Summary of findings
Description of the condition
Sepsis is defined as clinical evidence of infection, accompanied by a systemic inflammatory response such as fever. When associated with organ dysfunction, decreased blood flow in an organ (hypoperfusion) or abnormally low blood pressure (hypotension), sepsis is defined as severe (Bone 1992; Mandell 2004). Sepsis may be a response to direct microbial invasion or may be elicited by microbial signal molecules or toxin production. Bacterial infections may be lethal, with fatality rates ranging from less than 10% to more than 40% for those with severe sepsis (Moore 2001; Rangel-Frausto 1995; Russell 2000).
Description of the intervention
Antibiotic treatment for bacterial infection is usually initiated empirically, before the causative bacteria are identified and their susceptibility to antibiotic treatment is ascertained. Appropriate empirical antibiotic treatment, defined as that matching the in vitro susceptibility of subsequently identified bacteria, has been shown to halve the fatality associated with sepsis (Ibrahim 2000; Leibovici 1998; Paul 2010). Causative bacteria are identified only in about one-third of patients with sepsis overall (Paul 2006a). At this time, treatment is tailored according to the antibiotic susceptibilities of identified bacteria. Both empirically and after bacterial identification, single or combination antibiotic treatment may be given.
How the intervention might work
Combination antibiotic therapy offers several theoretical advantages. First, it can be used to broaden the spectrum of antibiotic coverage when used empirically to increase the chance of covering the causative bacteria. Second, the combination may possess an enhanced potential (synergism) when compared with the additive effect of each of the antibiotics assessed separately. Synergism between specific beta lactam antibiotics and aminoglycoside antibiotics has been shown in vitro for Gram-negative bacteria and specifically for Pseudomonas aeruginosa (Giamarellou 1986; Klastersky 1976; Klastersky 1982), Staphylococcus aureus, Enterococcus sp. and Streptococcus sp. (Bach 1980; Korzeniowski 1978; Levy 1979; Saleh-Mghir 1992; Sande 1974; Torres 1993). Third, combination therapy has been claimed to suppress the emergence of subpopulations of microorganisms resistant to antibiotics (Allan 1985; Milatovic 1987). Disadvantages of combination therapy may include additional costs, enhanced drug toxicity, possible induction of resistance caused by the broader antibiotic spectrum (Manian 1996; Weinstein 1985) and possible antagonism between specific drug combinations (Moellering 1986).
Why it is important to do this review
Several systematic reviews have addressed the clinical effects of beta lactam-aminoglycoside combinations for the treatment of sepsis, bacteraemia or specific types of infection. In previous versions of the current review (Paul 2003; Paul 2006), we found no advantage of combination therapy over monotherapy and an increased rate of renal toxicity with combination therapy. In a separate systematic review assessing the same intervention for cancer patients with neutropenia (excluded from the current review), similar results were found, with a small advantage of monotherapy when compared with a narrower-spectrum beta lactam combined with an aminoglycoside (Paul 2013). To fully examine the implications of in vitro synergism, we pooled all randomized controlled trials comparing one beta lactam antibiotic versus the same beta lactam with an aminoglycoside (Marcus 2011). Overall, no advantage emerged for combination therapy; the subgroup of P aeruginosa bacteraemia was too small to allow definitive conclusions. Safdar et al. focused on Gram-negative bacteraemia and compiled randomized trials and observational studies (Safdar 2004). In the subgroup of P aeruginosa bacteraemia, an advantage was reported for combination therapy, but aminoglycosides were used as monotherapy in some of the trials (Paul 2005). Kumar et al. performed a meta-regression analysis showing that an advantage of combination therapy involved mortality rates in randomized and observational studies (Kumar 2010), but how much of this was due to the inherent association between odds ratios and event rates and how much to a true clinical effect was unclear (Paul 2010a). When restricting the analysis to infective endocarditis (caused by Gram-positive bacteria), Falagas et al. found no advantage of combination therapy (Falagas 2006). The same group of authors reported no advantage of combination therapy with regard to emergence of antibiotic resistance following therapy (Bliziotis 2005).
Despite the large body of evidence pointing against a benefit for combination therapy, most recommendations and guidelines still recommend combination therapy, and combination therapy is frequently used in clinical practice. In the guidelines for the management of severe sepsis of the "Surviving Sepsis Campaign", initial combination therapy is recommended (Dellinger 2008). Narrowing the spectrum of coverage after three to five days is recommended, except for infections caused by P aeruginosa and infections among neutropenic patients, for whom continued combination treatment is advised. Beta lactam-aminoglycoside treatment is recommended for pneumonia caused by P aeruginosa (Sun 2011). Treatment of infective endocarditis has traditionally consisted of beta lactam-aminoglycoside combinations based on in vitro synergy studies and experimental studies. Although current guidelines, acknowledging the lack of evidence, recommend beta lactam monotherapy as first-line therapy for most pathogens, combination therapy is still suggested as optional treatment and is recommended for resistant bacteria, mainly Enterococcus sp. and S aureus (Baddour 2005).
Our objectives were to compare beta lactam monotherapy versus beta lactam-aminoglycoside combination therapy in patients with sepsis and to estimate the rate of adverse effects with each treatment regimen, including the development of bacterial resistance to antibiotics.
Criteria for considering studies for this review
Types of studies
We included randomized or quasi-randomized controlled trials.
Types of participants
We included hospitalized participants with sepsis acquired in the community or in the hospital. We defined sepsis as clinical evidence of infection plus evidence of a systemic response to infection (Bone 1992). We excluded neonates and preterm babies. We also excluded studies including more than 15% neutropenic patients.
Types of interventions
We considered studies comparing the antibiotic regimens described below.
- Any intravenous beta lactam antibiotic given as monotherapy, including:
- beta lactam drugs plus beta lactamase inhibitors (e.g. co-amoxiclav);
- cephalosporins (e.g. ceftazidime, cefotaxime); or
- carbapenems (e.g. imipenem, meropenem).
- Combination therapy of a beta lactam antibiotic (as specified) with one of the following aminoglycoside antibiotics:
- isepamicin; or
Types of outcome measures
All-cause mortality by the end of the study follow-up.
- Treatment failure defined as death and/or one or more serious morbid events (persistence, recurrence or worsening of clinical signs or symptoms of presenting infection; any modification of the assigned empirical antibiotic treatment; or any therapeutic invasive intervention required and not defined in the protocol). If defined differently, the study definitions were documented.
- Length of hospital stay.
- Superinfection: recurrent infections, defined as new, persistent or worsening symptoms and/or signs of infection associated with the isolation of a new pathogen (different pathogen or same pathogen with different susceptibilities) or the development of a new site of infection.
- Colonization by resistant bacteria: the isolation of bacteria resistant to the beta lactam antibiotic, during or following antibiotic therapy, with no signs or symptoms of infection.
