Results of the search

The original search strategy resulted in 331 references. After reviewing all abstracts, we retrieved 42 studies for full‐text inspection. We updated the search in 2013 and 3515 new references were screened, but no new trials were identified for inclusion. The current 2017 update revealed 2723 new references with one study identified for inclusion (Bucaneve 2014). See Figure 1.

Figure 1

---

---

Study flow diagram.

Included studies

We included 22 publications representing 14 individual randomised controlled trials (RCTs) corresponding to our inclusion criteria. The trials were conducted between 1979 and 2014.

Glycopeptides were tested in nine trials (vancomycin five, teicoplanin four) (see Characteristics of included studies). These trials were performed between 1984 and 2000. Other anti‐gram‐positive (antiGP) drugs were tested in five trials: cephalothin in two (Lawson 1979; Verhagen 1987), flucloxacillin and trimethoprim‐sulphamethoxazole in one trial each (de Pauw 1985; Menichetti 1986, respectively), and tigecycline in the last trial identified in the 2017 update (Bucaneve 2014). The last two antiGP antibiotics cover also gram‐negative bacteria (Bucaneve 2014; Menichetti 1986). The basic antibiotic regimens are specified in the table 'Characteristics of included studies'. Ceftazidime, the most commonly used beta‐lactam used as a basic regimen, was used alone in five trials and with amikacin in two trials.

The antiGP antibiotic was tested at the onset of antibiotic treatment as the first line (empirical) regimen in all trials but two, which assessed its addition for persistently febrile patients (first modification) after 72 to 96 hours of imipenem monotherapy (Erjavec 2000), or after 48 to 60 hours of piperacillin‐tazobactam monotherapy (Cometta 2003). We did not identify studies assessing pre‐emptive antiGP treatment.

Nine trials randomised 1993 patients. Five studies allowed patient re‐entry for separate neutropenic febrile episodes, thus randomising episodes instead of patients. Three trials included 352 episodes representing 292 patients (Del Favero 1987; Erjavec 2000; Menichetti 1986), and two trials included 437 episodes without specifying the number of patients (Lawson 1979; Marie 1991). Overall, 2782 febrile episodes were included and 2549 were evaluated.

All trials included patients with haematological malignancies, except for one trial that was restricted to patients with solid tumours (Molina 1993). One trial included only patients with haematological malignancies (Bucaneve 2014). Two trials did not specify patients' age. In the remaining trials, children less than 16 years were included in six trials, and the mean age ranged between 38 to 48 years. With regard to exclusion of patients at risk for infections due to gram‐positive bacteria, the two first modification trials excluded patients with documented or suspected catheter‐related infections (Cometta 2003; Erjavec 2000). Two empirical studies excluded patients with a documented focus of infection (Marie 1991; Novakova 1991), of which one also excluded patients in septic shock (Marie 1991). Otherwise, no restrictions related to the criteria suggested for empirical antiGP treatment (see Background) were imposed on patient inclusion.

The rate of single‐agent gram‐positive bacteraemia varied between 6% and 28% (Table 1) and did not correlate with the study year.

Excluded studies

Twenty studies were excluded. Two studies were excluded on account of a high percentage of dropouts. An EORTC trial randomised 841 patients and evaluated 419 patients (EORTC 1983). Martino and colleagues reported outcomes for a 10‐month period and 158 patients of a trial which was conducted for 15 months and included 232 patients (Martino 1992). The reasons for exclusion of the remaining studies are listed in the Characteristics of excluded studies table.

Mortality

Eight studies, including 1242 participants, reported overall mortality. The adjusted mean mortality rate in these studies was 9.0%. The risk ratio (RR) for death was 0.9 (95% confidence interval (CI) 0.64 to 1.25; 8 studies, 1242 participants; Analysis 1.1), values lower than 1 favouring the antiGP arm. The quality of this evidence was rated as moderate, downgraded for imprecision (summary of findings Table for the main comparison). Two trials used a glycopeptide empirically, two used a glycopeptide semi‐empirically, and four used another antiGP antibiotic empirically. No difference in mortality was seen in each of these groups. Considering only studies with low‐risk allocation concealment, or those reporting mortality by ITT results, the results were similar, below 1 (Analysis 1.2; Analysis 1.3). Overall, no heterogeneity was seen with this comparison, which was performed using the fixed‐effect model (I² = 0%).

