Summary of findings
Cancer patients are prone to infection. Low blood cell count (neutropenia) and disruption of normal barriers to infection, such as skin and mucous membranes, are caused by chemotherapy or underlying malignancy. Both disrupt the normal immune response and predispose patients to infection (Bodey 1966). Pathogens implicated in these infections are Gram-negative bacteria, including Pseudomonas aeruginosa, Gram-positive bacteria and fungi (Chow 1991; Hughes 1997). The considerable morbidity and mortality associated with these infections in neutropenic patients led to the routine use of empirical antibiotic treatment, which is given upon suspicion of infection (e.g. fever), before the causative pathogen/s or their susceptibilities are identified (Hughes 1997; Schimpff 1986).
Initial effective empirical treatment for patients with fever and neutropenia consisted of combinations of antibiotics, including double beta-lactam regimens and, more recently, aminoglycoside-beta-lactam combinations (Hughes 1990; Hughes 1997; Schimpff 1971). In the 1980s, third-generation cephalosporins and carbapenems having bactericidal activity against Enterobacteriaceae, Pseudomonas aeruginosa and many Gram-positive organisms became available, making monotherapy a reasonable alternative to combination therapy. Neither combination therapy nor monotherapy provides full coverage for the spectrum of infections encountered among neutropenic patients. Notably, resistant Gram-positive bacteria and fungi are left untreated. Nevertheless, current guidelines recommend beta-lactam monotherapy in clinically stable patients (Freifeld 2011; Tam 2011).
An evident advantage of combination therapy over monotherapy is the higher probability that the infecting pathogen will be covered by at least one of the components of the regimen. Furthermore, the interaction between two antibiotics may be synergistic, resulting in enhanced bacterial kill activity compared with the additive activities of the antibiotics when assessed separately (Giamarellou 1984; Giamarellou 1986; Klastersky 1976; Klastersky 1982). Finally, use of combination therapy has been claimed to suppress the emergence of resistant subpopulations of bacteria (Allan 1985; Milatovic 1987; Wade 1989). On the other hand, benefits of monotherapy may include a lower probability of adverse effects and narrower-spectrum treatment, possibly reducing the chance of developing a super-infection with resistant bacteria (Weistein 1985). Adverse effects may be related to administration of aminoglycosides per se (e.g. nephrotoxicity) or to interactions between antibiotic and underlying disease and/or other drugs. Neutropaenic participants not responding to the initial antibiotic regimen will be given modified treatment, which usually includes vancomycin to cover resistant Gram-positive bacteria and/or amphotericin-based preparations or azoles to treat fungal infection (Hughes 1997), thus increasing the chance for adverse events and drug interactions.
Although neutropenia itself is the single most important risk factor for infection, other factors can alter the risk. The probability and severity of infection are inversely proportional to the absolute neutrophil count, and patients with neutrophil counts below 100/mm³ are at highest risk for severe infection (Bodey 1966; Schimpff 1986). Underlying malignancy may affect outcome. Patients with acute leukaemia and other haematological malignancies have a worse prognosis than solid tumour patients (Rolston 1992; Rossini 1994; Talcott 1992). The severity and nature of the infection (e.g. bacteraemia, Gram-positive and Pseudomonas aeruginosa infections, resistant organisms) as well as the patient's age may underlie heterogeneity (Elting 1997; Hann 1997; Rolston 1992). More recent guidelines for empirical treatment of febrile neutropenia have emphasized the importance of risk stratification, both for deciding on the setting of therapy (out-patient versus hospitalisation) and for choosing among empirical antibiotics (monotherapy versus combination therapy) (Freifeld 2011; Tam 2011).
We undertook this systematic review to assess the evidence for combination therapy versus monotherapy in patients with febrile neutropenia in clinical trials. In 2002, the first version of this review was published. Results showed no advantage of combination therapy with regard to all cause mortality, the primary outcome assessed and an increased rate of nephrotoxicity with the combined regimen. Most trials compared a broad-spectrum beta-lactam with an older beta-lactam combined with an aminoglycoside; however comparisons performed to directly assess our research question, that is, trials comparing the same beta-lactam with or without an aminoglycoside, were rare. We called for further studies assessing directly the clinical implications of synergism, and further trials comparing different beta-lactams were discouraged in our recommendations (Paul 2003). In 2008 we updated our systematic review with new evidence that had accumulated since publication of the first version of our review; no significant differences were presented in terms of outcomes or subsequent recommendations. At present we are undertaking to update the review to include new evidence that has accumulated since the previous version.
To compare the effectiveness of beta-lactam monotherapy versus that of beta-lactam-aminoglycoside combination therapy in febrile neutropenic cancer patients. In addition, to compare the effectiveness of the two treatment modalities in the following subgroups of neutropenic participants:
- Participants with an absolute neutrophil count of less than 100/mm³
- Participants with microbiologically documented infection
- Participants with documented Pseudomonas aeruginosa infection
- Bacteraemic participants
- Participants with an underlying haematological malignancy or bone marrow transplantation
The following hypotheses were tested for the comparison between participants treated with beta-lactam monotherapy and those treated with beta-lactam-aminoglycoside combination therapy:
- There is no difference in the number of deaths in febrile neutropenic patients
- There is no difference in the number of deaths in the above subgroups of febrile neutropenic patients
- There is no difference in the number of treatment failures in all febrile neutropenic patients and in the defined subgroups
- There is no difference in the number and severity of adverse effects among all patients
- There is no difference in the rate of resistant colonisation and super-infection among all neutropenic patients
Criteria for considering studies for this review
Types of studies
Randomised or quasi-randomised trials comparing any beta-lactam antibiotic monotherapy with any combination of a beta-lactam and an aminoglycoside antibiotic, for the treatment of febrile neutropenia in cancer patients. Allocation to these regimens had to occur initially, before administration of any other antibiotics for the specific febrile episode and, empirically, before detection of pathogen/s or their susceptibilities.
