Colony‐stimulating factors for prevention of myelosuppressive therapy‐induced febrile neutropenia in children with acute lymphoblastic leukaemia

Abstract

Background

Acute lymphoblastic leukaemia (ALL) is the most common cancer in childhood and febrile neutropenia is a potentially life‐threatening side effect of its treatment. Current treatment consists of supportive care plus antibiotics. Clinical trials have attempted to evaluate the use of colony‐stimulating factors (CSF) as additional therapy to prevent febrile neutropenia in children with ALL. Individual trials have not demonstrated significant benefit. Systematic reviews provide the most reliable assessment and the best recommendations for practice.

Objectives

To evaluate the safety and effectiveness of the addition of granulocyte colony‐stimulating factors (G‐CSF) or granulocyte macrophage colony‐stimulating factors (GM‐CSF) to myelosuppressive chemotherapy in children with ALL in an effort to prevent the development of febrile neutropenia. Evaluation of number of febrile neutropenia episodes, length to neutrophil count recovery, incidence and length of hospitalisation, number of infectious disease episodes, incidence and length of treatment delays, side effects (flu‐like syndrome, bone pain and allergic reaction), relapse and overall mortality (death).

Search methods

The search covered the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, CANCERLIT, LILACS, and SciElo. We manually searched records of conference proceedings of ASCO and ASH from 1985 to 2003 and used the electronic databases of the ASCO and ASH web sites to search for abstracts from 2003 to September 2008, as well as databases of ongoing trials. We consulted experts and scanned references from the relevant articles.

Selection criteria

We looked for randomised controlled trials (RCTs) comparing CSF with placebo or no treatment as primary or secondary prophylaxis to prevent febrile neutropenia in children with ALL.

Data collection and analysis

Two authors independently selected and critically appraised studies and extracted relevant data. The end points of interest were: 
 
 * Primary end points: number of febrile neutropenia episodes and overall mortality (death) 
 * Secondary end points: time to neutrophil count recovery, incidence and length of hospitalisation, number of infectious diseases episodes, incidence and length of treatment delays, side effects (flu‐like syndrome, bone pain and allergic reaction) and relapse. 
 
 We conducted a meta‐analysis of these end points and expressed the results as Peto odds ratios. For continuous outcomes we calculated a weighted mean difference and a standardised mean difference. For count data, we conducted a meta‐analysis of the logarithms of the rate ratios using generic inverse variance.

Main results

We scanned more than 6800 citations and included six studies with a total of 333 participants in the analysis. There were insufficient data to assess the effect on survival. The use of CSF significantly reduced the number of episodes of febrile neutropenia episodes (Rate Ratio = 0.63; 95% confidence interval (CI) 0.46 to 0.85; P = 0.003, with substantial heterogeneity), the length of hospitalisation (weighted mean difference (WMD) = ‐1.58; 95% CI ‐3.00 to ‐0.15; P = 0.03), and number of infectious disease episodes (Rate Ratio = 0.56; 95% CI 0.39 to 0.80; P = 0.002). Despite these results, CSF did not influence the length of episodes of neutropenia (WMD = ‐1.11; 95% CI ‐3.55 to 1.32; P = 0.4) or delays in chemotherapy courses (Rate Ratio = 0.75; 95% CI 0.47 to 1.20; P = 0.23) .

Authors' conclusions

Children with ALL treated with CSF benefit from shorter hospitalisation and fewer infections. However, there was no evidence of shortened duration of neutropenia nor fewer treatment delays. There was also no useful information about survival. 
 
 The role of CSF in the context of febrile neutropenia episodes is still uncertain. Although current data show statistical benefit with CSF use, substantial heterogeneity between included trials does not allow this conclusion.

Plain language summary

Prophylactic administration of colony‐stimulating factors reduces hospital stay and risk of infections in children with acute lymphoblastic leukaemia

The authors evaluated the efficacy of adding colony‐stimulating factors (CSF) after chemotherapy in children with acute lymphoblastic leukaemia (ALL) to prevent febrile neutropenia, which is a potentially life‐threatening side effect of treatment. 
 
 There is a lack of studies to determine the best CSF dose for children and only a small number of RCTs evaluating the role of CSF in children's ALL. The prophylactic administration of CSF reduces hospital stay, and risk of infections. The authors did not find evidence that CSF reduces febrile neutropenia episodes, their duration, or treatment delays in children with ALL undergoing chemotherapy. Follow up was too short to provide useful information on any possible effect on relapse or survival.

Background

The administration of more intensive chemotherapy (Pui 1998; Pui 2001) combined with advances in supportive care (Gmur 1991; Kaushansky 2000; Kritz 1991; Lok 1994) have led to improved survival rates among children with acute lymphoblastic leukaemia (ALL). Chemotherapy‐induced febrile neutropenia remains a life‐threatening complication for patients with cancer and causes prolonged hospitalisation (Alexander 1999; Chanock 1996; Lee 1998; Petrilli 1993; Pizzo 1982; Pizzo 1991), increased costs of therapy (Aquino 1997; Aquino 1997(2); Bash 1994; Elting 2002; Rosenman 2002) and delay of chemotherapy courses (Pizzo 1999). Current standard practices of instituting empiric broad‐spectrum antibiotic therapy in the setting of fever with neutropenia have reduced mortality rates to around 2% to 5% (Alexander 1999; Pizzo 1993). The use of hematopoietic growth factors (Vose 1995) has led to shorter periods of chemotherapy‐induced neutropenia (Furman 1991; Laver 1998; Lifton 1996; Pui 1997; Riikonen 1994; Saarinen 1992; Welte 1996), post‐transplant neutropenia after both allogeneic and autologous bone marrow transplants (Madero 1995; Nemunaitis 1992; Saarinen 1996) and neutropenic fever (Riikonen 1994). 
 