- Adverse effects:
- life-threatening or associated with permanent disability (severe nephrotoxicity; ototoxicity; anaphylaxis; severe skin reactions);
- serious: requiring discontinuation of therapy (other nephrotoxicity; seizures; pseudomembranous colitis; other allergic reactions); or
- any other (other gastrointestinal; other allergic reactions).
Search methods for identification of studies
We formulated a comprehensive search strategy in an attempt to identify all relevant studies regardless of language or publication status (published, unpublished, in press and in progress).
We searched the Cochrane Infectious Diseases Group specialized trials register for relevant trials up to September 2011 using the following search terms: ((aminoglycoside* OR netilmicin* OR gentamicin* OR amikacin* OR tobramycin* OR streptomycin* OR isepamicin* OR sisomicin*) AND (pneumonia* OR infection OR infect* OR sepsis OR bacter* OR bacteremia OR septicemia).
In this updated review, we searched the Cochrane Central Register of Controlled Trials (CENTRAL, 2013, Issue 11; see Appendix 1 for a detailed search strategy), PubMed (1966 to November 2013; see Appendix 2), EMBASE (Ovid SP, 1980 to November 2013; see Appendix 3) and LILACS (via BIREME interface, 1982 to November 2013; see Appendix 4). We combined our PubMed search strategy with the Cochrane highly sensitive search strategy for identifying randomized controlled trials (RCTs), as suggested in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We modified this RCT filter for use in EMBASE and LILACS. In our previous review (Paul 2006), we searched the databases until July 2004.
Searching other resources
We searched the Interscience Conference of Antimicrobial Agents and Chemotherapy conference proceedings (1995 to 2012) and the European Congress of Clinical Microbiology and Infectious Diseases (2000 to 2013) for relevant abstracts.
We contacted the first or corresponding author of each included study and researchers active in the field for information regarding unpublished trials or for complementary information on their own trials.
We also checked the citations of major reviews and of all trials identified by the above methods for additional studies.
We did not have a language restriction.
Data collection and analysis
Selection of studies
One review author (MP for the original review; AL for the 2013 update) inspected the abstract of each reference identified in the search and applied the inclusion criteria. When relevant articles were identified, the full article was obtained and was inspected independently by two review authors (MP, AL, IS or LL).
Data extraction and management
Two review authors (MP, Ishay Silbiger or SG-G) independently extracted data from included trials in the original review, and AL and MP extracted data for the 2012 update. In case of disagreement between the two review authors, a third review author (LL) independently extracted the data. A third review author (LL) also extracted the data in 10% of the studies, selected at random. We discussed data extraction, documented decisions and contacted authors of all studies for clarification. We resolved differences in the data extracted by discussion. We also documented the justification for excluding studies from the review.
We identified the trials by the name of the first author and the year in which the trial was first published, and we listed them in chronological order. We extracted, checked and recorded the following data.
Characteristics of trials
- Date, location and setting of trial.
- Publication status.
- Country of origin.
- Design (intention-to-treat, method of randomization).
- Duration of study follow-up.
- Performance of surveillance cultures (routine cultures for the detection of colonization).
- Sponsor of trial.
Characteristics of participants
- Number of participants in each group.
- Age (mean and standard deviation, or median and range).
- Number of participants with renal failure before treatment.
- Number of participants with shock.
Characteristics of infection
- Number of participants with infection caused by bacteria resistant to the administered beta lactam antibiotic.
- Number of participants with nosocomial infection.
- Number of participants with bacteraemia.
- Number of participants with bacteriologically documented infection.
- Number of participants with infection caused by Gram-negative bacteria.
- Number of participants with Gram-negative bacteraemia.
- Number of participants with documented Pseudomonas infection (Pseudomonas isolated in blood or specimen(s) obtained from suspected site(s) of infection).
- Number of participants with:
- urinary tract infection;
- intra-abdominal infection;
- skin and soft tissue infection; and
- infection of unknown origin.
Characteristics of interventions
- Antibiotic type and dose.
- Duration of therapy (mean).
Characteristics of outcome measures
- Number of deaths at the end of the follow-up period.
- Number of participants failing treatment (as defined).
- Adverse reactions (as defined) in each group.
- Loss of follow-up (dropouts) before the end of the study in each group.
- Number of participants developing superinfection.
- Number of participants developing colonization (as defined) with resistant bacteria.
- Duration of fever and hospital stay.
We collected outcome measures on an intention-to-treat basis whenever possible. When such data were not presented, we sought information from the trial authors, and, if unavailable, per-protocol results were used. For failure outcome, we performed sensitivity analyses comparing these results with a 'presumed all intention to treat', which we achieved by counting all dropouts as failures. We could not make such an assumption in studies that did not specify the number of dropouts per study arm, and we analysed these studies separately.
Assessment of risk of bias in included studies
We assessed the risk of bias of the trials to be included for random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, blinding and incomplete outcome data (for mortality and failure outcomes), selective reporting, intention-to-treat analysis and number of participants excluded from the outcome assessment. Two review authors (MP, AL, Ishay Silbiger and Karla Soares-Weiser ) independently performed the risk of bias assessment by classifying each item separately as low, unclear or high risk of bias according to the criteria suggested by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).
Measures of treatment effect
We calculated risk ratios for dichotomous outcomes. For length of stay, we extracted the measure reported in the study (mean or median) along with its dispersion measure.
Unit of analysis issues
We expected all studies to be individually randomly assigned and to recruit each participant only once into the trial; thus no unit of analysis issues were expected for this review.
Dealing with missing data
We contacted the first or corresponding author of each included study and researchers active in the field to ask for information regarding unpublished trials or complementary information on their own trials.
Assessment of heterogeneity
We initially assessed heterogeneity by visual inspection of the forest plots. Statistical assessment was based on the Chi
Assessment of reporting biases
We visually examined a funnel plot of standard error (SE) (log(risk ratio)) versus risk ratio of each study to estimate small study effects, including publication bias or other. Statistical testing for funnel plot asymmetry was conducted using Egger's regression (Comprehensive Meta Analysis, version 2.2). The funnel plot was examined for the outcomes of mortality and failure.
We used the Mantel-Haenszel fixed-effect model to pool risk ratios. We did not plan to pool results for length of stay because this variable is not normally distributed.
Subgroup analysis and investigation of heterogeneity
We explored heterogeneity by subgroup analysis of the different types of infection.
- Infections caused by Gram-negative bacteria and P aeruginosa.
- Gram-negative bacteraemia.
- Urinary tract infection and non–urinary tract infection, assuming that the latter might be more serious and thus might benefit more from combination therapy.
- Gram-positive infection and endocarditis.
For subgroup analyses, we analysed the outcomes of mortality and failure. For Gram-positive infection, we also analysed microbiological failure.
We analysed separately studies at low risk of bias with regard to allocation concealment, generation, blinding and incomplete outcome data reporting. We based conclusions regarding the effect of risk of bias on results on the evidence of a strong association between poor allocation concealment and overestimation of effect (Schulz 1995).