Trials with randomly assigned participants with microbiologically documented infection (e.g. Pseudomonas aeruginosa infection, Gram-negative bacteraemia) were excluded, as were trials comparing short versus long courses of aminoglycoside treatment, because in both cases randomisation to combination treatment versus monotherapy did not occur empirically (referred to as semi-empirical studies).
Types of participants
Febrile cancer patients with neutropenia, as defined in the study, induced by chemotherapy or bone marrow transplantation. Neonates and preterm babies were excluded.
Types of interventions
The following antibiotic regimens were compared:
- Intravenous beta-lactam antibiotic given as monotherapy, including:
- Antipseudomonal carboxy-penicillins or ureido-penicillins ± beta-lactamase inhibitor (piperacillin, piperacillin/clavulanate, ticarcillin-clavulanate, azlocilin, mezlocillin)
- Cephalosporins (ceftazidime, ceftriaxone, cefoperazone, cefoxitin, cefuroxime, cefepime, cefpiramide)
- Carbapenems (imipenem/cilastatin, meropenem)
Studies comparing the same beta-lactam, with the addition of an aminoglycoside to one arm ('same beta-lactam'), were analysed separately from studies comparing different beta-lactams ('different beta-lactam').
- Combination duotherapy of an intravenous beta-lactam antibiotic (as specified) with one of the following aminoglycosides given intravenously:
Types of outcome measures
Death at end of follow-up for the infectious episode, up to 30 days (all cause mortality).
- Treatment failure: a composite end point comprising one or more of the following: death; persistence, recurrence or worsening of clinical signs or symptoms of presenting infection; any modification of the assigned empirical antibiotic treatment.
- Infection related mortality, as reported in the study.
- Duration of hospital stay.
- Dropouts before end of study.
- Super-infection: new, persistent or worsening symptoms and/or signs of infection associated with the isolation of a new pathogen (different, or different susceptibilities) or the development of a new site of infection.
- Colonisation: isolation during or after therapy of Gram-negative bacteria resistant to the beta-lactam included in the empirical regimen, without symptoms or signs of infection.
- Life threatening or associated with permanent disability.
- Serious－requiring discontinuation of therapy.
- Any other.
Search methods for identification of studies
Relevant randomised trials were identified by searching the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library, Issue 7, 2012), LILACS to August 2012, Database of Abstracts of Reviews of Effects (DARE) (Issue 3, 2012) and MEDLINE and EMBASE to August 2012. We conducted a wide search targeting all randomised trials for the treatment of infection in neutropenic patients for this and other systematic reviews conducted by our group. The detailed search strategies for each database are provided in Appendix 1, Appendix 2 and Appendix 3.
Searching other resources
References of all identified studies as well as major reviews were inspected for more studies. We checked the conference proceedings of the Interscience Conference of Antimicrobial Agents and Chemotherapy (ICAAC) 1995 to 2011, the European Congress of Clinical Microbiology and Infectious Diseases (ECCMID 2001 to 2012) and the American Society of Hematology (ASH) 2003 to 2011. Letters, abstracts and unpublished trials were accepted to reduce the influence of publication bias. Additionally, the first or corresponding author of each included study and pharmaceutical companies were contacted for complementary information or information regarding unpublished trials.
Data collection and analysis
Selection of studies
One review author inspected the abstract of each reference identified by the search and applied inclusion criteria. For possibly relevant articles, the full article was obtained and inspected by two review authors.
Data extraction and management
Two review authors independently extracted data from included trials. In cases of disagreement between the two review authors, a third review author extracted the data. In addition the third review author extracted 10% of the studies, selected randomly. Data extractions were discussed, decisions documented and all authors of included studies contacted for clarification. Justification for excluding studies from the review was also documented. Differences in the data extracted were resolved by discussion. All data were collected on an intention-to-treat (ITT) basis whenever possible.
Assessment of risk of bias in included studies
Trials fulfilling the review inclusion criteria were assessed for risk of bias by two review authors working independently. For the 2012 update, this was done using the criteria described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We primarily assessed the effect of allocation concealment on results, based on the evidence of a strong association between poor allocation concealment and overestimation of effect (Schulz 1995), as defined below:
- Low risk of bias (adequate allocation concealment).
- Moderate risk of bias (uncertainty regarding allocation concealment).
- High risk of bias (inadequate allocation concealment).
In addition to the adequacy of allocation concealment, methods of allocation generation, blinding, incomplete outcome data, selective reporting, the unit of randomisation (patient or febrile episode) and publication status were recorded independently by the two review authors.
Assessment of heterogeneity
Heterogeneity in the results of the trials was initially graphically inspected and assessed by calculating a test of heterogeneity (Chi-square). We anticipated between-trial variation in estimation of morbidity and mortality for studies comparing the same beta-lactam and studies comparing different beta-lactams (Elphick 2001). These were separated when heterogeneity was observed. Further heterogeneity was explored through subgroup analysis, assessing the above-defined patient subgroups (Objectives).
A funnel plot estimating the precision of trials (plots of the log of the risk ratio for efficacy against the sample size) was examined to estimate potential selection bias (such as publication bias) and to assess whether effect estimates are associated with study size.
Adjusted means were calculated and corrected by the inverse of the variance. We searched for the correlation between mortality and treatment failure, to assess the clinical relevance of treatment failure and infection related mortality outcomes in these studies. Correlations were tested for significance using a non-parametric test (Spearman) using the Statistical Package for the Social Sciences (SPSS) version 14.0. Numbers needed to treat or harm were calculated as 1/(CER-CER*RR), where CER is the control event rate and RR is the risk ratio.
Dichotomous data were analysed by calculating the risk ratio (RR) for each trial with the uncertainty in each result expressed with the use of 95% confidence intervals (CIs). A fixed-effect model was used throughout the review, unless significant heterogeneity was observed (P < 0.1 or I
Description of studies
The computerised search strategy identified a large number of randomised trials assessing the treatment of febrile neutropenia－not all of which were relevant for the present review. These were screened for trials assessing beta-lactam-aminoglycoside combination therapy versus beta-lactam monotherapy. Ninety-five publications of RCTs were considered eligible for this review.
