Introduction

Febrile neutropenia (FN) is the severe adverse events that can occur during chemotherapy for cancer [1]. In allogeneic hematopoietic stem cell transplantation (allo-HSCT) recipients especially, bloodstream infection (BSI) during FN remains a severe and life-threatening problem [2,3,4]. According to current guidelines, fluoroquinolone (FQ) prophylaxis, including levofloxacin (LVFX) prophylaxis, should be considered for high-risk patients, including allo-HSCT recipients, as the preventive strategy against bacterial infection [5, 6].

However, in this era where multidrug-resistant gram-negative bacteria are increasing worldwide, there are also concerns about the association between inducing or selecting multidrug-resistant strains, including extended-spectrum β-lactamase (ESBL)-producing pathogens and exposure to FQs [7, 8]. Some studies involving patients with hematological malignancy reported higher mortality rates associated with ESBL-producing Enterobacteriaceae (ESBL-E) bacteremia compared with bacteremia caused by non-ESBL-E [9]. A recent meta-analysis suggested that cefepime (CFPM) as empiric therapy for patients with FN was associated with a higher risk of all-cause death and this might be caused by inappropriate therapy for ESBL-E [10]. Reporting on the epidemiology and treatment strategy for gram-negative bacteremia (GNB), including ESBL-producing pathogens, in the setting of allo-HSCT might help to prevent poor outcome. To this end, in this study, we investigated the epidemiology of GNB and the treatment strategy used at the first episode of FN in allo-HSCT recipients on LVFX prophylaxis.

Methods

Study design

This retrospective analysis of BSI focused on GNB among allo-HSCT recipients (aged ≥ 20 years) who received standard prophylaxis with 500 mg/day oral LVFX at the 890-bed Toranomon Hospital in Tokyo between January 2011 and December 2016. Medical and microbiological records were reviewed. This study was approved by the Human Ethics Review Committee of Toranomon Hospital.

Study population

We included all recipients of allo-HSCT who fulfilled two inclusion criteria: LVFX prophylaxis 500 mg/day oral started before day − 7 (7 days before allo-HSCT); and LVFX prophylaxis stopped at the first episode of FN and changed to another antimicrobial regimen immediately after obtaining ≥ 2 sets of blood cultures.

A central venous catheter was inserted before starting the conditioning regimen. Trimethoprim-sulfamethoxazole was given from day − 7 through day − 2 for Pneumocystis pneumonia prophylaxis.

Monotherapy with an anti-pseudomonal β-lactam (APBL) agent, as recommended by the Infectious Disease Society of America guidelines, is the routine empiric antimicrobial regimen initially administered at the onset of FN [6]. Our clinicians could choose any APBL, such as CFPM, piperacillin/tazobactam (PIPC/TAZ), and meropenem (MEPM). Amikacin (AMK) could be combined with an APBL regimen for suspected severe sepsis caused by β-lactam-resistant gram-negative rod. Anti-gram-positive agents were combined with the APBL regimen for specific clinical indications, including suspected catheter-related infection, skin and soft tissue infection, or severe sepsis caused by β-lactam-resistant gram-positive organisms. Most of the clinicians chose vancomycin as the first-line anti-gram-positive agent.

Definitions

The definition of FN was body temperature ≥ 37.5 °C measured at the axillary fossa, as routinely taken in Japan [11], and absolute neutrophil count (ANC) < 500 cells/μL or ANC that decreased to < 500 cells/μL with 48 h following fever onset [6].

All hematological disorders were defined as either standard risk or high risk [12]. Conditioning regimens were classified based on a report by the Center for International Blood and Marrow Transplant Research [13]. Recipients with prior history of allo-HSCT are those who have received allo-HSCT two or more times.

BSI was defined as isolation of a bacterial or fungal pathogen from at least 1 blood culture. However, coagulase-negative staphylococci, Corynebacterium species, and unidentified gram-positive cocci were considered contaminants unless cultured from ≥ 2 separate blood culture bottles. Co-infection was defined as identification of ≥ 2 bacterial species in multiple blood culture bottles of samples collected within 24 h.

Appropriate empiric therapy was defined as treatment with an antimicrobial agent to which the causative organism was susceptible within 24 h at the first episode of FN. Severity at BSI onset was evaluated by the Pitt bacteremia score (PBS) [14]. Thirty-day mortality was defined as death within 30 days after developing FN.