Description of studies
Results of the search
The search strategy resulted in 6562 references. We filtered double references and screened 3629 different abstracts for inclusion. We did not further evaluate studies in which the comparator antibiotic regimens were clearly incompatible with inclusion criteria. We similarly excluded non-randomized and non-human studies. We retrieved 159 studies for full-text inspection, of which we excluded 72 publications. Eighty-three articles were deemed eligible for inclusion, of which 14 were secondary publications. One is ongoing (Characteristics of ongoing studies); thus we have included 69 trials in this review (Figure 1). Five trials are included in the current update that were not included in the original review (Banasal 2006; Damas 2006; Hasali 2005; Figueroa-Damian 1996; García Ramírez 1999).
|Figure 1. Study flow diagram.|
Main study characteristics are detailed in the table Characteristics of included studies. The included studies were performed between the years 1968 and 2006. Twenty-two were multi-centred. Twenty-one were performed in the United States or Canada, 35 in Europe and 14 in other countries. The studies included 7863 participants. The median number of included participants per trial was 80 (range 20 to 580). Four trials (Banasal 2006, Cardozo 2001; Hasali 2005, Naime Libien 1992) included children, and all other trials were restricted to or included mostly adults.
The studies differed by the type of population and the type of infection targeted (see Characteristics of included studies). Forty-five trials included participants with severe sepsis, suspected Gram-negative infection or pneumonia (designated as 'sepsis'). The adjusted mean fatality rate in these studies was 8.5%.Twelve trials included participants with intra-abdominal infection, related mainly to the biliary tract (designated 'abdominal'). The mean fatality rate in these trials was 1.7%. Seven trials were restricted to participants with urinary tract infection (UTI), all hospitalized, mainly women. Five of these studies reported on mortality, and no deaths occurred in four. Finally, five of the studies included in the review targeted participants with Gram-positive infection. Four studies addressed participants with endocarditis caused by S aureus (Abrams 1979; Korzeniowski 1982; Ribera 1996) or streptococci (Sexton 1998). One study included any staphylococcal infection (Coppens 1983).
Most studies compared beta-lactam monotherapy vs. beta-lactam-aminoglycoside combination therapy as the initial, empirical antibiotic treatment administered to participants. Four studies assessed the empirical and definitive treatment of a specific infection by randomly assigning participants empirically and evaluating only those who subsequently fulfilled criteria for the specific infection. Two such studies randomly assigned participants with suspected endocarditis and evaluated only those with S aureus bacteraemia and proven endocarditis (Abrams 1979; Korzeniowski 1982). The other two randomly assigned participants with suspected biliary tract infection and evaluated only participants with a surgically proven diagnosis (Gerecht 1989; Yellin 1993). Non-evaluated participants in these studies were not counted as dropouts because the study protocol a priori defined evaluation only for participants fulfilling definitive criteria. Eight studies, focusing on participants with specific infections or pathogens (e.g. cholecystitis, staphylococcal infections), tested the effect of monotherapy versus combination therapy semi-empirically. In these studies (designated 'semi-empirical', see Characteristics of included studies), randomization occurred after the specific infection was documented and participants could have received prior antibiotic treatment for this infection. Analysis of empirical and semi-empirical studies was not separated.
The specific antibiotic regimens used are detailed in the table Characteristics of included studies. Forty-seven studies compared a single beta lactam drug versus a different, narrower-spectrum beta lactam combined with an aminoglycoside (designated 'different BL'). Sixteen 'different BL' studies reported baseline susceptibility rates of the pathogens isolated on admission to the beta lactam. The beta lactam used in the combination arm covered fewer pathogens than the monotherapy beta lactam in 13 studies, and the opposite occurred in two studies only. Twenty-two studies compared the same beta lactam in the combination and monotherapy arms (designated 'same BL'). Results obtained from studies comparing same and different beta lactams were kept separated throughout all efficacy analyses. The beta lactam monotherapies used in the studies and their dosing are detailed in Table 1. The aminoglycoside was administered once daily in nine trials (Cardozo 2001; Damas 2006; García Ramírez 1999, Hasali 2005; Jaspers 1998; Rubinstein 1995; Sandberg 1997; Sexton 1998; Speich 1998). Other trials administered the aminoglycosides multiple times daily (49 trials) or did not specify the administration schedule (11 trials). Mean antibiotic treatment duration ranged between three and 17.5 days in the sepsis studies, between 6.8 and 11.9 days in the abdominal studies, between 4.1 and seven days in the UTI studies and between two and four weeks in the endocarditis studies.
We excluded 72 publications, representing 69 studies (Figure 1; Characteristics of excluded studies). Several studies compared monotherapy versus combination therapy among participants with cystic fibrosis. Participants in these studies typically did not have fever or other signs of sepsis when entering the trial and thus did not fulfil inclusion criteria for this review. These studies are included in a separate review (Elphick 2005).
Risk of bias in included studies
|Figure 2. Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.|
|Figure 3. Risk of bias summary: review authors' judgements about each risk of bias item for each included study.|
Thirty-two percent of the studies (22/69) reported adequate allocation concealment and thus considered were at low risk of bias. Four studies were graded as high risk of bias (Duff 1982; García Ramírez 1999; Hasali 2005; Landau 1990). No information was available for the other studies (37 studies), or envelopes were used but were not described as sealed or opaque (six studies).
Allocation generation was considered at low risk of bias in 51% of the studies (35/69). No information was available for 30 studies. Two studies were at high risk of bias, using participant identification numbers (Duff 1982; Landau 1990). Both allocation generation and concealment were at low risk in 29% of the studies (20/69).
Most studies were open. Although these were considered at high risk of bias, for mortality assessment the lack of blinding after randomization probably did not introduce bias. Two studies, including 226 participants, were double-blinded (Sanfilippo 1989; Smith 1984). Outcome assessors were blinded in four studies (Brown 1984; Dupont 2000; Rubinstein 1995; Verzasconi 1995). Clinicians were blinded to treatment in one study (Yellin 1993).
Incomplete outcome data
We separated included studies into four different study types with relation to outcome reporting.
- Full Intention-to-treat analysis (all randomly assigned participants included in the analysis).
- Per-protocol analysis, in which the number of dropouts was given per study arm.
- Per-protocol analysis, in which the number of dropouts was known but was not given per study arm.
- Studies that did not distinguish between the number of randomly assigned participants and the number of evaluated participants. These studies did not refer to dropouts, but the study authors did not define the study explicitly as intention-to-treat.
The distribution of included studies by study type was as follows.
- All-cause fatality (reported in 44 studies).
- Type one: 20 studies (45%).
- Types two and three: 18 studies (41%).
- Type four: six studies (14%).
- Treatment failure (reported In 66 studies).
- Type one: 15 studies (23%).
- Type two: 23 studies (375%).
- Type three: 16 studies (254%).
- Type four: 12 studies (18%).