 Colony‐stimulating factors (CSF) are a family of cytokines that regulate the proliferation and differentiation of hematopoietic cells (Griffin 2001). Among them, granulocyte colony stimulating factors (G‐CSF) and granulocyte‐macrophage colony stimulating factors (GM‐CSF) have been tested in clinical trials involving participants with febrile neutropenia. G‐CSF regulates the production of the neutrophil lineage. The administration of G‐CSF results in a dose‐dependent increase in circulating neutrophils (Griffin 2001; Petros 2001). This is mainly due to a reduced transit time from stem cell to mature neutrophil (Griffin 2001). GM‐CSF is a growth factor for the myeloid lineage that stimulates the growth of granulocytes, macrophages and eosinophil colonies (Griffin 2001; Petros 2001). Administration of GM‐CSF results in a dose‐dependent increase in neutrophils, eosinophils, macrophages and sometimes lymphocytes (Griffin 2001; Petros 2001). 
 
 CSF can be used in two prophylactic ways. In primary prophylaxis, CSF is used before the onset of any neutropenia or febrile neutropenia episode, while secondary prophylaxis is the use of CSF after at least one episode of febrile neutropenia to prevent a new episode or to avoid dose reductions and delays in chemotherapy courses (Ozer 2000). 
 
 A variety of phase I and phase II studies in children with cancer receiving intensive chemotherapy have documented that primary administration of G‐CSF reduces the duration of neutropenia, lowers rates of neutropenic fever, decreases use of antibiotics and diminishes the need for hospitalisation compared to historical control subjects who did not receive G‐CSF (Lifton 1996; Welte 1997). Other larger RCTs showed a lower incidence of febrile neutropenia and confirmed infection as well as a shorter duration of total antibiotic use compared with those receiving a placebo (Riikonen 1995; Welte 1996). However, not all studies have shown a benefit from the use of G‐CSF (Laver 1998; Pui 1997; Rubino 1998). RCTs of GM‐CSF administered in supportive care in childhood malignancies as primary prophylaxis of neutropenic fever have usually resulted in lesser duration or severity of neutropenia, but clinical benefits have not been as consistently positive as those observed with G‐CSF (Armitage 1998; Burdach 1995; Lifton 1996; Wexler 1996). Although some RCTs have compared the prophylactic use of CSF with no treatment or placebo in children, none of these has shown a significant benefit (Lee 1998; Little 2002; Ohno 1993; Riikonen 1995; Wexler 1996). The long‐term benefits of using growth factors during intensive phases of ALL therapy remain controversial (Pui 1997; Welte 1996). Moreover the safety of CSF is still not well established, as receptors for some CSFs found on the membranes of lymphoblastic cells (Inukai 1998; Trus 2003; Benko 2001) could, hypothetically, influence treatment outcomes. For these reasons, as well as financial considerations, G‐CSF and other growth factors are not routinely used in standard risk ALL regimens. They are more often administered in cases of bone marrow transplants and relapsed ALL and occasionally in front‐line high‐risk patients undergoing exceptionally intensive chemotherapy (Saarinen‐Pihkala2000). The routine use of these agents after lesser myelosuppressive chemotherapy is not supported by the literature. Therefore the use of CSF in prophylaxis is not uniform and depends on physician preference (Parsons 2000) and available data (Ozer 2000). The perception of high‐dose intensity and anticipated toxicity of paediatric chemotherapeutic regimens also tend to promote CSF use (Parsons 2000). 
 
 Such issues have led to reviews evaluating the role of CSFs in other types of haematological malignancies and populations. Bohlius 2008 performed a systematic review to evaluate the role of G‐CSF and GM‐CSF in the prevention of adverse effects secondary to lymphoma treatment. Although there were reduced risks of neutropenia, febrile neutropenia and infection, there is still no evidence that G‐CSF or GM‐CSF provide a more favourable response rate, freedom of treatment failure or overall survival. 
 
 The lack of any conclusive data derived from seemingly conflicting studies of the prophylactic use of CSF to prevent myelosuppressive therapy‐induced neutropenia in children with ALL demands a systematic review to equip clinicians with reliable data upon which to base clinical decisions.

Objectives

To evaluate the safety and effectiveness of the addition of G‐CSF or GM‐CSF to chemotherapy in children with ALL, in an effort to prevent the development of febrile neutropenia. In particular, we seek evaluation of the number of febrile neutropenia episodes, length of time to neutrophil count recovery, length of hospitalisation, number of infectious disease episodes, number of infectious diseases episodes, incidence and length of treatment delays, side effects (flu‐like syndrome, bone pain and allergic reaction), relapse 
 and overall mortality (death).

Methods

Criteria for considering studies for this review

Types of studies

RCTs with a parallel design that compare CSF versus placebo or no treatment given prior to the installation of neutropenia related to chemotherapy in children with ALL. We included information from ongoing studies and interim analyses with the same design. There was no restriction on language of publication.

Types of participants

Children (0 to 18 years) with ALL receiving myelosuppressive chemotherapy, excluding situations related to bone marrow transplantation. We included trials with upper age range of 19, 20, or 21 where we could not obtain separate results for the 0 to 18 year age range.

