Introduction

Febrile neutropenia (FN) in patients with cancer is recognized as a common, potentially fatal complication of cytotoxic chemotherapy [7]. Hospitalization and prompt initiation of empiric broad-spectrum intravenous antibiotic therapy has dramatically reduced infection-related morbidity and mortality over the past 40 years [37, 40]. The intensity of this approach, however, is not without undesirable sequelae. Prolonged antibiotic exposure and hospitalization are potent risk factors for the emergence of resistant microorganisms [14] and secondary infection [20]. Inpatient therapy is also associated with inferior health-related quality of life for children with cancer [42]. Finally, hospitalization is significantly less cost-effective when compared to outpatient management of FN [44].

Children with FN are a heterogeneous group [38], and a subset at low-risk of adverse events might benefit from less intensive therapy [3]. Numerous prediction rules developed for risk stratification in pediatrics [36] have made it possible for the use of less intensive strategies in the management of low-risk patients. Two less-intensive strategies that have been employed in many studies are outpatient management and oral antibiotic administration. Outpatient management, using oral or parenteral antibiotics, can also include step-down management involving a short hospital admission followed by early discharge with home antibiotic therapy. Oral antibiotic regimens, in the inpatient or outpatient setting, can also include step-down management involving intravenous therapy for a short period followed by oral antibiotics.

It is generally accepted that adult inpatients with low-risk FN can be managed effectively with oral antibiotics [15, 22, 46], and there is increasing evidence that outpatient management of this population is an equally safe alternative [10]. As a result, risk-adapted international treatment guidelines have been established in adult oncology [16, 23]. However, despite several prospective studies addressing the question, there is no consensus on whether therapeutic intensity can be reduced safely in children with low-risk FN [4].

We recently conducted a systematic review of inpatient versus outpatient and oral versus parenteral antibiotic management of FN by analyzing randomized controlled trials (RCTs) and found no differences in efficacy or safety [45]. However, only two studies of outpatient management with 278 subjects were conducted in children, and one of these consisted of high-risk patients. Thus, estimates were highly unreliable and poorly generalizable to children with low-risk FN. We realized that more information was available from prospective single-arm studies which could supplement data from randomized trials, possibly resulting in more complete and robust information about the efficacy and safety of these strategies in pediatrics.

Our primary objective was to determine if there is a difference in efficacy between initial or step-down outpatient management and inpatient management of children with low-risk FN. Our secondary objective was to determine if there is a difference in efficacy between initial or step-down oral antibiotic therapy and parenteral therapy in children with low-risk FN.

Materials and methods

Data sources and searches

We developed a protocol for the review following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting observational studies [47]. We performed electronic searches of OVID Medline (1980 to March 7, 2011); EMBASE (1980 to March 7, 2011); and the Cochrane Central Register of Controlled Trials (until the first quarter of 2011). The search strategy (Appendix 1) included the Medical Subject Headings (MeSH) and text words “fever” and “neutropenia”, and was limited to studies conducted after 1980 and those published in the English language. All trial designs were included in the search.

Study selection

We defined inclusion and exclusion criteria a priori. Studies were included if: (1) the study examined any infection outcome of a homogeneous initial empiric regimen, (2) the population consisted of children or results were abstractable for the pediatric sub-group, and (3) the study was conducted prospectively (to avoid bias associated with retrospective studies). Exclusion criteria were: (1) conference proceeding only, (2) not published in English, (3) not a study, (4) retrospective, (5) population did not consist of children or data not abstractable for children, (6) cohort did not consist of patients with initial presentation of FN (i.e., enrolled ≥24 h after initial empiric treatment), (7) antibiotics studied were not initial empiric therapy, (8) heterogeneous empiric therapy regimens, (9) pharmacokinetic studies, (10) no infection outcomes reported, and (11) duplicate publications. Among this set of studies, those of low-risk patients only, as defined by each study protocol, were then selected.