Protocols were not available for most studies because they were conducted before mandatory trial registry. In general, the primary outcome specified as planned and reported in all trials was "treatment success" ( a variously defined composite outcome including clinical response and the need for antibiotic modifications). All-cause mortality was not specifically defined as an outcome in most trials, and if reported, it was a safety measure reported in the results. Since this was uniformly observed across trials, we have not specified selective reporting for each trial in the Risk of bias tables.
Other potential sources of bias
No other potential sources of bias were identified.
Effects of interventions
Forty-four trials including 5577 participants were included in this comparison ( Analysis 1.1). Thirteen studies including 1431 participants compared the same beta lactam. These studies showed near equivalence (risk ratio (RR) 0.97, 95% confidence interval (CI) 0.73 to 1.30), and studies comparing different beta lactams tended non-significantly in favour of monotherapy (RR 0.85, 95% CI 0.71 to 1.01). No heterogeneity was present for these comparisons (I
The funnel plot analysis for all-cause mortality showed that small studies favouring combination therapy may be missing (Figure 4; Egger's regression two-sided P value 0.05). Mortality outcome was unavailable from 36% of the trials.
|Figure 4. Funnel mortality.|
No significant difference between monotherapy and combination therapy was apparent when analysis was restricted to participants with any Gram-negative infection (eight studies) or Gram-negative bacteraemia (five studies; Analysis 2.1; Analysis 2.2). Only four studies permitted mortality outcome extraction among participants with P aeruginosa infection, and these did not show a difference (total of 60 evaluable participants and 15 deaths; graph not shown). Five UTI studies reported mortality, and mortality was null in three studies. Excluding participants with UTI from the analysis ('non-UTI' subgroup; Analysis 2.3) strengthened the advantage of monotherapy in studies comparing different beta lactams (RR 0.70, 95% CI 0.52 to 0.95).
Three studies addressing Gram-positive infection reported on mortality, all comparing the same antibiotics, with a small sample size and few deaths ( Analysis 2.4). No difference was observed between monotherapy and combination therapy, with the point estimate in the direction favouring monotherapy (RR 0.44, 95% CI 0.12 to 1.58).
Low-risk allocation concealment and generation were associated with risk ratios closer to one than studies with unclear methods for studies comparing different beta lactams but without a statistically significant difference between subgroups ( Analysis 3.1; Analysis 3.2). Combination therapy was significantly better among studies comparing different beta lactams classified as unclear allocation concealment. Blinding was performed in too few studies to assess its effect on mortality. The combined RR for studies comparing the same beta lactam and reporting mortality by intention-to-treat was 0.57 (95% CI 0.28 to 1.19) compared with 1.09 (95% CI 0.80 to 1.51) for studies reporting mortality per-protocol ( Analysis 3.3; P value 0.11 for subgroup difference). Comparison of intention-to-treat versus per-protocol studies for different beta lactams did not reveal a difference. Reanalysis of the mortality comparison by the random-effects model was very similar for same beta lactams (RR 0.99, 95% CI 0.74 to 1.33) and for different beta lactams (RR 0.85, 95% CI 0.69 to 1.05).
We included 66 trials in the clinical failure analysis, comprising 6803 participants ( Analysis 1.3). We detected no difference between monotherapy and combination therapy among studies comparing the same beta lactam (RR 1.11, 95% CI 0.95 to 1.29). We found a significant advantage of monotherapy among studies comparing different beta lactams (RR 0.75, 95% CI 0.67 to 0.84). No heterogeneity was present (I
The funnel plot for treatment failure generated a nearly symmetrical 'funnel distribution' (Figure 5).
|Figure 5. Funnel failure.|
We analysed 28 studies including 1835 participants with Gram-negative infection and 18 studies including 426 participants with P aeruginosa infection ( Analysis 2.5; Analysis 2.6). We observed no significant differences between the study groups for studies comparing the same or different beta lactams. For studies comparing the same beta lactam, the RR was 1.23 (95% CI 0.90 to 1.68) for Gram-negative infection and 1.02 (95% CI 0.68 to 1.51) for P aeruginosa infection. We observed no difference between study groups among participants with Gram-negative bacteraemia or any bacteraemia ( Analysis 2.7; Analysis 2.8). The latter comparison mainly comprised participants with Gram-negative bacteraemia but was available from a larger number of studies and showed a large advantage of monotherapy among studies comparing different beta lactams. Both subgroups of participants with UTIs ( Analysis 2.9) and participants without UTIs maintained the trends seen previously ( Analysis 2.10).
All five studies targeting Gram-positive infection reported on clinical failure. All compared the same beta lactam. The combined risk ratio for clinical failure was 0.69 (95% CI 0.40 to 1.19, five studies, 305 participants; Analysis 2.11). Measures of treatment failure in these studies included persistence of bacteraemia or signs of endocarditis, relapse, need for valve replacement, need for surgery in endocarditis and death. The time of outcome determination was predefined in all trials and follow-up was long (one to six months). The need for surgery in endocarditis was reported in all four trials including participants with endocarditis, with no statistically significant difference noted between treatment groups ( Analysis 2.12).
The quality of allocation concealment and generation did not affect the risk ratios for treatment failure among studies comparing the same or different beta lactams ( Analysis 3.4; Analysis 3.5). All studies at high risk for bias (quasi-randomized) compared different beta lactams, and their results were highly heterogenous.
Several studies comparing different beta lactams used some type of blinding. The advantage of monotherapy was non-significantly larger among these studies compared with non-blinded studies ( Analysis 3.6; P value 0.05 for subgroup difference).
Among studies comparing the same beta lactam, we observed an advantage of combination therapy in the presumed intention-to-treat analysis (type two studies), in which we imputed failure for dropouts. Among studies comparing different beta lactams, intention-to-treat, presumed intention-to-treat and per-protocol results were similar, favouring monotherapy ( Analysis 3.7). Analysis by the random-effects model did not change the results (RR 1.09, 95% CI 0.94 to 1.27 for same beta lactams; RR 0.76, 95% CI 0.67 to 0.87 for different beta lactams).
Bacteriological cure occurred more frequently with monotherapy among studies comparing different beta lactams (RR 0.81, 95% CI 0.69 to 0.94) but did not differ significantly in studies comparing the same beta lactam ( Analysis 1.6).
In an analysis restricted to the studies assessing Gram-positive infection, no difference in microbiological failure rates was reported ( Analysis 2.13),
Length of hospital stay
Eleven studies published data for the comparison of hospital stay. Significant heterogeneity precluded their combination. Duration of hospitalization was longer with monotherapy in three studies (McCormick 1997; Figueroa-Damian 1996; McCormick 1997; 331 participants), shorter in four studies (Arich 1987; Biglino 1991; Damas 2006; Sexton 1998; 186 participants) and similar in four studies (García Ramírez 1999; Mouton 1990; Wing 1998; Yellin 1993; 540 participants).