Twenty-three publications of 22 trials were excluded (Characteristics of excluded studies). Allocation to monotherapy versus combination therapy was non-random in five studies, randomisation to monotherapy versus combination therapy was semi-empirical in three trials (Bodey 1976; EORTC 1987; Pegram 1989), the comparator regimens were incompatible with our inclusion criteria in nine trials, and non-neutropenic patients were included in three trials (D'Antonio 1992; Fainstein 1983; Hoepelman 1988), in which results for neutropenic patients only could not be extracted. One trial randomly assigned participants to treatment with ticarcillin-clavulanate versus ticarcillin-clavulanate+amikacin; however participants who had undergone bone marrow transplantation were allocated to combination therapy only, over-riding the random allocation (Bru 1986); another trial comparing imipenem versus ceftazidime versus amikacin was excluded, because it was presented as an ongoing study in a conference in 1986, no further publication of the study was found and we were not able to contact the authors (Moreno-Sanchez 1992).
Seventy-one trials described in 89 publications are included in the review (Characteristics of included studies; secondary publications are listed under their primary reference). The trials were published between 1983 and 2012. Three trials were added since the previous version of this review, all published between 2007 and 2012. Forty-three trials reported data on all cause mortality and 41 reported on infection related mortality. Data regarding treatment failure were available for all trials. Thirty-one trials contained usable information for super-infections, and 49 trials are included in the adverse event analysis.
Eight included trials, presented in conference proceedings between 1987 and 2002, were published in abstract form only. Supplementary data from the authors were available for two of these (Cornely 2001, Hense 2000). Additional information on trial methods and/or on mortality was available from 24 full-text publications ('unpublished data' in the reference description).
Patient and infection characteristics
Most trials included adult cancer patients. Fourteen trials included only children, and another 14 trials included both adults and children. Most trials included participants with haematological cancer: 35 trials included only patients with haematological malignancies, and in another 32 trials most patients had haematological cancer. Bone marrow transplant patients were excluded from three trials. Patients with septic shock were specifically excluded from four trials; most trials did not refer to patients with septic shock, and in the few trials that did report patients with septic shock, only a few patients were included (1% to 6% of patients in five trials reporting the number of patients with shock on admission).
The ratio between Gram-negative and Gram-positive bacteria among all included studies was 0.69. The adjusted mean rate of infection caused by Gram-negative bacteria was 11.5% of participants. Pseudomonas aeruginosa, a commonly implicated pathogen of febrile neutropenia in the past, was isolated in only 1.7% of included participants, constituting 15.3% of all documented Gram-negative isolates.
Surveillance cultures were performed in nine trials.
The same beta-lactam was compared in 16 of 71 included trials. In these trials the beta-lactam was ceftazidime (seven trials), piperacillin-tazobactam (four trials), cefepime (three trials), imipenem (two trials－one of which included four arms and assessed both ceftazidime and imipenem) and cefoperazone (one trial). All other trials compared one beta-lactam (usually a new drug) with a narrower-spectrum beta-lactam combined with an aminoglycoside. The most common mono-combi beta-lactam comparison was between a carbapenem and a cephalosporin (18 trials). Other comparisons included cephalosporin-cephalosporin (11 trials), cephalosporin-penicillin (nine trials), carbapenem-penicillin (nine trials), penicillin-cephalosporin (four trials) and penicillin-penicillin (three trials), respectively.
The most commonly tested aminoglycoside was amikacin (43 trials), followed by tobramycin (14 trials), gentamicin (11 trials) and netilmicin (three trials). Aminoglycosides were administered once daily in 16 trials. Aminoglycosides were administered for the duration of treatment in all trials, except Tamura 2004, where amikacin was administered only for the first 3 days of combination therapy.
Treatment duration was reported as means or medians. The mean treatment duration ranged from 7 to 15 days (most commonly 9 days); median treatment duration varied between 4 and 9 days (most commonly 9 days).
Risk of bias in included studies
Adequate allocation concealment, using sealed opaque envelopes or central randomisation, was described in 27 trials (Ahmed 2007; Akova 1999; Alanis 1983; Behre 1998; Cometta 1996; Cornely 2001; De la Camara 1997; Del Favero 2001; De Pauw 1994; Gibson 1989; Gorschluter 2003; Hess 1998; Jimeno 2006; Kinsey 1990; Leyland 1992; Lieschke 1990; Marie 1991; Matsui 1991; Norrby 1987; Novakova 1991; Novakova 1990; Petrilli 1991; Pickard 1983; Tamura 2002; Tamura 2004; Wrzesien-Kus 2001; Yamamura 1997). Allocation generation was adequate in a similar number of studies. These studies used tables of random numbers or computer-generated lists. Allocation concealment was inadequate in two trials describing the randomisation only as consecutive (Corapcioglu 2005; Zengin 2011). Randomisation methods were not described in all other trials. Four trials were double-blinded (Del Favero 2001; Ozyilkan 1999; Schuchter 1988; Wade 1989), four single-blinded (Cometta 1996; Duzova 2001; Leyland 1992; Rolston 1992) and the remainder open-randomised trials.
Intention-to-treat (ITT) analysis was presented in 23 of 68 trials included for treatment failure analysis and in 25 of 47 trials included for mortality analysis. Dropouts were reported by their allocation group in 26 of the 45 trials presenting per protocol analysis for treatment failure, permitting a secondary ITT analysis in which dropouts were assumed to be failures (see later, sensitivity analyses for failure). The number of patients excluded from analysis in studies in which ITT analysis was impossible ranged between 3% and 30% and the median rate of excluded patients was 10%. Twelve trials, mostly presented as conference proceedings, addressed 'treated' or 'evaluated' patients, without specifying a different figure for the number of randomly assigned participants (Agaoglu 2001; Borbolla 2001; Duzova 2001; El Haddad 1995; Esteve 1997; Gaytan-Martinez 2002; Kliasova 2001; Marie 1991; Pegram 1984; Pellegrin 1988; Schuchter 1988; Wade 1987). The analysis presumed for these studies was per-protocol.