One reviewer (LS) evaluated the titles and abstracts of publications identified by the search strategy, and any publication thought to be potentially relevant was retrieved in full. Two independent reviewers (AM and LS) then assessed full publications for eligibility; reviewers were not blinded to study authors or outcomes. Final inclusion of studies into the meta-analysis was by agreement of both reviewers. Agreement between reviewers was evaluated using a kappa statistic. Strength of agreement as evaluated by the kappa statistic was defined as slight (0.00–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), or almost perfect (0.81–1.00) [25].

Data extraction and quality assessment

Data extraction from included trials was performed independently by two reviewers (AM and LS) using a standardized data collection form.

The primary outcome measure for both objectives was treatment failure at 30 days when antibiotic modification was included as a criterion for failure. Although the trials used heterogeneous definitions of treatment failure, most definitions included persistence, recurrence or worsening of fever/infecting organisms, new infections, any modification of antibiotics, readmission, or death during study drug treatment. Secondary outcome measures were 30-day overall mortality and infection-related mortality, treatment failure when antibiotic modification was excluded from the failure definition, fever duration, recurrent infection (reappearance of infection or fever after initial resolution), sepsis, secondary infection (occurrence of new infection during treatment), adverse events leading to antibiotic discontinuation, and readmission to hospital. An outcome was presented only if each sub-group contained at least one study reporting that outcome.

Study quality was assessed using a modified version of an instrument previously developed to describe quality in studies of prognosis [18]. This quality assessment instrument examines four potential sources of bias: study participation, study attrition, confounding variables, and measurement of outcomes. Each element was rated as having low, medium, or high risk of bias for each study.

Data synthesis and analysis

This meta-analysis combined data at the study level and not at the individual patient level. All outcomes were described as percentages, with the exception of duration of fever which was described using the mean. For this outcome, we made the following assumptions to facilitate data synthesis: the mean can be approximated by the median and the range contains six standard deviations. Each study was weighted by the inverse variance. Given the anticipation of heterogeneity between studies, a random effects model [12] was used for all analyses. Because our outcomes were single percentages and not within-study comparisons, we did not test for publication bias as we did not believe it was relevant in this context.

The meta-analysis was performed using Review Manager (RevMan) (Version 5.1.0, The Cochrane Collaboration, Oxford, England). Agreement was calculated using the SAS statistical program (SAS-PC, version 9.1; SAS Institute Inc., Cary, NC). Sub-groups were defined based on treatment setting (outpatient or inpatient) and route of administration (oral or parenteral). A test for heterogeneity across subgroup results was used to determine if outcomes were modified based upon treatment setting or route of administration, with statistical significance defined as P < 0.10.

Results

Figure 1 illustrates the flow diagram of trial identification and selection. A total of 7,281 titles and abstracts were reviewed, and 380 full articles were retrieved. Of these, 66 satisfied pre-defined inclusion criteria. Reasons for excluding 314 articles are provided in Fig. 1. The reviewers had almost perfect agreement on articles for inclusion (kappa = 0.98; 95 % CI 0.96, 1.00).

Fig. 1
figure 1

Flow diagram of trial identification and selection

Full size image

Of the 66 studies, 16 studies were restricted to low-risk patients [1, 5, 6, 9, 13, 17, 21, 24, 26, 28, 29, 31–34, 41] and were thus included in this analysis. For these 16 studies, the number of studies that demonstrated low risk of bias was as follows: 13 (81 %) for study participation, 14 (88 %) for study attrition, 2 (13 %) for confounding, and 7 (44 %) for measurement of outcomes.

Clinical characteristics of the 16 studies are presented in Table 1. Eight were RCTs and eight were prospective non-randomized studies. There were no RCTs that randomized the trial setting (outpatient versus inpatient). Five out of eight RCTs randomized oral versus parenteral antibiotics. The 16 studies described 24 treatment regimens and their outcomes. None of these regimens mandated the use of granulocyte colony-stimulating factor or antibiotic prophylaxis.