Types of interventions

The administration of CSF to prevent febrile neutropenia induced by chemotherapy in children with ALL, given concomitantly or after 24 to 48 hours, concluded the chemotherapy course as primary or secondary prophylaxis to prevent febrile neutropenia. G‐CSF or GM‐CSF administered subcutaneously in a dose of at least 5 mcg/kg body weight per day, given until absolute neutrophil counts reached more than 0.5 x 10⁹/L. We excluded trials of high‐dose myeloablative chemotherapy regimens followed by bone marrow transplantation due to the length of severe neutropenia that differs considerably compared to standard dose regimens for paediatric ALL. We also excluded trials using CSF for treatment of febrile neutropenia because this is a different situation.

Types of outcome measures

Primary outcomes

• Number of febrile neutropenia episodes

• Overall mortality (death)

Secondary outcomes

• Time to neutrophil count recovery

• Incidence and length of hospitalisation

• Number of infectious diseases episodes

• Incidence and length of treatment delays

• Side effects (flu‐like syndrome, bone pain and allergic reaction)

• Relapse

Search methods for identification of studies

We identified relevant trials in any language through:

Electronic searches

We performed a wide search in the main computerised databases of interest from 1985 to 2008:

• The Cochrane Central Register of Controlled Trials (CENTRAL)

• Cancer Literature On line (CANCERLIT)

• Excerpta Medica Database (EMBASE)

• Literatura Latino‐Americana e do Caribe em Ciências da Saúde (LILACS)

• Medilars Online (MEDLINE)

• Scientific Electronic Library Online (SciELO)

• American Society of Clinical Oncology Website (ASCO) ‐ searched from 2003 to September 2008

• American Society of Hematology Website (ASH) ‐ searched from 2003 to September 2008

Search strategy in electronic databases: the methodological search strategy used in MEDLINE is that reported by Dickersin 1994. We used adaptations of this strategy in EMBASE (Egger 2001) and LILACS (Castro 1999) searches (see Appendix 1). For the ASCO and ASH websites we used combinations of terms used in MEDLINE.

Searching other resources

Search of databases of ongoing trials

We performed a search of the following databases:

• Current Controlled Trials Register: http://www.controlled‐trials.com;

• European Organisation for Research and Treatment of Cancer (EORTC): http://www.eortc.be;

• United Kingdom Co‐ordinating Committee on Cancer Research (UKCCCR): http://www.ctu;mrc.ac.uk/ukcccr/;

• UK National Research Register of all NHS‐funded research: http://www.doh.gov.uk/research/nrr.htm;

• National Cancer Institute, America: http://www.cancertrials.nci.nih.gov/researchers/index.html;

• National Cancer Institute, Canada: http://www.ctg.queensu.ca/ctg_home.htm;

• National Health and Medical Council of Australia: http://www.ctc.usyd.edu.au/.

• Cochrane Central Register of Controlled Trials: http://www.mrw.interscience.wiley.com/cochrane/cochrane_clcentral_articles_fs.html

Contact

We consulted experts in oncology and haematology about ongoing or currently unpublished studies. We also contacted Roche do Brasil (Brazilian division of La Roche Pharmaceuticals) and Aventis Pharma do Brasil (Brazilian division of Aventis Pharma) which are involved in the manufacture of CSF about ongoing or unpublished randomised trials.

Manual search

We searched citations from identified trials and relevant review articles as well as conference proceedings until 2003 of the

• American Society of Clinical Oncology (ASCO),

• American Society of Hematology (ASH ),

• European Society of Medical Oncology (ESMO).

Data collection and analysis

Two authors (ECS, OC) independently screened titles and abstracts of studies identified from the above sources. Where we found this to be unsatisfactory, we obtained a full text version for assessment. We resolved disagreements through a consensus meeting. We called upon a third author (ADS) to resolve disagreement where necessary.

Data extraction

Two authors (ECS, OC) independently extracted the data from articles. For identification purposes, we used the name of the first author and the publication year. We extracted all data directly from the text or, where possible, made calculations based on the information available. 
 
 We extracted the following quality domains empirically linked to bias (Egger 2001): methods of randomisation, data regarding allocation concealment, blinding of patients, blinding of care‐givers, blinding of outcome evaluators, description of withdrawals and drop‐outs, details regarding intention‐to‐treat analysis, power analysis, and source of funding. We retrieved selected clinical outcomes and additional data.

Selected clinical outcomes and definitions

Primary endpoints

• Number of febrile neutropenia episodes: we defined febrile neutropenia (Ozer 2000) as oral temperature above 38.3ºC unrelated to transfusions associated with grade IV neutropenia (absolute neutrophil count (ANC) less than 500/mm³);

• overall mortality or death: we included deaths related or not related to chemotherapy, use of CSF or progression of disease.

Secondary endpoints

• Time to neutrophil count recovery: we considered the number of days needed for the ANC to rise from less than 500/mm³ to 500/mm³ after each course of chemotherapy.

• Incidence and length of hospitalisation: we considered the number of episodes and days of hospitalisation due to chemotherapy infusions or any complication.

• Number of infectious disease episodes: we considered an episode of infection as every objective sign of infection of any site confirmed by microbiologic assay or radiologic exam associated with compatible clinical signs in the groups studied.

• Incidence and length of treatment delays: we defined an episode as delayed when a subsequent course of chemotherapy started after the scheduled day due to complications resulting from treatment. We determined the length of treatment delay to be the number of days elapsed since the originally scheduled start of a course of chemotherapy.

• Side effects: we considered each episode of flu‐like syndrome, bone pain, allergic reaction as a side effect related to CSF use and evaluated both groups.

• Relapses: we considered the recurrence of ALL after positive response to myelosuppressive chemotherapy at any time after starting treatment.

