptly guiding semitargeted antimicrobial therapy, thereby reducing the duration and incidence of inappropriate empirical antimicrobial therapy at BSI onset.

In this study, a trend of better day-28 survival and a lower risk of clinical failure on days 7, 14, and 28 with an aminoglycoside-based regimen was observed. Existing evidence suggests the use of aminoglycosides in the treatment of uncomplicated UTI, with insufficient evidence to support their use in other infections due to CRE [29]. The guidelines for treating infections caused by multidrug-resistant GNB, published by the European Society of Clinical Microbiology and Infectious Diseases, suggest that aminoglycosides can be employed in a combination regimen for multidrug-resistant GNB provided that they demonstrate in vitro activity [28]. Based on a surveillance of multicenter antimicrobial resistance in Taiwan [38], which included antimicrobial susceptibility data from 758 clinical isolates, amikacin showed high in vitro activity against both Enterobacterales and P. aeruginosa, with resistance rates below 8%. Due to the limited number of cases in this study, the treatment benefit of an aminoglycoside-based regimen in CRGNB-BSI cannot be concluded. However, given the unavailability of novel β-lactam/β-lactamase inhibitors in some regions and considering the use of an old antibiotic as part of antibiotic stewardship [28], the use of aminoglycoside combination should still be reserved as a treatment option, particularly in patients with ESRD and lower concern for renal toxicity.

This study had several limitations. First, the case number was relatively small to draw a definite conclusion about the treatment benefits of the investigated antibiotics, and large-scale studies are needed. Second, this study investigated the prognostic effect of older antibiotic agents and could not extend the conclusions to other novel antibiotic agents in CRGNB-BSI among critically ill patients with ESRD. Third, there were variations in antimicrobial regimens in this retrospective study, including differences in the timing of initiation and duration of treatment, influenced by treatment response and patient condition. Along with the small sample size, we analyzed the prognostic effect of different antibiotic-based regimens rather than the cumulative dose of a specific antibiotic or regimen. Fourth, most patients in this study were mechanically ventilated and may have received sedative agents. However, we did not record the use of sedative agents, which could influence the evaluation of the neurologic components of the SOFA score.

Conclusions

CRGNB-BSI led to high mortality in critically ill patients with ESRD. Day-28 mortality was independently predicted by a higher SOFA score and shock status. In patients with higher disease severity and A. baumannii-related BSI, colistin-based therapy improved treatment outcomes.