Development of resistance and adverse events
We merged studies comparing same and different beta lactams for assessment of development of resistance and adverse events. These outcomes are intended to assess the antibiotic class effect of aminoglycoside-beta lactam combinations versus beta lactams alone, whether same or different.
Bacterial superinfections occurred more frequently with combination therapy (RR 0.75, 95% CI 0.57 to 0.99, 28 studies, 3135 participants; Analysis 1.7). We detected no significant difference in the rates of fungal superinfection ( Analysis 1.8). Bacterial colonization was non-significantly more common with combination therapy in all studies reporting on colonization ( Analysis 1.9) and in studies in which surveillance cultures were performed routinely ( Analysis 1.10). Few studies monitored development of resistance among pathogens isolated initially ( Analysis 1.11). We observed no significant difference between monotherapy and combination therapy.
Any adverse event occurred non-significantly more frequently with combination therapy (RR 0.92, 95% CI 0.83 to 1.01; Analysis 1.12), and this analysis was slightly heterogenous. No significant difference was reported with regard to adverse events requiring treatment discontinuation, but these were reported in a minority of studies ( Analysis 1.13). We found nephrotoxicity to be more common in the combination arm in nearly all studies, with a highly significant combined risk ratio in favour of monotherapy (RR 0.30, 95% CI 0.23 to 0.39 Analysis 1.14). A significantly increased rate of nephrotoxicity was seen in studies administering the aminoglycoside once daily and in studies with a multiple-day regimen. Vestibular symptoms and ototoxicity, other known serious side effects of aminoglycoside treatment, were not reported routinely and could not be analysed. Different definitions and detailing of specific adverse events precluded a meaningful meta-analysis of other adverse events, individually or grouped.
Summary of main results
The present review compares beta lactam-aminoglycoside antibiotic combinations versus beta lactam monotherapy among non-neutropenic participants with sepsis. The primary outcome that we assessed was all-cause mortality. Twenty-two of the 69 included studies used the same beta lactam in both study arms. Most studies compared one beta lactam versus a different, narrower-spectrum beta lactam combined with an aminoglycoside. Special emphasis should be placed on studies comparing the same beta lactam. These studies directly test the hypothesis that the addition of an aminoglycoside to the beta lactam is beneficial. Among these studies, all-cause mortality did not differ between study arms (RR 0.97, 95% CI 0.73 to 1.30). Treatment failure occurred more frequently in the monotherapy arm, reaching statistical significance only among the group of 'sepsis' studies. In studies comparing different beta lactams, both failure and mortality were more common in the combination treatment arm. Failure was highly significant, and mortality reached significance only with subgroup analyses. These studies demonstrate an advantage of broad-spectrum beta lactam monotherapy when compared with a narrower-spectrum beta lactam combined with an aminoglycoside, despite equal in vitro coverage of the culprit pathogens in both arms.
Development of resistance was assessed by the occurrence of superinfection and colonization, assuming that bacteria appearing under antibiotic treatment are resistant to the antibiotic administered. Bacterial superinfection occurred significantly more frequently with combination therapy (RR 0.75, 95% CI 0.57 to 0.99). Adverse events occurred more frequently with combination therapy. Specifically, nephrotoxicity occurred significantly more frequently in the combination treatment arm (RR 0.30, 95% CI 0.23 to 0.39). The major adverse event associated with combination therapy was nephrotoxicity. During the past decade, once-daily administration of aminoglycosides has come into use, with similar efficacy but lower nephrotoxicity (Barza 1996). Most studies in our review used multiple-day administration schedules for the complete duration of antibiotic therapy or until modification. The RR of 0.30 for any nephrotoxicity that we observed may, therefore, be an overestimation. However, the RR among the few studies that did administer the aminoglycoside once daily was also highly significant in favour of monotherapy (RR 0.17, 95% CI 0.06 to 0.53).
A small subset of studies in our review addressed participants with Gram-positive infection, mainly S aureus endocarditis. No study assessed enterococcal infection specifically. In these, also, no outcome was improved by the addition of an aminoglycoside.
Overall completeness and applicability of evidence
We defined all-cause mortality as the primary outcome, and most studies assessed and reported treatment failure as a main outcome. Obviously, the most significant outcome for the patient is survival following the infectious episode, and this is the ultimate goal in the treatment of sepsis. Available evidence shows that the addition of an aminoglycoside to a beta lactam does not reduce mortality. Replacing beta lactam monotherapy with a narrower-spectrum beta lactam combined with an aminoglycoside may be associated with increased mortality. Failure was commonly defined as lack of clinical improvement, deterioration, relapse and/or modifications to the antibiotic treatment. These endpoints are highly subjective and do not necessarily translate to detriments experienced by the patient. Detection bias is a concern in open trials that compared the same beta lactam and in trials comparing a 'new' broad-spectrum monotherapy versus a conventional antibiotic regimen. Thus, the advantage of monotherapy in studies comparing different beta lactams, and the opposing advantage of combination therapy in studies comparing the same beta lactam, may be largely biased. Failure was poorly correlated with mortality, despite the fact that the clinical failure definition most commonly included infection-related deaths. In 42 trials reporting both deaths and failures, the Pearson correlation coefficient was 0.36 (RR of 1.0 was assumed for studies with no events in both groups).
The rationale for administering combination therapy arose from in vitro studies showing synergistic bactericidal activity of specific beta lactam-aminoglycoside antibiotic combinations. Synergy has been observed for P aeruginosa (Giamarellou 1984), other Gram-negative bacteria (Giamarellou 1986; Klastersky 1976) and staphylococci (Sande 1975; Sande 1976). Assessment of antibiotic efficacy against specific infections in randomized trials must be limited to definitive treatment (randomization performed when infection is microbiologically documented) or must be performed as a subgroup analysis to assess empirical treatment (randomly assigning participants empirically and assessing those with documented infection). Eight studies assessed definitive treatment (semi-empirical studies), and most assessed empirical treatment. We did not find an advantage of combination therapy among participants with any Gram-negative infection, Gram-negative bacteraemia or P aeruginosa infection. Lack of data precluded the assessment of P aeruginosa bacteraemia. Why does synergy, observed in vitro, not translate into clinical benefit? Specific growth conditions in vitro, unattainable in vivo, may induce synergism. Pharmacokinetic and pharmacodynamic properties involving specific antibiotics, sites of infection, timing and intervals of administration may prevent synergism in vivo. Adverse events related directly to the aminoglycoside, or to the combination, may interfere with an in vivo benefit, amounting altogether to no benefit.
Combination therapy in endocarditis similarly relies on in vitro and in vivo data. Animal studies have shown that sterilization of cardiac vegetations may be achieved more rapidly with combination therapy (Sande 1975; Sande 1976). One clinical study included in our review showed that combination therapy shortened the duration of bacteraemia, but this comparison was performed according to the empirical antibiotic regimen, while randomization occurred empirically or semi-empirically (Korzeniowski 1982). We could not show an advantage of combination therapy by combining all trials in humans. On the contrary, all outcomes tended to favour monotherapy, although statistical significance was not reached.