Identification and antimicrobial susceptibility

Organisms cultured from blood samples were identified using the VITEK2 system (bioMérieux, Marcy l’Etoile, France) or the WalkAway 96 SI system (Siemens Healthcare, Deerfield, IL). The Clinical and Laboratory Standards Institute (CLSI) criteria were used to define susceptibility or resistance to the antimicrobial agents studied. ESBL production was detected using the Mast® D68C test (Kanto Kagaku, Tokyo, Japan) and/or the disk-diffusion method according to the CLSI recommendations [15, 16]. The breakpoint of APBL was determined according to CLSI M100-S29 [16].

Statistical analysis

Among GNB, GPB, and non-bacteremia groups, categorical variables were compared using Fisher’s exact test. Additionally, post hoc testing was performed to investigate significant group differences in variables with P values of < 0.05. Thirty-day mortality rates were estimated using Kaplan-Meier analysis and the groups were compared using the log-rank test. All statistical analysis was performed with EZR (Saitama Medical Center, Jichi Medical University, Japan), a graphical user interface for R (The R Foundation for Statistical Computing) [17].

Results

Patients characteristics

FN occurred in 414 of the 417 allo-HSCT patients who received prophylactic LVFX during the study period. Prophylactic LVFX was changed to any one of the APBLs in all 414 cases immediately after obtaining ≥ 2 sets of blood cultures at the onset of the first FN episode. Four of the 414 cases were excluded from this study due to GNB and GPB mixed infection (3/4 cases) and fungal bacteremia (1/4 case). The remaining 410 allo-HSCT cases comprised 253 (61.7%) men and 157 (38.3%) women, with a median age of 55 years (range, 20–75 years). The most common underlying disease was acute myeloid leukemia (n = 191; 46.6%). Overall, 299 (72.9%) received cord blood transplantation (CBT), 69 (16.8%) received bone marrow transplantation, and 43 (10.2%) received peripheral blood stem cell transplantation.

Incidence of BSI and isolates

Bacteremia in the first FN episode occurred in 169 (41.2%) of the 410 cases; polymicrobial bacteremia was evident in 26 of these cases. Table 1 lists the patients’ clinical characteristics. In total, 29 (17.2%) of 169 cases had GNB and 140 (82.8%) had gram-positive bacteremia (GPB). The median time to onset of BSI was 4 days (range − 7 to 12 days) after HSCT. A total of 194 causative organisms were identified from blood samples in the 169 cases; 165 (85.1%) were gram-positive bacteria and 29 (14.9%) were gram-negative bacteria.

Table 1 Patients characteristics among three groups including GNB, GPB, and non-bacteremia
Full size table

Among the GNB, GPB, and non-bacteremia groups, there were significant differences in CBT, HLA4/6, severe neutropenia, and neutropenia of > 7 days. Post hoc testing identified CBT, HLA4/6, and neutropenia of > 7 days as risk factors for GPB only and severe neutropenia as a risk factor for both GPB and GNB.

Of the 29 GNBs documented, causative organisms were E. coli in 21 cases including 10 ESBLs (12.4% of total, 72.4% of gram-negative), Klebsiella pneumoniae in 2 cases (non-ESBLs), Pseudomonas aeruginosa in 2 cases, Helicobacter cinaedi in 2 cases, and other in 2 cases (Table 2). Among the GPBs, the common causative organisms were viridans group streptococci (38.5% of the total, and 46.4% of gram-positive) and coagulase-negative streptococci (21.3% of total, and 25.7% of gram-positive).

Table 2 Gram-negative bacteria from bloodstream isolates in HSCT recipients during FN on LVFX prophylaxis
Full size table

All isolates that had established clinical breakpoints were resistant to LVFX. No carbapenem-resistant Enterobacteriaceae (CRE) was documented.

Therapeutic regimens and outcomes

Among the 410 FN patients, the CFPM-based regimen (89%) was most frequently used for empiric therapy, followed by PIPC/TAZ-(8.8%) and MEPM-based (2%) regimens. Among the 29 GNB cases, treatment was started with CFPM monotherapy in 25 cases (86.2%), with CFPM and AMK in 2 cases (6.9%), with PIPC/TAZ monotherapy in 1 case (3.4%), and with PIPC/TAZ and AMK in 1 case (3.4%). For patients with GNB, the 30-day crude mortality rate was 6.9% (2/29), and this rate did not differ significantly among those with GNB (6.9%), those with GPB (7.1%), or those without bacteremia (5.4%; P = 0.78; Fig. 1).