A pre-determined, defined follow-up period was available from the publication or through author contact for 14 included trials (Behre 1998; Cometta 1996; De la Camara 1997; Del Favero 2001; Gorschluter 2003; Hess 1998; Kojima 1994; Leyland 1992; Norrby 1987; Ozyilkan 1999; Smith 1990; Tamura 2002; Tamura 2004; Yamamura 1997). Follow-up ranged from 72 hours to 1 month following the end of treatment. The observation time was longer than 1 month in two trials (De la Camara 1997; Ozyilkan 1999), both of which reported the outcomes at 1 month post-therapy. In five trials the time of outcome assessment was described more generally as end of treatment, fever, episode or neutropenia (De Pauw 1994; Erjavec 1994; Lieschke 1990; Matsui 1991; Piguet 1988). Two additional trials reported the average follow-up period of their trials (8 and 14 days) but a fixed time for outcome assessment was not specified (Akova 1999; Rolston 1992).
The unit of randomisation was the patient in 23 of the 71 trials (Characteristics of included studies). Episodes comprised the unit of randomisation in all the other trials, which allowed patient re-entry for recurrent episodes of fever and neutropenia. The number of participating patients was given in 74% of trials analysing episodes, and the mean episode-to-patient ratio in these trials was 1.3 (range 1.02 to 2.07). Trials that allowed repeat randomisation of participants for separate episodes of febrile neutropenia did not adjust their analyses to the 'cluster' effect of episodes within single participants and did not provide an intra-patient correlation estimate to allow for adjusted analyses in the meta-analysis. All trials were included in the main analysis and the effect of episode randomisation was assessed through sensitivity analyses.
Effects of interventions
All cause mortality :
All cause mortality was reported in 44 trials, including 7186 episodes. A difference in favour of monotherapy was observed overall (RR 0.87, 95% CI 0.75 to 1.02) ( Analysis 1.1). This difference was not statistically significant, but there was no heterogeneity (P = 0.95, I
No significant differences between monotherapy and combination therapy were observed for the planned subgroups. The trend observed was similar for all comparisons, with RRs favouring monotherapy, with no statistical significance. Moreover, effect estimates favouring monotherapy were larger in subgroups designating participants with a potential worst prognosis:
- Participants with microbiologically documented infection: 13 trials, 1188 episodes, RR 0.81 (95% CI 0.56 to 1.17) ( Analysis 4.1).
- Participants with bacteraemia: 14 trials, 676 episodes, RR 0.74 (95% CI 0.46 to 1.18) ( Analysis 5.1).
- Participants with microbiologically documented Gram-negative infection: 16 trials, 376 episodes, RR 0.64 (95% CI 0.37 to1.11) ( Analysis 6.1).
- Participants with documented Pseudomonas aeruginosa infection: 9 trials, 71 episodes, RR 0.87 (95% CI 0.34 to 2.24) ( Analysis 7.1).
- Participants with haematological cancer: 22 trials 3463 episodes, RR 0.88 (95% CI 0.68 to 1.13) ( Analysis 8.1).
- Participants with severe neutropenia on admission: 6 trials, 737 episodes, RR 0.68 (95% CI 0.37 to 1.24) ( Analysis 9.1).
When the analysis was separated by the monotherapy beta-lactam ( Analysis 10.1), only piperacillin-tazobactam was associated with significantly improved survival compared with combination therapy (RR 0.62, 95% CI 0.40 to 0.96, 5 trials, 1093 episodes). In studies including only children, the RR was 0.80 (95% CI 0.29 to 1.64), and in trials including only adults, the RR was 0.90 (95% CI 0.75 to 1.09) ( Analysis 11.1).
In summary, monotherapy was associated with a trend toward improved survival overall and in all subgroups assessed.
Infection related mortality and treatment failure
Infection related mortality was reported in 41 trials ( Analysis 1.2). No deaths related to infection were reported in nine trials (which did not contribute to the meta-analysis). Monotherapy was associated with a significantly lower rate of infection related mortality compared with combination therapy (RR 0.80, 95% CI 0.64 to 0.99). Results were similar for trials comparing same and different beta-lactams. The number of participants needed to treat with monotherapy to prevent one death related to infection was 95 participants, but 95% CIs were large (49 to 1241 participants).
Studies performed in recent years based their definitions for treatment success and failure on recommendations of the Immuncompromised Host Society (Consensus 1990). Treatment failure reported here is the inverse of "success without modification". It should be noted that we defined treatment failure more broadly in our protocol as death, lack of clinical improvement or any modification of the assigned empirical antibiotic treatment (see earlier, outcomes). Death judged as unrelated to infection was not included in the consensus definitions for failure. Thus other than infection related deaths, treatment failure reflected mainly treatment modifications in trials that were open-label in the vast majority.
In trials comparing the same beta-lactam, a significant advantage was seen with combination therapy (RR 1.11, 95% CI 1.02 to 1.20) with minor heterogeneity (I
Subgroup analyses for trials comparing the same beta-lactams ( Analysis 4.2; Analysis 5.2; Analysis 6.2; Analysis 7.2; Analysis 8.2; Analysis 9.2) demonstrated significant differences in favour of combination therapy for patients with Gram-negative infection (RR 1.34) and severe neutropenia (RR 1.48). No significant differences were observed for the subgroups of participants with any microbiologically documented infection, Pseudomonas aeruginosa infection, bacteraemia and haematological cancer. No specific beta-lactam monotherapy was associated with increased risk for failure ( Analysis 10.2). All subgroup analyses for trials comparing the same beta-lactam were limited by the paucity of trials and participants included.