Quality of the evidence
Among studies comparing the same beta lactam, the quality of evidence of mortality was graded as low, mainly because of imprecision. The 95% confidence intervals range from 27% improved survival to 30% higher risk of death with monotherapy. The quality of evidence for failure was graded as very low because of the indirectness of the outcome and the risk of detection bias associated with assessment of a subjective outcome in open trials (Wood 2008).
In studies using different beta lactams, the quality of evidence for mortality was low because the advantage of monotherapy was derived from studies at unclear risk of bias in relation to allocation concealment and due to suspected publication or reporting bias. The advantage of monotherapy with regard to treatment failure was graded as very low quality of evidence, again because of the indirectness of the failure outcome and the high risk of bias in non-blinded trials.
Potential biases in the review process
A major limitation of existing studies and thus of the compiled analysis is the lack of data for all-cause mortality from more than a third of included studies. This was probably not due to selective reporting bias in that all-cause mortality was not defined as an outcome in included studies. However, the funnel plot for mortality was asymmetrical. Data for subgroups most likely to benefit from combination therapy were also not available from all studies. In our analysis, we did not correct for the appropriateness of antibiotic treatment, which has been shown conclusively to correlate with survival (Ibrahim 2000; Leibovici 1998). Data were not fully available to perform such an analysis. However, among studies comparing the same beta lactam, combination therapy by definition broadened the spectrum of coverage without improving outcomes. In studies comparing different beta lactams, an inappropriate beta lactam was used more frequently in the combination arm, which may partially explain the advantage of monotherapy.
Agreements and disagreements with other studies or reviews
Observational studies tend to show an advantage of combination therapy for severe infection caused by Gram-negative bacteria or P aeruginosa. Combination therapy was claimed to be superior to monotherapy in a prospective observational study of participants with P aeruginosa bacteraemia, but most participants in the monotherapy group received an aminoglycoside (Hilf 1989). Kumar et al. conducted a large multi-centre retrospective study, including 4662 critically ill participants in the intensive care unit (ICU) with culture-positive, bacterial septic shock. In a propensity-matched analysis (1223 propensity-matched pairs), combination therapy was associated with lower mortality than monotherapy, with an overall hazard ratio for 28-day all-cause mortality of 0.77 (95% CI 0.67 to 0.88; Kumar 2010b). This included all infections (Gram-negative and Gram-positive), and any antibiotics of any class could be included in the combination and monotherapy arms. An analysis restricted to beta lactam-aminoglycosides as combination therapy showed an advantage of combination therapy, but an analysis of beta lactam-aminoglycoside versus beta lactam alone (same or different) is not presented. Bliziotis et al. compared combination therapy versus monotherapy for P aeruginosa bacteraemia in a retrospective cohort study and found no significant difference between the regimens, although both mortality and treatment failure were more common with monotherapy (Bliziotis 2011). In contrast, in a prospective study of bacteraemic participants with Gram-negative bacteraemia, we found no significant difference with regard to in-hospital mortality between appropriate beta lactam monotherapy and appropriate beta lactam aminoglycoside combination therapy, both empirically and semi-empirically. Appropriate single aminoglycoside monotherapy was associated with increased mortality (Leibovici 1997). Participants included in observational studies are different from those included in randomized trials. It is possible that an effect that was not observed in randomized controlled trials exists. However, observational studies to date do not provide clear enough conclusions.
Implications for practice
We conclude that the addition of an aminoglycoside to a beta lactam does not improve the clinical efficacy achieved with the beta lactam alone. Substituting a narrow-spectrum beta lactam with an aminoglycoside for a single broad-spectrum beta lactam will result in increased failure rates and may be associated with increased mortality. Adverse events occur more frequently with combination treatment. Short-term combination therapy for sepsis does not prevent the development of resistant bacteria, as assessed by superinfection or colonization rates following antibiotic treatment. Thus, the use of beta lactam-aminoglycoside combination therapy for sepsis should be discouraged.
Clinicians usually face the dilemma of selecting an antibiotic treatment on two occasions during an uncomplicated infectious episode. On the initial encounter with a patient, the clinician must prescribe empirical antibiotic treatment because the causative pathogen and its susceptibilities are generally unknown. Most studies addressed this situation, and the results show no difference in overall mortality whether monotherapy or combination therapy is used. Adverse effects, most significantly nephrotoxicity, will occur more frequently with combination therapy. If the choice is between a narrower-spectrum beta lactam combined with an aminoglycoside versus a broad-spectrum beta lactam, our results show that treatment will ultimately have to be modified more frequently if the combination is chosen. We have not identified a specific site of infection or level of disease severity for which combination treatment provides an advantage.
The second decision point occurs when the causative pathogen is identified. Here, the choice of antibiotic treatment is dictated by known susceptibility results. However, the question remains whether for specific bacteria, beta lactam-aminoglycoside combination treatment offers an advantage over single beta lactam treatment. We addressed this question through subgroup analyses of participants with documented infection caused by specific pathogens (Gram-negative pathogens, P aeruginosa, S aureus). In addition, several semi-empirical studies addressed this question specifically. We have not identified a specific pathogen, or pathogen group, for which combination therapy is advantageous. However, data for these subgroups are very limited.
Overall, appropriate beta lactam monotherapy should be used. Beta lactam-aminoglycoside combination therapy does not offer an advantage and is associated with an increased rate of adverse events.
Implications for research
Innovative trial designs are needed to allow the assessment of participants with severe Gram-negative infection and P aeruginosa bacteraemia (Paul 2009). Similarly, the question is still open for endocarditis caused by Gram-positive bacteria, including mainly S aureus and Enterococcus sp. (Leibovici 2010). The pragmatic randomized trial design using electronic health records might serve as a solution for identification and recruitment of a necessary sample size in multi-centred trials (Staa 2012).
Future trials should differentiate between empirical and definitive antibiotic treatment. Appropriate antibiotic treatment has been shown to significantly reduce mortality and should therefore be reported, with results adjusted. Outcomes relevant to patients, such as survival and duration of hospitalization, should be assessed. Survival, if not assessed as a primary outcome, must at least be reported as a safety measure in all clinical trials.
We would like to thank all the authors who responded to our requests for additional data (see 'unpublished data' and 'unpublished data sought but not used', 'References to studies'). Dr Solomkin (Solomkin 1986) and Dr Sexton (Sexton 1984) supplied supplementary data for their studies, which were not included in the review. Dr Finer and Dr Goustas of the GlaxoSmithKline Company supplied detailed data for their study (Finer 1992). Dr Kora Huber sent completed trial results for Kljucar 1990 and supplied requested additional information. Ms Mary Forrest (Managing Editor, Journal of Chemotherapy) sent several publications that were not available to us. We would also like to warmly thank Ms Rika Fujiya, who translated the Japanese studies (Sukoh 1994; Takamoto 1994).