Additional data evaluated in sub‐group analysis

• Diagnostic criteria of neutropenia (ANC);

• use of CSF in first cycle or after first neutropenia episode;

• treatment schedule and CSF used;

• CSF concurrent with chemotherapy or in intervals between cycles;

• ANC median and mean in each arm at the beginning of study;

• length of study.

We resolved disagreements by a consensus meeting. Where necessary we invited a third author to give an opinion.

Analysing and presenting the results

We entered data regarding the endpoints of interest into Review Manager software (RevMan). 
 
 We performed a formal statistical analysis of heterogeneity using the Chi² test (DerSimonian 1986) and calculated the percentage of the variability in the effect estimates using I² as described by Higgins et al (Higgins 2002; Higgins 2003). We set the statistical level of heterogeneity (Chi² P) at 0.10 and considered I² below 30% to represent homogeneity. We assessed any identified heterogeneity in an effort to explain it. If we were unable to find an explanation we stressed such a finding in the review and highlighted that caution in the interpretation of these data was appropriate. Where we found that a cause for heterogeneity was apparent and justified a separate analysis of the studies with a particular characteristic, we undertook such analysis and presented it.

Where we found no heterogeneity, we included the studies in a meta‐analysis for the outcomes selected above. We performed meta‐analysis using RevMan 5. 
 
 For dichotomous data we calculated relative risk, odds ratio and Peto odds ratio (OR) (Yusuf 1985). If these data showed the same conclusion, we used Peto OR in the description of results. We set a significance level of 5% to test our hypothesis. We reported the 95% confidence interval (CI) in the statistical summary. When overall results were significant, we calculated the absolute risk reduction and the number needed to treat (NNT) or the number needed to harm (NNH) to produce or prevent an outcome based on the inverse of the absolute risk reduction (McQuay 1997).

For continuous outcomes we used a fixed‐effect method with the inverse variance approach. We calculated a weighted mean difference (WMD) when outcome measurements in all trials were made on the same scale. We used a standardised mean difference (SMD) when the trials all assessed the same outcome, but measured the outcome in a variety of ways. We did so irrespective of whether the data for each individual were single assessments or changed from baseline measures.

For purposes of count data, we calculated the logarithms of the rate ratios of each trial and combined them using the generic inverse variance method (Hasselblad 1995) when a participant experienced an event more than once. We calculated an approximate standard error of the log rate ratio given by √(1/A + 1/C) whenever necessary.

We planned to assess the possibility of publication bias using the funnel plot method (Egger 1997) for the primary endpoints.

Improving and updating reviews

We will update every two years.

Results

Description of studies

Results of the search

Two authors independently scanned more than 6800 citations identified in the computerised databases using the search strategy previously outlined. We identified 12 randomised controlled trials that met our inclusion criteria for full text reading and analysis of quality.

Included studies

Six studies (Calderwood 1994; Clarke 1999; Dibenedetto 1995; Michel 2000; Pui 1997; Welte 1996) met our inclusion criteria (see Characteristics of included studies) and we included their data in the meta‐analysis.

Characteristics of included studies

Five trials described the effects of G‐CSF (Clarke 1999; Dibenedetto 1995; Michel 2000; Pui 1997; Welte 1996). Only one trial studied the effects of GM‐CSF (Calderwood 1994). There was a cross‐over study (Clarke 1999) where children were randomised to receive G‐CSF starting four days after completion of either the first or the second block of intensification therapy. In order to avoid the carry‐over situation that interferes with the effects of the intervention, we decided to include available data of the first period. We found only one kind of data extractable for meta‐analysis from this trial: delay of chemotherapy. 
 One trial defined neutropenia as absolute neutrophil count (ANC) of less than 1000/mm³ (Clarke 1999) while the other five considered neutropenia an ANC of less than 500/mm³. One study (Clarke 1999) included one participant with advanced T‐cell Hodgkin's lymphoma who underwent a treatment protocol similar to patients with ALL and was therefore considered eligible for the trial. 
 
 Two of the included studies evaluated some outcomes after only one cycle of treatment (Clarke 1999; Pui 1997). Welte et al (Welte 1996) evaluated outcomes coming from observation after nine courses of chemotherapy, while Michel (Michel 2000) analysed six courses, Dibenedetto (Dibenedetto 1995) studied three cycles and Calderwood (Calderwood 1994) analysed data after two courses of chemotherapy. 
 
 Three studies used 5 mcg/kg body weight per day (Clarke 1999; Michel 2000; Welte 1996) while two others (Dibenedetto 1995; Pui 1997) used 10 mcg/kg body weight per day of G‐CSF. GM‐CSF was used at a dose of 5.5 mcg/kg on days 5 to 11 and 19 to 25 of each treatment cycle (Calderwood 1994). 
 
 Most of the included studies used CSF as primary prophylaxis and only one (Dibenedetto 1995) used CSF as secondary prophylaxis. The objective of treatment is to prevent subsequent episodes of febrile neutropenia. 
 
 We retrieved no eligible citations from either the search of databases of ongoing trials or contact with the pharmaceuticals companies.

Excluded studies

Six studies did not meet the inclusion criteria. The reasons for exclusion were: involvement of adult participants; data non‐extractable for children; and non‐randomised design (Chen 1998; Heath 2003; Laver 1998; Little 2002; Ohno 1993; Saarinen‐Pihkala2000) (see Characteristics of excluded studies).