Similar subgroup analyses for trials comparing different beta-lactams showed that the significant advantage associated with monotherapy persisted in all tested subgroups, except for cases of documented Pseudomonas aeruginosa infection, severe neutropenia and haematological cancer. Similar RRs in favour of monotherapy were observed with the different specific beta-lactams.
No correlation was noted between rates of treatment failure and all cause or infection related mortality in these studies (r = 0.27, P = 0.11, 38 trials, and r = 0.21, P = 0.27, 30 trials, respectively). As expected, infection related mortality was significantly correlated with all cause mortality (r = 0.63, P < 0.001, 29 trials). No significant correlation was noted between publication year and the RRs for mortality or treatment failure.
Twenty-nine trials, including 4961 episodes, reported on the development of bacterial super infections during and after antibiotic treatment ( Analysis 2.1), and 20 trials, including 3437 episodes, reported on fungal super infections ( Analysis 2.2). Equivalence was demonstrated with regard to bacterial super infections (RR 1.02, 95% CI 0.87 to 1.19). Fungal super infections developed more frequently in the combination treatment group (RR 0.70, 95% CI 0.49 to 1.00). Data concerning resistant colonisation were scarce. Five trials supplied data regarding any colonisation (Alanis 1983; Cornelissen 1992; Erjavec 1994; Kojima 1994; Norrby 1987), and comparison of colonisation with resistant Gram-negative bacteria was possible in only two studies (Cornelissen 1992; Norrby 1987). In these studies, resistant Gram-negative bacteria were detected in 5 of 152 participants in the monotherapy group versus 1 of 152 in the combination group. Notably, none of the newer trials included in the updated review performed surveillance cultures, nor did they report on colonisation with resistant bacteria.
Adverse events were significantly more frequent in the combination treatment group. The difference was most remarkable when development of renal failure was compared (RR 0.45, 95% CI 0.35 to 0.57) for any nephrotoxicity ( Analysis 3.3) and (RR 0.16, 95% CI 0.05 to 0.49) for severe nephrotoxicity ( Analysis 3.4). Nephrotoxicity was more common in the combination therapy than in the monotherapy arm also in studies using a once-daily dosing regimen for the aminoglycoside (RR 0.31, 95% CI 0.15 to 0.63, 8 trials, 1707 participants). In assessment of any adverse effect in all trials and in studies grouped by their monotherapy ( Analysis 3.1), an advantage of monotherapy was seen overall (RR 0.87, 95% CI 0.81 to 0.94), and with ceftazidime monotherapy (RR 0.64, 95% CI 0.53 to 0.76). Likewise, discontinuation of study medication due to adverse events occurred more often in the combination group ( Analysis 3.2) (RR 0.61, 95% CI 0.40 to 0.93). The number needed to harm with combination therapy was 34 participants (95% CI, 20 to 104) with regard to any adverse event and 31 participants (95% CI, 24 to 42) with regard to nephrotoxicity.
Duration of hospital stay was non-significantly shorter in the monotherapy group in each of the four trials that reported this outcome: mean 24.8 days (standard deviation (SD) 21 to 31) versus 27.3 days (SD 23 to 56) (De la Camara 1997, data availed through personal correspondence), median 8.6 ± 4 versus 11.8 ± 5.6 (Corapcioglu 2005), mean 9.96 versus 11.93 days (Jimeno 2006) and mean 12.6 ± 5.3 versus 10.6 ± 4.7 (Yildirim 2008) for monotherapy versus combination therapy, respectively. The data were not pooled because variable reporting measures were used.
Funnel plot analyses were undertaken for the two main comparisons: failure and mortality. The funnel plot for mortality was symmetrical (Figure 1). The funnel plots for trials comparing same and different beta-lactams for failure were separated. Among trials comparing the same beta-lactam, the funnel plot was approximately symmetrical (Figure 2); among trials comparing different beta-lactams, an indication that small trials favouring combination therapy are missing may be present (Figure 3).
|Figure 1. All cause mortality.|
|Figure 2. Failure-same BL.|
|Figure 3. Failure-different BL.|
Sensitivity analyses were performed for the primary outcomes－mortality and failure－to assess the impact of study quality on our results.
For mortality, results from studies with adequate allocation concealment (RR 0.88) were similar to results from studies with unclear allocation concealment (RR 0.87; Analysis 12.1), as were results for trials reporting ITT (RR 0.87) versus efficacy analysis (RR 0.88; Analysis 12.2). The effect size was smaller in trials assessing episodes (RR 0.90) compared with trials assessing participants (RR 0.84), although the 95% CI overlapped ( Analysis 12.3). Small and large trials provided similar results, with no study size effect for mortality (comparison 12.5). Unpublished trials and those published only in conference proceedings showed no advantage of monotherapy (RR 1.07, 95% CI 1.07 to 0.72 to 1.59), and trials published in peer reviewed journals showed an advantage of monotherapy (RR 0.84, 95% CI 0.71 to 1.00) ( Analysis 12.4).
For failure among trials comparing the same beta-lactams, no significant differences in the pooled effect estimate were observed for the different methodological measures assessed. In an ITT analysis counting all dropouts as failures, the advantage of combination therapy decreased (RR 1.07; Analysis 12.8). Analysis by episodes was associated with a larger effect estimate in favour of combination therapy (RR 1.16; Analysis 12.10). The only double-blinded trial showed similar results for combination therapy versus monotherapy (Del Favero 2001, Analysis 12.11).