Fig. 1
figure 1

Crude 30-day mortality rate among the GPB, GNB, and non-bacteremia groups

Full size image

Characteristics of the appropriate and inappropriate empiric therapy groups for GNB

The results of in vitro susceptibility testing indicated that empiric therapy was inappropriate in 10 of 25 GNB cases (40%; excluding 4 pathogens that do not have CLSI breakpoints [2 H. cinaedi, 1 Campylobacter spp. and 1 Fusobacterium spp.]). Table 3 shows the characteristics of patients with inappropriate empiric therapy. Inappropriate empiric therapy consisted of CFPM in all cases and all causative organisms were ESBL-producing E. coli (ESBL-EC). Appropriate therapy (AMK, PIPC/TAZ, or MEPM) was initiated in all cases as soon as the response to CFPM was confirmed or gram-negative bacteria were detected from blood culture (median 1.0 day [0–4 days] after infection onset); eventually all cases were treated with MEPM. All the 10 inappropriate empiric therapy cases showed low PBS (≤ 2), of these 1 (10%) patient died due to intracranial hemorrhage 21 days after onset of bacteremia. The crude 30-day mortality rate (1/10, 10%) was not significantly different from that of the appropriate empiric therapy group (1/15, 6.7%; P = 0.61; Fig. 2).

Table 3 Patient characteristics with ESBL producing E. coli bacteremia
Full size table
Fig. 2
figure 2

Crude 30-day mortality rate among those with inappropriate empiric therapy (ESBL-EC) and those with appropriate empiric therapy (non-ESBL-EC)

Full size image

Discussion

Three major findings can be highlighted in this study. First, we clarified the characteristics of GNB during the first episode of FN among allo-HSCT recipients on LVFX prophylaxis. Second, GNB did not cause significant mortality. Third, the use of CFPM as empiric therapy, even if the causative organism was ESBL-EC, did not cause death in less severe cases (PBS ≤ 2).

Some previous studies have discussed the epidemiology and efficacy of FQ prophylaxis in allo-HSCT [18,19,20,21]. However, to our knowledge, this is the first study focusing on GNB in the first episode of FN under LVFX prophylaxis.

The 30-day mortality rate in cases of GNB did not significantly differ with that in cases of GPB or non-bacteremia. On the other hand, GNB is a major cause of illness and death in the previous study [22]. Girmenia et al. reported that pre-engraftment GNB after allo-HSCT represented an independent prognostic factor and was the cause of death in 39.1% of allo-HSCT recipients. The poor prognostic impact of GNB was related mainly to infection with carbapenem non-susceptible enterobacteria and P. aeruginosa [19]. There were no carbapenem non-susceptible enterobacteria and few P. aeruginosa in our study. This might be one of the reasons for the lack of significant differences among the GNB, GPB, and non-bacteremia groups. We analyzed only the first episode of FN under LVFX prophylaxis during the pre-engraftment phase and this may have affected the lack of CRE extracted as a result of exposure to various antibiotics.

Among the 25 GNB (excluding the 4 pathogens that had no CLSI breakpoints), there were 10 (40%) cases of inappropriate empiric therapy and all involved bacteremia due to ESBL-EC in which we started empiric antibiotic therapy with CFPM. Nevertheless, 30-day mortality in this group was not significant different from that in the appropriate empiric therapy group (1/10 vs. 1/15, respectively, P = 0.61). Furthermore, in only 1 recipient who died was death considered to be remotely related to ESBL-EC bacteremia (cause of death was intracranial hemorrhage). This is probably because all cases that received inappropriate empiric therapy for ESBL-EC bacteremia were of mild severity. In fact, all cases showed low PBS (≤ 2). Therefore, in empiric therapy for ESBL-EC bacteremia, inappropriate therapy for mild cases might have no effect on mortality as long as the treatment is switched to an effective regimen for ESBL-EC, such as carbapenem, immediately after documentation of gram-negative bacteria from the blood sample or soon after signs of deterioration such as hypotension.