We thank Dr Vittoria Lutje, Dr Harriet G. MacLehose and Ms Rieve Robb (Managing Editor) of the Cochrane Infectious Diseases Group. We thank Dr Harald Herkner, Prof Nathan Pace, Kathie Godfrey, Janet Wale and Jane Cracknell (Managing Editor) of the Cochrane Anaesthesia Review Group. Both groups supported and provided helpful revisions for this review.
This review was initially developed within the Infectious Diseases Group and was supported by a grant from the Department for International Development, UK. The review was transferred to the Anaesthesia Group in May 2005.
Ishay Silbiger participated in the first version of this review: applied inclusion and exclusion criteria and performed risk of bias assessment, data extraction and analysis.
Karla Soares-Weiser participated in the first version of this review: assisted with inclusion and exclusion of studies; performed quality assessment, data extraction and analysis; and assisted with the writing and reviewed all versions of the protocol and the review. We thank Karla for her mentorship on systematic reviews and for guidance provided on the initial protocol and on initiation of this review.
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. Search strategy for CENTRAL, T he Cochrane Library
#1 MeSH descriptor beta-Lactams, this term only
#2 MeSH descriptor Penicillins, this term only
#3 MeSH descriptor Cephalosporins, this term only
#4 MeSH descriptor Carbapenems, this term only
#5 MeSH descriptor Imipenem, this term only
#6 MeSH descriptor Ceftazidime, this term only
#7 MeSH descriptor Cefotaxime, this term only
#8 MeSH descriptor Amoxicillin-Potassium Clavulanate Combination, this term only
#10 co?amoxiclav* or cephalosporin* or ceftazidim* or cefotaxim* or carbapenem* or imipenem* or meropenem*
#11 beta-lactam* near (combin* or mono*)
#12 (#1 OR #2 OR #3 OR #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11)
#13 MeSH descriptor Aminoglycosides, this term only
#14 (aminoglycoside* or netilmicin* or gentamicin* or amikacin* or tobramycin* or streptomycin* or isepamicin* or sisomicin* or combinat*):ti,ab
#15 (#13 OR #14)
#16 (#12 OR ( #12 AND #15 ))
#17 MeSH descriptor Pneumonia, this term only
#18 MeSH descriptor Sepsis, this term only
#19 MeSH descriptor Shock, Septic, this term only
#20 MeSH descriptor Bacteremia explode all trees
#21 MeSH descriptor Infection, this term only
#22 MeSH descriptor Endocarditis, Bacterial explode all trees
#23 MeSH descriptor Endocarditis, this term only
#24 MeSH descriptor Staphylococcus, this term only
#25 MeSH descriptor Streptococcus, this term only
#26 (pneumonia* or infect* or sepsis* or septic* or bacter*):ti,ab
#27 (#17 OR #18 OR #19 OR #20 OR #21 OR #22 OR #23 OR #24 OR #25 OR #26)
#28 (#16 AND #27)
Appendix 2. Search strategy for MEDLINE (Ovid SP)
1. Beta-Lactams/ or Penicillins/ or Cephalosporins/ or Carbapenems/ or Imipenem/ or Ceftazidime/ or Cefotaxime/ or Amoxicillin-Potassium Clavulanate Combination/ or beta-lactam*.ti,ab. or (co?amoxiclav* or cephalosporin* or ceftazidim* or cefotaxim* or carbapenem* or imipenem* or meropenem*).mp. or (beta-lactam* adj5 (combin* or mono*)).mp.
2. Aminoglycosides/ or (aminoglycoside* or netilmicin* or gentamicin* or amikacin* or tobramycin* or streptomycin* or isepamicin* or sisomicin* or combinat*).ti,ab. or Drug-Therapy-Combination/
3. 1 or (1 and 2)
4. pneumonia/ or exp Sepsis/ or Shock, Septic/ or exp Bacteremia/ or Infection/ or exp Endocarditis, Bacterial/ or Endocarditis/ or exp Endocarditis, Subacute Bacterial/ or Staphylococcus/ or Streptococcus/ or (pneumonia* or infect* or sepsis* or septic* or bacter*).ti,ab.
5. 3 and 4
6. ((randomized controlled trial or controlled clinical trial).pt. or randomized.ab. or placebo.ab. or clinical trials as topic.sh. or randomly.ab. or trial.ti.) not (animals not (humans and animals)).sh.
7. 6 and 5
Appendix 3. Search strategy for EMBASE (Ovid SP)
1. beta lactam/ or penicillin derivative/ or cephalosporin derivative/ or carbapenem derivative/ or imipenem/ or ceftazidime/ or cefotaxime/ or amoxicillin plus clavulanic acid/ or beta-lactam*.ti,ab. or (co?amoxiclav* or cephalosporin* or ceftazidim* or cefotaxim* or carbapenem* or imipenem* or meropenem*).mp. or (beta-lactam* adj5 (combin* or mono*)).mp.
2. aminoglycoside/ or (aminoglycoside* or netilmicin* or gentamicin* or amikacin* or tobramycin* or streptomycin* or isepamicin* or sisomicin* or combinat*).ti,ab. or drug combination/
3. 1 or (1 and 2)
4. pneumonia/ or exp sepsis/ or septicemia/ or bacteremia/ or infection/ or exp bacterial endocarditis/ or endocarditis/ or Staphylococcus/ or Staphylococcus/ or (pneumonia* or infect* or sepsis* or septic* or bacter*).ti,ab.
5. 4 and 3
6. (placebo.sh. or controlled study.ab. or random*.ti,ab. or trial*.ti,ab.) not (animals not (humans and animals)).sh.
7. 5 and 6
Appendix 4. Search strategy for LILACS (via BIREME)
(co?amoxiclav$ or cephalosporin$ or ceftazidim$ or cefotaxim$ or carbapenem$ or imipenem$ or meropenem$ or beta-lactam$) and (aminoglycoside$ or netilmicin$ or gentamicin$ or amikacin$ or tobramycin$ or streptomycin$ or isepamicin$ or sisomicin$ or combinat$ or pneumonia$ or infect$ or sepsis$ or septic$ or bacter$ or endocardit$ or Staphylococ$ or Streptococ$) and (trial$ or Random$ or placebo$ or ((single or double or triple) and (blind$ or mask$)) or (ensayo controlado) or (experimentação controlada) or multicenter or multicentre or prospective or (estudio anticipado) or (estudo em perspectiva))
Obtaining data on all-cause mortality, 16 June 2013
Thank you for taking on the large amount of data surrounding topic of aminoglycoside and beta-lactam combination therapy in the treatment of sepsis. With the large volume of studies spanning such a long time period this was no small task. With this in mind we still have some one question regarding the primary outcome analysis of all cause mortality between the two treatment arms.