Among trials comparing different beta-lactams, adequate allocation concealment was associated with a smaller effect estimate in favour of monotherapy than was seen with unclear methods (RR 0.94 versus RR 0.87, respectively; Analysis 12.6). ITT analysis in the publication was associated with a smaller effect estimate than was seen with efficacy analysis (RR 0.80, 95% CI 0.71 to 0.91 versus RR 0.95, 95% CI 0.88 to 1.01, respectively; Analysis 12.7), and an ITT analysis assuming that all dropouts were failures did not alter results significantly (RR 0.92, 95% CI 0.86 to 0.97; Analysis 12.8). Analysis by episodes was associated with a smaller effect estimate than analysis by participants (RR 0.95 versus RR 0.89; Analysis 12.10). Smaller trials were associated with a significantly larger effect estimate than was noted in the bigger trials (RR 0.75, 95% CI 0.67 to 0.84 versus RR 0.98, 95% CI 0.92 to 1.03; Analysis 12.9), pointing at the same small studies for effects observed in the corresponding funnel plot analysis (Figure 3). No advantage was seen with monotherapy in double-blind trials ( Analysis 12.11).
For trials comparing same and different beta-lactams, unpublished trials showed no difference between monotherapy and combination therapy, but published trials showed a significant difference favouring combination therapy for trials comparing the same beta-lactams, and favouring monotherapy for trials comparing different beta-lactams ( Analysis 12.12).
Seventy one trials that included more than 10,000 participants were analysed to compare beta-lactam monotherapy with beta-lactam-aminoglycoside combination therapy for the empirical treatment of febrile neutropenic cancer patients. The same beta-lactam was compared in 16 trials, but all other trials compared a broad-spectrum beta-lactam with a narrower-spectrum beta-lactam combined with an aminoglycoside. Most of the participants included in these trials were haematological cancer patients. We assessed all cause mortality as the primary outcome.
Monotherapy was associated with a statistically non-significant lower all cause mortality rate at end of follow-up (30 days) (RR 0.87, 95% CI 0.75 to 1.02). Results for trials comparing same and different beta-lactams were similar. Appropriate trial methods (adequate allocation concealment, ITT analysis and analysis by participants) were associated with similar effect estimates in favour of monotherapy, and no small studies effect was observed. Mortality attributed in the primary studies to infection was significantly lower with monotherapy (RR 0.80, 95% CI 0.64 to 0.99).
Treatment failure was assessed as the primary outcome in all included trials. By definition, its main addition on the rather subjective outcome of infection related mortality is treatment modifications (Consensus 1990). Among trials comparing the same beta-lactams, treatment failure was significantly more frequent with monotherapy. This difference likely reflects mainly physicians' tendency for treatment modifications in open trials comparing one antibiotic regimen with a broader-spectrum regimen. Among trials comparing different beta-lactams, a significant advantage was seen with monotherapy. Adequate trial methods were associated with smaller effect estimates for both 'same' and 'different' comparisons. Notably, in the single double-blind trial comparing the same beta-lactams, failure was equal with combination treatment and with monotherapy, and in three double-blind trials assessing different beta-lactams, the RRs were in the opposite direction compared with those in the other trials. We detected a small studies effect for trials comparing different beta-lactams. This may reflect a publication bias related to trials that assessed a newer monotherapy without showing its advantage.
Bacterial super infections occurred with equal frequency with monotherapy and combination therapy. Fungal super-infections were more common with combination therapy. All adverse events were more common with combination therapy, with a highly significant difference for nephrotoxicity. The pooled effect estimate translated to a number needed to harm of 34 participants (95% CI 20 to 104 participants).
To explain the advantage of monotherapy with regard to all cause mortality, several of the secondary outcomes may be used. Infection related mortality was significantly lower with monotherapy, and fungal super infections occurred more frequently with combination treatment. Fungal infections developing during neutropenia are highly lethal (Lin 2001). Thus, the improvement in survival may indeed be infection related. On the other hand, nephrotoxicity associated with combination therapy is a risk factor for subsequent adverse outcomes. Given these results and those of the methodological quality assessment, it is likely that the both mechanisms contribute to an unbiased advantage in overall survival with monotherapy.
Several hypotheses underlie the use of beta-lactam-aminoglycoside combination therapy for patients with neutropenia and suspected infection. Synergism is usually claimed as the major reason for combination therapy. Synergism was assessed most directly in trials comparing the same beta-lactam. We did not detect the beneficial effects of synergism. A wider spectrum of coverage may be the incentive for the addition of an aminoglycoside depending on local patterns of resistance. Studies included in the review did not supply enough data to allow determination of whether coverage is indeed improved with combination therapy. However, the efficacy of aminoglycosides alone for the treatment of neutropenic patients is doubtful (Bodey 1972; Klastersky 1986); therefore this potential advantage does not seem substantial. Finally, combination therapy is claimed to prevent emergence of resistant pathogens. Development of resistance after antibiotic treatment is difficult to quantify. We intended to extract data regarding colonisation with resistant pathogens following antibiotic treatment, but these data were rarely available. Resistance was therefore indirectly examined through super infections, under the assumption that infection that develops under antibiotic treatment involves resistant pathogens. No difference was noted in the rate of bacterial super infections between monotherapy and combination therapy, and this analysis resulting in an RR close to 1. Fungal super infections developed more frequently with combination therapy, perhaps as a reflection of increased antibiotic spectrum or burden with combination therapy. Thus we could not show an advantage of combination therapy from this aspect.
We chose all cause mortality as the primary outcome, rather than treatment failure or infection related mortality, and have drawn our conclusions from the analysis for all cause mortality. Only a small part of the variance in mortality is explained by infection and its treatment; however, appropriate randomisation should ensure similar distribution of non–infection related risk factors for death between the study groups. Infection related mortality may be prone to bias in that the cause of death is difficult to determine in severely ill cancer patients. Moreover, ignoring deaths due to treatment-related adverse effects and super infections is inappropriate. Early empirical antibiotic treatment is the standard of practice for febrile neutropenic patients because it has been proven to decrease mortality (Hughes 1997; Schimpff 1986). Survival is indeed the objective when an acute infection is treated in cancer patients. Treatment failure indicates mainly modifications of the initial antibiotic regimen, and possibly a longer time to defervescence. The implications of such an outcome are not clear from the clinical point of view. Finally, deaths are objective, but failures cannot be objective when the trials are open. It is important to note that we could demonstrate in this review that assessing treatment failure is probably inappropriate, because no correlation between failure and mortality could be shown.