Table 1 Characteristics of included studies: prospective studies of initial empiric regimens for children with low-risk fever and neutropenia
Full size table

Outpatient vs. inpatient management

Table 2 compares outcomes of outpatient and inpatient management. Sixteen outpatient regimens, including three step-down regimens, were evaluated in 1,078 FN episodes while eight inpatient regimens were evaluated in 317 episodes. Nine out of 16 outpatient regimens and two out of eight inpatient regimens consisted of oral antibiotics. The following outcomes were not reported due to a lack of studies reporting these outcomes in either treatment setting: recurrent infection, sepsis, secondary infection, and readmission.

Table 2 Outcomes of initial or step-down outpatient versus inpatient management (any route of antibiotics)
Full size table

Treatment failure, when its definition included modification of antibiotics, was less likely to occur in the outpatient setting compared with the inpatient setting (15 % versus 28 %, P = 0.04) as demonstrated by the forest plot in Fig. 2. Overall and infection-related mortality were not significantly different between the inpatient and outpatient groups. There were two infection-related deaths in total; both patients had received inpatient therapy. Among 953 outpatient episodes, no infection-related deaths were reported. Treatment failure, when its definition excluded antibiotic modification, was not significantly different in both settings (Table 2).

Fig. 2
figure 2

Forest plot of treatment failure, including antibiotic modification, comparing outpatient with inpatient empiric regimens. Squares indicate percentages with horizontal lines representing 95 % CIs. Diamonds represent overall percentages from the meta-analysis with corresponding 95 % CIs. A test for heterogeneity [19] across subgroup results was used to determine if outcomes were modified based upon treatment setting or route of administration

Full size image

Oral vs. parenteral antibiotics

Table 3 compares outcomes of oral and parenteral antibiotic management. Eleven oral regimens, including five step-down regimens, were evaluated in 785 FN episodes, while 13 parenteral regimens were evaluated in 610 episodes. Nine out of 11 oral regimens and 7 out of 13 parenteral regimens were delivered in the outpatient setting. Oral antibiotic regimens consisted of: fluoroquinolone monotherapy (7 regimens, 581 episodes), fluoroquinolone and amoxicillin-clavulanate (3 regimens, 159 episodes), and cefixime (1 regimen, 45 episodes). Of the 10 studies using fluoroquinolones, 6 used ciprofloxacin (332 episodes), 3 used ofloxacin (207 episodes), and 1 used gatifloxacin (201 episodes).

Table 3 Outcomes of initial or step-down oral versus parenteral antibiotic management (any treatment setting)
Full size table

Treatment failure, when its definition included modification of antibiotics, was not significantly different between oral and parenteral antibiotic regimens (20 % versus 22 %, P = 0.68) as demonstrated by the forest plot in Fig. 3. There were two infection-related deaths in total; both had been treated with parenteral antibiotics. Among 676 episodes treated with oral regimens, no infection-related deaths were reported. There was no significant increase in treatment failure excluding antibiotic modification, duration of fever, recurrent infection, sepsis, secondary infection, adverse events, or readmission associated with oral therapy.

Fig. 3
figure 3

Forest plot of treatment failure, including antibiotic modification, comparing oral with parenteral empiric regimens. Squares indicate percentages with horizontal lines representing 95 % CIs. Diamonds represent overall percentages from the meta-analysis with corresponding 95 % CIs. A test for heterogeneity [19] across subgroup results was used to determine if outcomes were modified based upon treatment setting or route of administration

Full size image

Discussion

When the results of all prospective studies are combined, we were unable to demonstrate any difference in the efficacy of outpatient and oral antibiotic therapy compared with inpatient and parenteral antibiotic therapy, respectively. In particular, among all pediatric published studies of low-risk pediatric FN, 0 out of 953 episodes treated in the outpatient setting and 0 out of 676 episodes treated with oral antibiotics were associated with infection-related mortality, thereby underscoring the potential safety of these regimens in low-risk children.