Risk of bias in included studies

Calderwood et al and Welte et al (Calderwood 1994; Welte 1996) described an adequate method of randomisation but only one (Calderwood 1994) had a clear and adequate allocation concealment. Although Michel 2000 had described adequate randomisation, the two arms of comparison were unbalanced: there were more children with a specific cytogenetic feature (t(4;11)) in the non‐G‐CSF group than in the G‐CSF group (six patients versus one patient, respectively; P = 0.05). This difference explained the higher mean white blood cell count at diagnosis in the non‐G‐CSF group (223 x 10⁹ versus 136 x 10⁹/L, respectively) and the higher incidence of infants less than one year old in the same group (six patients versus one patient respectively). 
 
 Four studies (Clarke 1999; Dibenedetto 1995; Michel 2000; Welte 1996) used no treatment in the control group. Two other trials (Calderwood 1994; Pui 1997) used a placebo in the group which was also double blind. 
 
 There was a variation in CSF doses used. Calderwood 1994 used 5.5 mcg/kg body weight per day of GM‐CSF in his trial. Three studies used 5 mcg/kg body weight per day of G‐CSF (Clarke 1999; Michel 2000; Welte 1996) while two others (Dibenedetto 1995; Pui 1997) used 10 mcg/kg body weight per day. 
 
 Alpha‐error and beta‐error were pre‐determined in two trials (Pui 1997; Welte 1996) but the number of participants was not reached in one of them (Welte 1996). Intention‐to‐treat analysis was performed in only two of the included studies (Clarke 1999; Welte 1996). 
 
 Only one trial included in the meta‐analysis was referred to as multicentre (Welte 1996).

Effects of interventions

Our analysis included results of six trials (Calderwood 1994; Clarke 1999; Dibenedetto 1995; Michel 2000; Pui 1997; Welte 1996) with a total of 333 participants. One hundred and sixty‐two children with ALL were randomised to use CSF as the treatment group and 171 were allocated to the control group (placebo or no treatment).

Primary end points

Number of febrile neutropenia episodes

Febrile neutropenia was defined (Ozer 2000) as oral temperature above 38.3ºC unrelated to transfusions associated with grade IV neutropenia (ANC less than 500/mm³). Only two studies comprising 451 cycles of treatment had extractable data about febrile neutropenia episodes (Pui 1997; Welte 1996). There were 68 episodes of febrile neutropenia in 226 cycles of chemotherapy in the CSF group and 112 episodes in 228 cycles of treatment in the control group. We considered that each participant could experience an episode of febrile neutropenia episode and may also experience it more than once. For this reason we calculated the logarithms of the rate ratios in each trial to perform meta‐analysis (Hasselblad 1995). Meta‐analysis showed significant difference to the CSF group (Rate Ratio 0.63, 95% Confidence Interval (CI) 0.46 to 0.85, Figure 1). We detected substantial statistical heterogeneity between the studies (x² = 4.80, degrees of freedom (df): 1; P = 0.03). There was 47% of risk reduction (Risk Reduction ‐0.47, 95% CI ‐0.77 to ‐0,16) with a number needed to treat (NNT) of 2 to prevent a febrile neutropenia episode. A funnel plot was not generated because only two studies were included in the analyses and publication bias could not be assessed in this case (Table 1).

1.

Forest plot of comparison: 1 Febrile neutropenia episodes, outcome: 1.1 Febrile neutropenia episodes.

Death

Only two studies (Calderwood 1994; Michel 2000) described overall mortality rate with a small number of deaths or no deaths occurring during the trial period. Calderwood reported one death due to septic shock, and Michel described three deaths from therapy‐related toxicity.

Secondary end points

Time to neutrophil count recovery

We considered the number of days needed to rise from an ANC less than 500/mm³ after each chemotherapy course with or without CSF. Three studies (Calderwood 1994; Dibenedetto 1995; Michel 2000) with 134 participants had extractable data related to the length of neutropenia episode. Patients receiving CSF had shorter duration of neutropenia compared to patients not receiving CSF (WMD = ‐3.44, 95% CI ‐4.76 to ‐2.12, P < 0.00001, Figure 2) but we detected a small heterogeneity between studies (x² = 5.22, df: 2, P = 0.07). The data from Michel 2000 was the main cause of this heterogeneity and we performed the analysis of data excluding this study (Michel 2000). 
 
 With two studies (Calderwood 1994; Dibenedetto 1995) including 67 participants, the group using CSF did not show shorter duration of neutropenia compared to the group not receiving CSF (WMD = ‐1.11, 95% CI ‐3.55 to 1.32, P = 0.37, Figure 3). We resolved the heterogeneity detected between studies after exclusion of the data from Michel et al (x² = 0.28, df: 1; P = 0.60).

2.

Forest plot of comparison: 2 Length of neutropenia, outcome: 2.1 Length of neutropenia.

3.

Forest plot of comparison: 2 Length of neutropenia, outcome: 2.2 Length of neutropenia excluding Michel 2000.

Incidence and length of hospitalisation

We considered the number of episodes and days of hospitalisation due to chemotherapy infusions or any complication and pooled them. There were no extractable data about the incidence of hospitalisation in the included trials. In the evaluation of the length of hospitalisation, meta‐analysis of three trials (Calderwood 1994; Dibenedetto 1995; Michel 2000) including 134 participants showed significant benefit of CSF addition to the chemotherapy compared to control groups (WMD = ‐1.58, 95% CI ‐3.00 to ‐0.15; P = 0.03, Figure 4). There was no evidence of heterogeneity between the studies (x² = 0.71, df: 2; P = 0.70).

4.

Forest plot of comparison: 3 Length of hospitalisation, outcome: 3.1 Length of hospitalisation.