Our results are congruent with those of several other analyses of beta-lactam-aminoglycoside combination therapy versus beta-lactam monotherapy, showing no advantage associated with combination therapy. We conducted a similar analysis in non-neutropenic participants with sepsis, showing an advantage of monotherapy in trials comparing different beta-lactams, and no difference in trials comparison the same beta-lactam (Paul 2004; Paul 2006a). In an analysis of all RCTs comparing the same beta-lactam in the combination and monotherapy arms, in both neutropenic and non-neutropenic participants, and including semi-empirical studies, we did not find a significant difference in all cause mortality, but we noted significantly more bacterial super infections and increased renal failure with the addition of aminoglycosides (Marcus 2011). An analysis focusing on the development of resistance did not find an advantage associated with combination therapy (Bliziotis 2005). Finally, an analysis of observational studies focusing onPseudomonas aeruginosa infection (mainly bacteraemia), a pathogen with special relevance to neutropenic cancer patients, did not find an advantage for combination therapy (Vardakas 2013).
The major limitations of this review include the lack of complete data concerning mortality (all cause mortality was available for 44 of 71 included trials, 62%) and the paucity of available data regarding specific patient subgroups, such as those with Pseuomonas aeruginosa infection. Other limitations stem from those of the primary studies. Allocation concealment was at low risk of bias in less than 35% of the trials, and nearly all were non-blinded. Many of the trials did not adhere to the principle of ITT analysis, resulting in incomplete data reporting. Most studies used febrile episodes as the unit of randomisation, although recurrent episodes are not independent for any for the outcomes assessed. Finally, follow-up did not seem pre-determined in many of the studies. Reported mortality may have been biased because the time of assessment was not defined in advance. We included trials regardless of their publication status. The differences detected in our review, namely, the advantage of monotherapy with regard to survival and the divergent advantages with regard to failure, existed with larger effect estimates in trials published in peer reviewed journals. The RRs were close to 1 for these outcomes in unpublished trials, mainly conference proceedings. Their inclusion in the meta-analysis tipped the overall RRs toward equivalence.
Implications for practice
Monotherapy can be regarded as the standard of care for the empirical treatment of febrile neutropenic patients. The addition of an aminoglycoside does not improve survival. On the contrary, it is associated with significant morbidity incurred mainly through aminoglycoside-associated nephrotoxicity.
The monotherapies assessed in recent years have included imipenem, meropenem, ceftazidime, piperacillin-tazobactam and cefepime. These beta-lactams have also been assessed in head-to-head trials comparing different monotherapies and have shown similar efficacies, but for cefepime this was associated with increased all cause mortality (Paul 2006). Thus, individual centres should select the best matching monotherapy according to local epidemiology and susceptibility patterns.
RCTs do not support an advantage of combination therapy for Pseudomonas aeruginosa infection and other more severely ill patient subgroups. However the paucity of data precludes firm conclusions regarding these patient subgroups.
Implications for research
Assessment of new beta-lactams for febrile neutropenia should not be performed by comparison with a narrower-spectrum beta-lactam combined with an aminoglycoside. The results of these trials are uniformly unfavourable for patients. Assessment of new beta-lactam monotherapies should be performed by comparison with established monotherapies for febrile neutropenia. This design can and does show the advantages and disadvantages of specific beta-lactams (Paul 2006).
The need for further trials assessing the addition of an aminoglycoside to the same beta-lactam is doubtful given the results of our review, spanning more than two decades of clinical trials in febrile neutropenia and without a change in RRs throughout the years. We can foresee such a need if a reduction in aminoglycoside-related adverse effects is expected, or if new data will point toward drug combinations with a marked synergistic effect－much greater than that observed in current studies. Trials targeting specific patient subgroups, such as those with severe sepsis and septic shock, documented Pseudomonas aeruginosa infection, etc. are warranted.
Future trials should report all cause mortality. The primary outcome used in these studies should be re-defined because with current definitions, no correlation can be noted between failure and the ultimate outcome: survival. This outcome should be defined in a consensus statement and applied universally to permit comparisons and compilation of different studies. The unit of randomisation should be the patient－not the episode. If recurrent episodes are allowed, results for the first randomisation of each patient should be reported separately, or the analysis should be adjusted to the clustering effect of patient episodes. Length of follow-up should be uniform and should be determined before the study is begun.
We would like to thank warmly the Cochrane Gynaecological Cancer Review Group, for their helpful advice and for their assistance in obtaining articles from abroad.
We would like to express our appreciation to all the authors who responded to our letters and supplied additional information on their studies: Drs. Will (Smith 1990 trial), Wrzesien-Kus, Ahmed, Jimeno, Pettrili, Tamura and Vallejos (Rodriguez 1995), Gibson, Donnelly and De Pauw (De Pauw 1983; De Pauw 1994 trials), Pickard and Rotstein (Yamamura 1997 trial), Kojima, Kinsey, Norrby, Matsui, Ozyilkan, Dincol, Doyen and Michaux (Doyen 1983 trial), Duzova, Agaoglu and Karakas (Agaoglu 2001 trial), Jacobs, De la Camara and Sarper (Zengin 2011), Drs. Glasmacher, Hense and Lieschke, who supplied their full unpublished manuscripts, Drs. Keddie and Wilks of the AstraZeneca Company for supplying their data for the Behre 1998 and De la Camara 1997 trials, Dr. Sawae for clarifying the details of his study, and Dr. Cornelly for supplying full results for his yet unpublished trial (Cornely 2001). We would also like to thank Professor Bodey for his response and comments on the previous version of this review.