Using our inclusion criteria, no trials were identified that randomized inpatient versus outpatient therapy. However, by combining data by regimen, we were able to calculate and compare outcomes between inpatient and outpatient protocols. Interestingly, treatment failure, including therapy modification, was more likely in the inpatient group. Since there was no difference in treatment failure between the two groups when antibiotic modification was excluded as a cause, it is likely that this difference was related to more frequent modification of therapy in the inpatient group. It is possible that inpatients were more likely to develop sepsis or hospital-acquired secondary infection which responded to a change in therapy. It is also possible that with closer clinical follow-up, modifications were more likely to occur in inpatients with persistent fever or suspected minor infections compared with outpatients, resulting in more frequent and potentially unnecessary antibiotic modification. However, we could not examine this in our meta-analysis as no inpatient studies reported these outcomes.

The secondary objective of this study was to evaluate whether the route of drug administration has an impact on the outcome of low-risk pediatric FN. We did not find a significant difference in treatment failure rates or infection-related mortality between oral and parenteral management. These findings are consistent with RCTs comparing oral and parenteral regimens in children [9, 17, 28, 31, 34]. When antibiotic modification was included as a cause of failure, treatment failure rates were about 20 % in both oral and parenteral groups.

Our results are important because although it is recognized that adult low-risk FN may be managed with outpatient or oral antibiotics as reflected by international treatment guidelines [16, 23], there is much more uncertainty in children. Initial management of low-risk pediatric FN in many oncology centers continues to involve inpatient parenteral antibiotics [8, 11, 35]. Previous evidence of the efficacy of outpatient or oral antibiotic therapy in this population has been limited by the very small number of randomized trials conducted in pediatrics, which has led to caution in extrapolating results to this population. However, our results, which are the most comprehensive to date, verify the efficacy of these approaches in pediatric patients. These good outcomes may be expected since at least one study has demonstrated that a highly selected group of adult and pediatric low-risk patients may be managed without antibiotics [30].

We limited eligibility to those studies which enrolled children within 24 h of FN presentation and to low-risk patients. We chose to restrict our analysis to those with new-onset FN as studies that enroll patients 24 h or more after FN onset might exclude patients who experience adverse outcomes early in the episode. We also limited the analysis to studies of low-risk patients as we believe that lower intensity therapy is most likely to be acceptable in this patient population in high-income countries. This approach resulted in the exclusion of two randomized studies [2, 39]. However, it is noteworthy that both studies showed similar results to ours.

One limitation of our analysis was that we relied on each study protocol’s definition of low-risk criteria. Risk in FN is likely to be a continuous variable [23], and as a result, the risk status of patients may have varied from trial to trial. Consequently, despite restricting the analysis to “low-risk” patients, it is possible that the outcomes in our comparison of treatment setting and route of administration may still be confounded by risk status. Nonetheless, in spite of this heterogeneity, we demonstrated no infection-related deaths in the outpatient and oral sub-groups irrespective of variable definitions of low-risk FN. Another limitation of this meta-analysis was the variability in definition of our primary outcome, treatment failure, across all studies. This composite endpoint encompassed multiple outcomes in most of the evaluated studies, resulting in significant heterogeneity. Furthermore, because treatment failure included readmission in most studies, outpatient failure rates may have been overestimated in our meta-analysis as readmission pertains only to outpatients. Finally, we restricted our review to studies published in the English language. However, a previous review found that restriction of systematic reviews to English, when compared to the inclusion of other languages, does not bias results [27].

In this study, we failed to demonstrate significant differences between sub-groups based upon tests of heterogeneity. It is possible that this failure to demonstrate significant heterogeneity was due to inadequate power; this is especially important for comparisons of outcome variables such as mortality in which the number of observed events was very low.

Our meta-analysis suggests that initial and step-down outpatient or oral antibiotic management of low-risk FN are effective in children. Centers should therefore adopt these lower-intensity strategies for low-risk pediatric FN patients if the infrastructure to safely manage outpatients can be established. While useful for clinical decision-making, the results of this study may also assist in educating and reassuring patients and families who might otherwise prefer inpatient management of FN [43]. Future research should focus on knowledge translation and effectiveness analysis of large numbers of patients treated outside of research protocols as more centers begin to adopt these lower-intensity strategies for low-risk pediatric FN.