Regarding the use of CFPM in patients with febrile neutropenia, it has been reported that patients treated with CFPM have no difference in mortality compared to patients treated with other antibiotics [23]. On the other hand, there is a report that the mortality rate was significantly higher in patients who received CFPM [24], and a most recent meta-analysis also showed that CFPM use as empiric therapy for FN might increase morbidity [10]. This was considered to be likely due to the effect of ESBL-EC bacteremia. Indeed, another study reported that patients with oncologic cancer, including those with hematological malignancy and those who undergone HSCT, had higher mortality when ESBL-EC bacteremia was confirmed during neutropenia compared with non-ESBL-EC bacteremia [25]. Patients with ESBL-EC bacteremia are more likely to receive inappropriate therapy compared with patients with non-ESBL-EC bacteremia, and in the state of neutropenia, the mortality rate is higher with ESBL-EC bacteremia than it is with non-ESBL-EC bacteremia [25]. Therefore, given the possibility of ESBL-EC bacteremia, carbapenems such as MEPM are recommended as empiric therapy for FN. Nevertheless, overuse of empiric carbapenem should be avoided because it might be associated with an increase in CRE [26].

For options other than carbapenems, β-lactam/β-lactamase inhibitors such as PIPC/TAZ can be considered. In an open-labeled randomized controlled trial, definitive therapy with PIPC/TAZ for ceftriaxone non-susceptible E. coli or K. pneumoniae was inferior to MEPM, although the study population did not include FN patients [27]. Thus, the empiric use of PIPC/TAZ might not help to avoid overusing carbapenem for treating FN.

Administering empiric carbapenem to only patients at high risk of developing ESBL-producing bacteremia among HSCT recipients could be one solution. It was reported that among HSCT recipients receiving FQ prophylaxis, stool colonization of ESBL-E prior to transplantation was associated with ESBL-E bacteremia developing, and ESBL-E colonization in stool has a 32.2% positive predictive value (PPV) and 99.6% of negative predictive value (NPV) for with ESBL-E bacteremia [28]. While the NPV is high, the PPV is relatively low, so unnecessary empiric therapy with carbapenems will be instituted for about 70% of patients who will not develop ESBL-E bacteremia. Moreover, Arnan et al. found no significant association between ESBL-EC bacteremia and colonization in stool among hematological malignancy patients, including HSCT recipients [29]. Therefore, the strategy of administering carbapenem based on stool screening may not be sufficient to ensure proper empiric carbapenem usage for FN.

Our findings indicate that empiric CFPM for less severe ESBL-EC bacteremia may not be associated with mortality. In this era of increasing ESBL-E worldwide, severity-directed empiric therapy or severity combined with ESBL-E colonization might offer a better therapeutic strategy while avoiding carbapenem overuse and preventing poor prognosis. In addition, anaerobic coverage was recently reported to be associated with risk and mortality of acute graft-versus-host disease [30]. Therefore, cefepime might be a more suitable empiric therapy of FN compared with carbapenems and PIPC/TAZ in allo-HSCT settings.

Our study has several limitations. First, it is a retrospective, single-center study. The prevalence of ESBL-E differs depending on the geographic areas, so practice in areas with high proportions of ESBL-E bacteremia remains controversial. Second, ESBL confirmatory testing was performed with the disk-diffusion method according to the Mast® D68C test and/or double-disk synergy test but did not involve PCR assay for β-lactamase genes. Third, this study has a small number of cases in the inappropriate therapy group. All the inappropriate therapy cases showed low PBS, so it was not possible to analyze the association between inappropriate therapy and worse prognosis in this group. However, in gram-negative BSI, high PBS is often associated with poor prognosis, suggesting that the inappropriate group was probably composed of milder cases [31]. Thus, the validity of the strategy where clinicians select empiric therapy on the basis of severity should be examined with a large number of patients. Fourth, although the limitations of FQ prophylaxis are recognized [4, 32, 33], it remains an accepted protocol for neutropenic patients with hematological malignancy and severe neutropenia [4, 20, 21]. Therefore, analyzing the impact of LVFX breakthrough infections is still important in real-world allo-HSCT settings currently.

In this study, we have reported on the epidemiology of GNB during the first FN episode among allo-HSCT recipients on LVFX prophylaxis. GNB was not a significant cause of death. In LVFX breakthrough ESBL-EC bacteremia among allo-HSCT recipients, the administration of CFPM as empiric therapy did not lead to significantly poor prognosis. This finding might contribute to establishing a strategy to avoid carbapenem overuse without worsening prognosis.