Number of infectious diseases episodes

We considered every objective sign of infection of any site confirmed by microbiologic assay or radiologic exam associated with compatible clinical signs in the groups studied. Five studies (Calderwood 1994; Dibenedetto 1995; Michel 2000; Pui 1997; Welte 1996) had data of 958 cycles of treatment which we included in meta‐analysis. There were 45 episodes of infection among 476 cycles of chemotherapy in the CSF group and 91 episodes in 482 cycles of treatment in the control group. Meta‐analysis showed a significant reduction in the incidence of infection (Rate Ratio=0.56, 95% CI 0.39 to 0.80, P = 0.002, Figure 5) without detectable statistical heterogeneity (x² = 4.91; df: 4; P = 0.30). There was a 58% of risk difference (RD ‐0.58, 95% CI ‐0.93 to ‐0.22) with a NNT of 2 to prevent an infectious disease episode.

5.

Forest plot of comparison: 5 Number of infectious disease episodes, outcome: 5.1 Number of infectious disease episodes.

Incidence and length of delays in chemotherapy courses

We considered any delay to restart the subsequent course of chemotherapy as an episode. Four studies (Calderwood 1994; Clarke 1999; Dibenedetto 1995; Welte 1996) including 118 patients had data related to delays in chemotherapy courses extractable for meta‐analysis. There were 30 delay episodes among 257 cycles of therapy in CSF groups compared to 46 delays in 281 cycles among control patients. The analysis did not show a difference between groups (Rate Ratio = 0.75, 95% CI 0.47 to 1.20, P = 0.23, Figure 6). There was no detectable heterogeneity in this analysis (x² = 1,24; df: 3; P = 0.74). 
 
 The trials did not report on the duration of delays and meta‐analysis of this outcome was not feasible.

6.

Forest plot of comparison: 4 Delays in chemotherapy courses, outcome: 4.1 Number of delays in chemotherapy.

Side effects

A lack of uniformity of adverse effects described between trials meant that we could not perform meta‐analysis. Overall we considered the adverse effects associated with the use of CSF in children to be mild. Calderwood detected that 15% of patients had localised redness or swelling at injection sites, low‐grade fever or muscle aches and pains. One patient in this study experienced an urticarial eruption which necessitated the patient's withdrawal from the study (Calderwood 1994). In the trial coordinated by Welte et al the authors described one adverse event (vasculitis of WHO grade III) that was related to the administration of G‐CSF (Welte 1996). Dibenedetto reported that G‐CSF was well tolerated with mild bone tenderness as the only side effect being observed in two patients (Dibenedetto 1995). 
 
 Data about the use of antibiotics were presented in three trials (Calderwood 1994; Pui 1997; Welte 1996). Of these, two (Pui 1997; Welte 1996) presented median values with insufficient data to calculate mean values for meta‐analysis. Calderwood et al reported no significant difference between the experimental and control groups in the number of days that patients needed antibiotics (P = 0.73) (Calderwood 1994).

Relapse

None of the included studies was long enough to verify a relapse rate and there were no data available. Therefore we could not perform meta‐analysis of this important outcome.

Additional information is shown in Table 1 and Table 2.

1. Total cycles of chemotherapy for febrile neutropenia, infections and delays.

Study ID	Group numbers	Number of courses	Febrile Neutrop data	Infection data	Count of infection	Chemotherapy delays	Total cycles
Calderwood	16 CSF vs 19 control	2	none available	13 CSF vs 15 control	13 vs 15	9 in CSF vs 15 in control	32 in CSF vs 38 in control
Clarke	8 CSF vs 9 control	2	none available	none available	none available	3 CSF vs 6 control	9 in CSF and 8 in control
Dibenedetto	14 CSF vs 18 control	1 (4 in total)	none available	2 vs 2	2 vs 2	13 in CSF vs 16 in control	56 in CSF vs 72 in control
Michel	34 CSF vs 33 control	6	none available	septicaemia 4% (CSF) vs 11% (control) per patient per course	8 vs 22	5 CSF vs 9 control	204 in CSF vs 198 in control
Pui	73 CSF vs 75 control	1	42 CSF vs 51 control	12CSF vs 27 control (documented infections)	12 vs 27	not available	73 in CSF vs 75 in control
Welte	17 vs 17	9	17% (26) vs 40% (61) per cycle	8% vs 15% cultures confirmed	12 vs 23	not available	153 vs 153