In addition we would like to thank the authors who responded even though additional data were unavailable: Drs. Morgan, Piccart, Rehm (Alanis 1983 trial),
The National Institute for Health Research (NIHR) is the largest single funder of the Cochrane Gynaecological Cancer Group.
The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the NIHR, NHS or the Department of Health
Data and analyses
- Top of page
- Summary of findings [Explanations]
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Index terms
Appendix 1. MEDLINE search strategy
1 exp Neoplasms/
2 Bone Marrow Transplantation/
3 (cancer* or tumor* or tumour* or neoplas* or malignan* or carcinoma* or adenocarcinoma* or leukemia* or leukaemia* or bone marrow transplant*).mp.
4 1 or 2 or 3
5 exp Agranulocytosis/
6 (agranulocytosis or neutropen* or neutropaen* or granulocytopen* or granulocytopaen* or granulopen* or granulopaen*).mp.
7 5 or 6
8 exp beta-Lactams/
9 exp Anti-Bacterial Agents/
10 (beta-lactam* or antibiotic* or antimicrob* or anti-microb* or antibacteria* or anti-bacteria*).mp.
11 8 or 9 or 10
12 exp Aminoglycosides/
13 (aminoglycoside* or gentamicin or gentamycin or amikacin or amikacyn or tobramicin or tobramycin or kanamicin or kanamycin or netilmicin or netilmycin).mp.
14 12 or 13
15 4 and 7 and 11 and 14
16 randomized controlled trial.pt.
17 controlled clinical trial.pt.
20 drug therapy.fs.
24 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23
25 15 and 24
mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier, pt=publication type, ab=abstract, sh=subject heading, ti=title
Appendix 2. EMBASE search strategy
1 exp neoplasm/
2 exp bone marrow transplantation/
3 (cancer* or tumor* or tumour* or neoplas* or malignan* or carcinoma* or adenocarcinoma* or leukemia* or leukaemia* or bone marrow transplant*).mp.
4 1 or 2 or 3
6 exp neutropenia/
7 (agranulocytosis or neutropen* or neutropaen* or granulocytopen* or granulocytopaen* or granulopen* or granulopaen*).mp.
8 5 or 6 or 7
9 exp antiinfective agent/
10 (beta-lactam* or antibiotic* or antimicrob* or anti-microb* or antibacterial* or anti-bacteria*).mp.
11 9 or 10
12 exp aminoglycoside antibiotic agent/
13 (aminoglycoside* or gentamicin or gentamycin or amikacin or amikacyn or tobramicin or tobramycin or kanamicin or kanamycin or netilmicin or netilmycin).mp.
14 12 or 13
15 4 and 8 and 11 and 14
16 crossover procedure/
17 double-blind procedure/
18 randomized controlled trial/
19 single-blind procedure/
22 (crossover* or cross over* or cross-over*).mp.
24 (double* adj blind*).mp.
25 (singl* adj blind*).mp.
29 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 or 25 or 26 or 27 or 28
30 15 and 29
[mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer, device trade name, keyword]
Appendix 3. CENTRAL search strategy
#1 MeSH descriptor Neoplasms explode all trees
#2 MeSH descriptor Bone Marrow Transplantation, this term only
#3 (cancer* or tumor* or tumour* or neoplas* or malignan* or carcinoma* or adenocarcinoma* or leukemia* or leukaemia* or bone marrow transplant*)
#4 (#1 OR #2 OR #3)
#5 MeSH descriptor Agranulocytosis explode all trees
#6 (agranulocytosis or neutropen* or neutropaen* or granulocytopen* or granulocytopaen* or granulopen* or granulopaen*)
#7 (#5 OR #6)
#8 MeSH descriptor beta-Lactams explode all trees
#9 MeSH descriptor Anti-Bacterial Agents explode all trees
#10 beta-lactam* or antibiotic* or antimicrob* or anti-microb* or antibacterial* or anti-bacteria*
#11 (#8 OR #9 OR #10)
#12 MeSH descriptor Aminoglycosides explode all trees
#13 (aminoglycoside* or gentamicin or gentamycin or amikacin or amikacyn or tobramicin or tobramycin or kanamicin or kanamycin or netilmicin or netilmycin)
#14 (#12 OR #13)
#15 (#4 AND #7 AND #11 AND #14)
Last assessed as up-to-date: 7 June 2013.
Protocol first published: Issue 2, 2001
Review first published: Issue 2, 2002
Contributions of authors
Mical Paul performed the search and article retrieval; applied inclusion and exclusion criteria; performed quality assessment and data extraction; contacted authors; analysed results and wrote the review. Yaakov Dickstein conducted the search for the 2012 update, extracted the data from new trials, entered data into RevMan and wrote the review for the 2012 update. Karla Soares-Weiser applied inclusion and exclusion criteria; performed data extraction; analysed results－all for the previous version of the review and commented on all drafts and final version of the review. Simona Grozinsky-Glasberg assisted with search; retrieved articles; applied inclusion and exclusion criteria and assisted in data extraction－all for the previous version of the review. Leonard Leibovici performed search; applied inclusion and exclusion criteria; performed data extraction; assisted with author correspondence; analysed results; assisted in writing the review and commented on all drafts and final version of the review.
Declarations of interest
Sources of support
- Rabin Medical Center, Beilison Campus, Skidal Foundation, Israel.
- Tel-Aviv University, Sackler Faculty of Medicine, Israel.
- EU 5th Framework: TREAT project (grant number: 1999-11459), Not specified.
Medical Subject Headings (MeSH)
Aminoglycosides [adverse effects; *therapeutic use]; Anti-Bacterial Agents [*therapeutic use]; Cause of Death; Combined Modality Therapy [adverse effects; methods]; Neoplasms [*complications]; Neutropenia [*drug therapy; mortality]; Randomized Controlled Trials as Topic; beta-Lactams [*therapeutic use]
MeSH check words
Adult; Child; Humans
* Indicates the major publication for the study