Data regarding mortality among patient subgroups were scarce. Only five studies were included in the comparison for patients in whom a gram‐positive infection was documented (Analysis 1.4). Only 13 deaths were recorded; hence although overall mortality among patients receiving antiGP treatment was almost twice that in the control group, this was not interpreted into a difference. The rate of single‐agent gram‐positive bacteraemia (Table 1) was reported only for five of the studies in the mortality analysis and no association was observed between this rate and the RRs for mortality, but the analysis was limited by the paucity of data with extreme 95% CI for the ratio of odds ratios (ORs). Excluding two studies in which the antiGP was active also against gram‐negative bacteria did not change results (Analysis 1.5). Data for the other pre‐defined subgroup analyses were not available. Inspecting the funnel plot did not reveal a small‐study effect (Figure 4).

Figure 4

---

---

Failure.

Eight trials compared infection‐related fatality, two of which did not report overall mortality. The RR was 1.15 (95% CI 0.76 to 1.75; 8 studies, 1810 participants; Analysis 1.6).

Treatment failure

Overall failure, disregarding treatment modifications, was assessed in seven studies including 943 participants and was similar in both study arms (RR 1.00, 95% CI 0.79 to 1.27; 7 studies, 943 participants; Analysis 2.1, low‐quality evidence). When modifications were counted as causes for treatment failure, an advantage in favour of antiGP treatment was evident (RR 0.72, 95% CI 0.65 to 0.79, 11 studies, 2169 participants; Analysis 2.2, very low‐quality evidence). The quality of the evidence for failure was downgraded for lack of blinding for both outcomes and indirectness of the outcome for treatment failure with modifications (summary of findings Table 2). The advantage originated from studies that assessed the initial empirical administration of antiGP antibiotics and was demonstrated both for empirical glycopeptides and other antiGP agents. No benefit was observed for the addition of glycopeptides for persistent fever (first modification), in two double‐blind studies (Cometta 2003; Erjavec 2000; Analysis 2.2). Results were not affected by randomisation methods, with similar direction of effects when the analysis was limited to studies with adequate allocation concealment (Analysis 2.3) or those permitting analysis by ITT(Analysis 2.4). Excluding the double‐blind trials decreased the risk ratio (RR 0.67, 95% CI 0.60 to 0.75), demonstrating the effect of the open design on treatment modifications. The funnel plot for failure was centred approximately symmetrically around the effect estimate (Figure 5).

Figure 5

---

---

Mortality.

In subgroup analyses of failure with modifications, when excluding antiGP antibiotics covering gram‐negative bacteria, there was no longer an advantage to non‐glycopeptide empirical antiGP antibiotics (Analysis 2.5). An analysis restricted to patients who ultimately had a gram‐positive infection, was composed of studies assessing empirical antiGP treatment and demonstrated a large advantage to this intervention, (RR 0.56, 95% CI 0.38 to 0.84; 5 studies, 175 patients; Analysis 2.6). Meta‐regression demonstrated no association between the rate of single gram‐positive bacteraemia in the studies and their risk ratio for failure with antiGP treatment. No data were available for the subgroup of patients with central catheters or those having received quinolone prophylaxis.

Duration of fever was not reported comparatively, but three studies compared the number of persistently febrile patients at 72 hours after the initiation of empirical antibiotic treatment (Analysis 2.7). An advantage to the antiGP arm was seen, but the number of patients evaluated was small (312 patients). Amphotericin was added more frequently to the control arm (RR 1.23, 95% CI 0.84 to 1.80, 5 studies, 1201 participants, Analysis 2.8). Substantial heterogeneity was seen in the comparisons of persistent fever and addition of amphotericin, which were analysed using the random‐effects model.

Superinfections and adverse events

AntiGP treatment did not increase superinfection rates (Analysis 3.1). Focusing on bacterial superinfections, we observed a decrease with antiGPs (RR 0.41, 95% CI 0.27 to 0.6, Analysis 3.2); specifically gram‐positive superinfections (RR 0.24, 95% CI 0.14 to 0.4, Analysis 3.3). The rate of fungal superinfections was similar in both arms (RR 1.04, 95% CI 0.60 to 1.82, Analysis 3.4). No study assessed the effect of additional antiGP treatment on the rate of colonisation with resistant microorganisms or development of resistance.

Adverse events were more frequent in the antiGP arm (Analysis 4.1) (summary of findings Table 3) with very low quality of evidence. However, this originated mainly from a difference in skin reactions (Analysis 4.2) rather than in adverse events incurring significant morbidity. Nephrotoxicity did not differ between the study groups (Analysis 4.3).

Other outcomes

Duration of hospital stay was inconsistently reported and summarised heterogeneously, as means or medians without appropriate CIs, in the included trials. Thus results could not be combined. Duration of fever and removal of central catheters were not reported in the studies.