2. Details of ALL treatment in the included studies.

Study ID	Cycle CSF given	Follow up period	Total cycles	Chemotherapy used	Protocol name
Calderwood 1994	GM‐CSF D5‐11, D19‐25; first cycle	28 days	1	(Intensification) cyclophosphamide 1g/m2 D0 and D13; cytosine arabinoside 75mg/m2 D1‐4, 8‐11, 15‐18 e 22‐25; 6‐mercaptopurine 60mg/m2 D0‐27	French‐American‐British L3 protocol
Clarke 1999	G‐CSF after first cycle and second cycle if intensification as cross‐over basis	unclear (2 blocks = 60 days?)	2	Induction ‐ vincristine, prednisolone ans asparaginase; Intensification ‐ unclear; Maintenance ‐ monthly vincristine and prednisolone, daily mercaptopurine and weekly oral methotrexate	MRC UKALL XI
Dibenedetto 1995	G‐CSF after cytarabine in phase 2	unclear (3 cycles)	3?	Induction‐ phase 1: prednisone D0‐28, vincristine 1,5mg/m2 D7,14,21 and 28; daunorubicin 30mg/m2 D7,14,21, and 28 and L‐asparaginase 10000UI/m2 D18,24,27,30,33,36 and 39. Phase 2: cyclophosphamide 1g/m2 D42,70; cytarabine 75mgm2 D44‐47,51‐54,58‐61,65‐68; 6‐mercaptopurine 60mg/m2 D35‐49	Berlin‐Frankfurt‐Munster Protocol
Michel 2000	G‐CSF during consolidation	median of 27 months	5	Consolidation (alternate phases until course 6): phase 1 ‐ cytarabine 2g/m2 D1‐2; etoposide 150mg/m2 D3‐5; Dexamethasone 20mg/m2 D2‐5. Phase 2: Vincristine 1.5mg/m2 D1; Methotrexate 8g/m2; Cyclophosphamide 375mg/m2 D2‐3; Doxorubicin 60 mg/m2 D2; Prednisone 60mg/m2 D1‐5.	Fralle 93
Pui 1997	G‐CSF one day after induction	unclear (1cycle)	1	Induction: prednisone 40mg/m2/day for 4‐weeks; vincristine 1.5mg/m2/week for 4 weeks; asparaginase 1.000UI/m2 3times a week for 3 weeks; daunorubicin 25mg/m2 D1, 8; etoposide and cytarabine 300mg/m2 each, D22,25 and 29.	Total Therapy Study XIIIA
Welte 1996	G‐CSF after first cycle, then for 9 cycles	approx. 30 weeks	9	HR1: Dexamethasone 20mg/m2/d D1‐5; 6‐mercaptopurine 100mg/m2/d D1‐5; vincristine 1.5mg/m2/d D1,6; cytosine‐arabinoside 2g/m2x2 D5; methotrexate 5g;m2 D1 (24hs); L‐asparaginase 25000UI/m2 D1.HR‐2: Dexamethasone 20mg/m2/d D1‐5; 6‐thioguanine 100mg/m2/d D1‐5; vindesine 3mg/m2/d D1; daunorubicin 50mg/m2 D5; methotrexate 5g;m2 D1 (24hs); ifosfamide 400mg/m2/d D1‐5; L‐asparaginase 25000UI/m2 D1. HR3: Dexamethasone 20mg/m2/d D1‐5; cytosine‐arabinoside 2g/m2x4 D1‐2; etoposide 150mg/m2/d D6; L‐asparaginase 25000UI/m2 D6.	Berlin‐Frankgurt‐Munster 90 Protocol

Discussion

Colony‐stimulating factors are being used to prevent febrile neutropenia and infection in patients undergoing myelosuppressive treatment for cancer. Their use is more established in adult populations where CSF is often used in cancer treatment. Some systematic reviews had already evaluated the effect of CSF in chemotherapy‐induced febrile neutropenia in cancer (Clark 2005) and more particularly in malignant lymphoma (Bohlius 2008). Meanwhile the effect of CSF in febrile neutropenia due to chemotherapy in childhood cancer remains unclear and inspired this review. 
 
 This systematic review presents evidence for the use of CSF to prevent febrile neutropenia in children with ALL during chemotherapy. There is still only a small number of RCTs involving paediatric subjects and most of them present methodological quality concerns. 
 
 A Paediatric Oncology Group trial found contradictions between guidelines and clinical practices about CSF use in the paediatric oncology field (Parsons 2000). In the absence of definitive data, paediatric oncologists adopt therapeutic patterns based on convenience. Once established, such patterns are difficult to transform. 
 
 In 1997, the US Food and Drug Administration Modernization Act showed the need for specific RCTs in paediatric populations to define the benefits of various medications. Because of the development of new therapeutic agents, their effects should be determined through well‐designed RCTs and determination of relevant outcomes. In the absence of this information, recommendations incorporate evidence derived from trials in adult populations (Ozer 2000; Schaison 1998). In addition, the opinions of paediatric oncologists have been influencing routine use of CSF. There is still a need for further, well‐designed RCTs in this field. 
 
 We made important observations relating to some aspects of the included studies in this systematic review. First, there is no consensus about the ideal dose of CSF for children. Three studies used 5 mcg/kg body weight per day (Clarke 1999; Michel 2000; Welte 1996) while two others (Dibenedetto 1995; Pui 1997) used double this dose. We tried to perform a subgroup analysis comparing two different doses but, because of the small number of included studies, this was a meaningless exercise. The dearth of RCTs to determine the ideal doses of CSF to children may lead to a waste of financial resources and unnecessary exposure of paediatric participants to potential side effects without any clear conclusion. This review showed the need for new trials to determine the ideal doses of CSF for children. 
 
 Second, most of the studies included were of short duration of evaluation periods. Welte et al (Welte 1996) evaluated outcomes coming from observation after nine courses of chemotherapy, while Michel (Michel 2000) analysed six courses, Dibenedetto (Dibenedetto 1995) analysed data from three cycles and Calderwood (Calderwood 1994) analysed data after two courses of chemotherapy. The others (Clarke 1999; Pui 1997) evaluated some outcomes after only one cycle of treatment. Such a short period of study can influence the result of this systematic review. Long‐term treatment required in ALL chemotherapy regimes often diminishes bone marrow reserves progressively. This can negate the potential benefit of CSF after a low reserve is reached. Regardless of the treatment protocol used, the use of CSF as primary or secondary prophylaxis should comprise the most myelosuppressive phases in ALL treatment that include induction, intensification and re‐intensification phases. These phases have higher prevalence of febrile neutropenia and other complications of chemotherapy than the maintenance phase. Depending on each treatment protocol, the duration of each phase could vary considerably. To evaluate outcomes such as mortality, relapse and response rates, a long‐term follow up of at least five years would be necessary. Such long follow ups could easily be achieved since most paediatric oncology centres do follow most of the patients for more than five years. 
 