The primary mortality analysis contained 43 of the total 64 studies included in the review and they were split into two separate subgroups, using either the same or a different beta-lactam agent as monotherapy as in combination therapy. Our concern is centered on the outstanding 21 studies not included in this analysis. We were wondering what attempts were made to collect mortality data from these remaining trials questioning whether the inclusion of those results would statistically alter the outcomes. We understand that a number of these studies were completed over 40 years ago and the data may be very difficult to obtain.
In the subgroup where a different beta-lactam was used the risk of mortality was non-significantly lowered in the monotherapy arm RR 0.85 (95% CI 0.71, 1.01). With the results being close to statistical significance we were wondering if the addition of data from the outstanding studies would actually make a statistical difference. If this were truly the case then your conclusion of “The addition of an aminoglycoside to beta-lactams for sepsis should be discouraged. All-cause fatality rates are unchanged. Combination treatment carries a significant risk of nephrotoxicity” would change and the call to avoid the use of these antibiotics would be much stronger.
We also pooled all of the data from both subgroups (same and different beta-lactams) and found that it did not change the results of the different beta-lactam group but greatly narrowed the confidence interval of the same beta-lactam group with a RR of 1.13 (0.97, 1.31) - increased risk of mortality in the combination group versus monotherapy. With this analysis we also found very little heterogeneity between same and different beta-lactam studies (I
We understand the beta-lactam agent selected, specifically in regards the spectrum of activity, greatly impacts the effects of empirical therapy but with this potentially increased risk of mortality that is consistent across this large number of studies leads us to believe that although statistically non-significant it seems plausible that the risk of mortality with combination therapy over beta-lactam monotherapy is real.
We also believe that the possibility of “emotional based medicine” is real in this patient population. As the majority of these studies were open-label despite being randomized, it is not unlikely that “sicker” patients would receive more drugs (i.e. combination therapy). If this were true then it is plausible that patients who were more likely to die received more antibiotics and were in the combination groups but with the effect remaining relatively consistent across this large number of studies, we feel that a true risk may actually exist.
After this long discussion, our question returns to whether or not mortality data are available from the remaining 21 studies and what attempts have been made to retrieve this information. A statistically significant increase in mortality, along with the increase in adverse events see with combination therapy would likely facilitate a rapid change in practice and removal of this therapeutic option. Just as an exercise we inserted the data provided by your review into Review Manager to test how many events it would take to make the difference in mortality. We understand this is not a truly scientific exercise but one based on curiosity.
What we found was that when we added two events (deaths) to the combination group in the most heavily weighted study (Felisart 1985) the outcome of mortality became statistically significantly higher in the combination group. We also combined all of the data between the same different beta-lactam subgroups and found that only eight more deaths in the combination group made the entire analysis (all 43 studies) statistically significant for an increase in mortality in the combination group. On the flip side, it took 80 events in the monotherapy group to swing the analysis the other way and statistically favour combination therapy in the outcome of mortality.
Thank you for your time
Dear Dr Amadio,
Thank you for your kind attention to our work and your input for the data analysis.
In response to your question regarding obtaining data on all-cause mortality, we mailed all authors of trials that did not report on this outcome asking for the data, as is routine in Cochrane reviews. We agree with you on the importance of the missing data on mortality and for this reason we made extra efforts to obtain the data. If we did not establish contact with the corresponding author, we tried to contact a second and third author. The data presented in our review are the result of this process and still we miss mortality data from a third of all randomized controlled trials (RCTs) that were conducted.
Selection bias should not occur in adequately conducted RCTs, those using appropriate allocation concealment. Allocation concealment is the procedure ensuring that no one is aware of the treatment assignment when the patient is recruited into the trial and before the patient is allocated to an intervention. We observed in our review that the advantage to combination therapy was larger in trials with unclear methods for allocation concealment (studies not reporting the methods for this procedure) compared to trials that used methods ensuring adequate allocation concealment. Therefore, it is possible that results were affected by selection of sicker patients to the combination therapy group. However the difference between trials with low and unclear risk of bias was not statistically significant and we have no actual data on whether bias could occur in the trials with unclear risk of bias. Most importantly to our view, the trials comparing different beta-lactams usually compared a new, broad-spectrum beta-lactam to an old, classical regimen; we believe that if selection bias crept in to some trials it would have worked in the opposite direction of recruiting the sicker patients to the novel monotherapy arm. The fact that most of the trials were open, might have led to a different type of bias, and dilution of effects, because physicians could add an aminoglycoside to failing patients in the monotherapy arm, while this could not occur in the combination therapy arm.
Methods exist to formally examine the possible effects of missing data in meta-analysis. We will consider adding such an analysis to an update of our review. More importantly, we will highlight the issue of missing data on all-cause mortality. Should your important correspondence result in any authors sending further data from their trials on mortality, these will be added to our review.
Anthony Amadio, BSc. Pharm, ACPR, RPh
Doctor of Pharmacy Student
Faculty of Pharmaceutical Sciences
University of British Columbia
Aaron M Tejani, BSc Pharm, PharmD
Therapeutics Initiative, University of British Columbia
2176 Health Sciences Mall
Vancouver, BC, Canada
Mical Paul, corresponding author
Last assessed as up-to-date: 4 November 2013.
Protocol first published: Issue 4, 2001
Review first published: Issue 1, 2006
Contributions of authors
Mical Paul (MP): performed the search and scanned abstracts; retrieved full-text articles and applied inclusion and exclusion criteria; performed risk of bias assessment, data extraction and analysis. MP communicated with authors and wrote the protocol and the review.
Adi Lador (AL): performed an update search (2011) and scanned the new abstracts; retrieved full-text articles and applied inclusion and exclusion criteria; performed risk of bias assessment, data extraction and analysis for the updated search; and wrote the updated review.
Simona Grozinsky-Glasberg (SG-G): extracted the data for the first version of this review; and reviewed and approved the final version of the updated review.
Leonard Leibovici (LL): assisted with inclusion and exclusion of studies; performed quality assessment, data extraction and analysis; assisted with communication with authors; and assisted with the writing and review of all versions of the protocol and the review.
Declarations of interest
Mical Paul: none known.
Adi Lador: none known.
Simona Grozinsky-Glasberg: none known.
Leonard Leibovici: none known.
We certify that we have no affiliations with or involvement in any organization or entity with a direct financial interest in the subject matter of this review (e.g. employment, consultancy, stock ownership, honoraria, expert testimony).
Sources of support
- Rabin Medical Center - Beilison Campus, Israel.
- EU 5th Framework - TREAT project (grant number: 1999-11459), Not specified.
- Department for International Development, UK.
Medical Subject Headings (MeSH)
Aminoglycosides [adverse effects; *therapeutic use]; Anti-Bacterial Agents [*therapeutic use]; Bacterial Infections [*drug therapy]; Cause of Death; Drug Therapy, Combination [methods]; Gram-Negative Bacterial Infections [drug therapy]; Gram-Positive Bacterial Infections [drug therapy]; Randomized Controlled Trials as Topic; beta-Lactams [*therapeutic use]
MeSH check words
* Indicates the major publication for the study