 In addition, the safety of CSF after long‐term use still needs to be determined. Inukai et al described the finding of receptors to G‐CSF, GM‐CSF and IL‐3 (Inukai 1998) on the membrane of leukaemic lymphoblasts. Some initial molecular studies found growth of ALL blast cell colonies after exposure to G‐CSF(Benko 2001), GM‐CSF and other cytokines (p. ex interleukin)(Gattei 1997) in in vitro assay. To verify the possibility of clinical consequences of these findings, well‐designed RCTs with long‐term follow up are needed to evaluate if frequent use of CSF during myelosuppressive therapy can influence the prognosis of ALL regarding response, relapse and overall mortality rates. Considering that most ALL chemotherapy regimens are highly intense and toxic to the bone marrow from the outset, it would be necessary to have studies with prophylactic use of CSF during entire intensive phases of chemotherapy and long‐term follow‐up to determine the most important outcomes regarding effectiveness and safety. 
 
 It is also useful to discuss some characteristics of the design of the included studies. Clark et al performed a cross‐over trial (Clarke 1999) which is known to be subject to potential risk of bias, particularly the 'carry‐over' effect that would underestimate the overall effect. To reduce the risk of bias, we used the common strategy of focusing the analysis on the first period only. Additionally, only one trial described a clear and adequate allocation concealment (Calderwood 1994), which is an important measure to prevent bias. Another study had different baseline characteristics between the treatment and control groups; this is the result of inadequate randomisation. In the study by Michel et al, more children with t(4;11) cytogenetic feature, with consequent higher mean white blood cell counts and a higher incidence of infants less than one year old were randomised to the non‐G‐CSF group (Michel 2000). Another factor which has an impact on the quality of studies is the pharmaceutical company funding of three RCTs (Calderwood 1994; Clarke 1999; Welte 1996). Pharmaceutical company funding is empirically linked to potential bias (Egger 2001), although many developed and developing countries still depend on this source of funding to promote research (Montaner 2001). There is a need to avoid pharmaceutical funding as much as possible to minimise results bias and publication bias (Lechxin 2003). 
 
 The meta‐analysis of data derived from two RCTs (Pui 1997; Welte 1996) showed statistical difference in favour of CSF use to prevent febrile neutropenia in children undergoing ALL treatment. Substantial heterogeneity was also detected through the chi² test (P = 0.03). We also found this to be the case when we quantified inconsistency across studies (I² = 79%). The results of this analysis should therefore be interpreted with caution and not considered definitive. The inclusion of data from other RCTs could reduce the statistical heterogeneity between the trials and confirm these results. There is a clear consensus among specialists that the major intention of the use of CSF is the prevention of febrile neutropenia. However, most of the included studies did not evaluate or report adequately on this important outcome. 
 
 Another important purpose of the addition of CSF to chemotherapy of paediatric ALL would be a reduction in the duration of neutropenia. The meta‐analysis of three studies (Calderwood 1994; Dibenedetto 1995; Michel 2000) showed that the length of episodes of neutropenia could be reduced by CSF use but this is mainly due to a heterogeneity coming from Michel 2000. As we analysed the quality of this trial, we found that the comparison groups were unbalanced and considered this the cause of the heterogeneity we detected. After analysis excluding Michel et al, both heterogeneity and the effect of reducing the duration of neutropenia disappeared. 
 
 On average the use of CSF reduced the length of hospital stay significantly by one day (mean of five days in the CSF group versus six days in the control group). To judge the economic impact of the results, a cost‐effectiveness analysis separated for different economic environments would be necessary (Elting 2002). ALL treatment leads to high frequency and prolonged hospitalisation (Aquino 1997; Bash 1994; Welte 1996) and a one‐day reduction in hospitalisation may be of no consequence when it comes to cost and quality of life. 
 
 When a subsequent chemotherapy course starts after the scheduled day due to complications from treatment, we defined this as a delay. We assessed a possible difference in the results by the fact that this meta‐analysis included data from one study (Clarke 1999) that defined neutropenia as an ANC of less than 1000/mm³, while the others considered neutropenia to be an ANC of less than 500/mm³. This was the only outcome available from Clarke et al for this systematic review. The small number of participants and the statistical homogeneity found between trials (I²=0%) prompted us to consider the inclusion of this trial in the analysis even though the definition of neutropenia was different. 
 
 Chemotherapy regimens based on multiple drugs are often administered with the aim of enhancing dose intensity (Michel 2000). Delays in the administration of scheduled chemotherapy may reduce the overall dose intensity which in turn may compromise long‐term results (Pui 2001). In this review, the meta‐analysis of four trials showed that there is no evidence that the addition of CSF in chemotherapy regimens in ALL diminishes treatment delay. However, the short duration of studies makes the long‐term effect uncertain. 
 
 In the present review the addition of CSF after chemotherapy courses significantly reduced episodes of infections in the group treated with CSF (Rate Ratio = 0.56, 95% CI 0.39 to 0.80, P = 0.002). As with the febrile neutropenia episodes outcome, we applied some indirect calculation to evaluate the total cycles of chemotherapy in each arm. The result must therefore be interpreted with caution. As severe neutropenia with concomitant infection is life‐threatening, the evidence of a lower rate of infection with the use of CSF raises hopes of a reduction in overall mortality rates. Only studies designed with long‐term follow up will allow such evaluation. In addition, infection leads to long hospital stays, use of broad‐spectrum antibiotics, laboratory and radiologic exams and consequently, higher treatment costs. High‐quality studies in the future could provide data for cost effectiveness studies. 
 
 A primary endpoint of this systematic review was death (mortality). However, it was not possible to combine the data in a meta‐analysis. Because of the short follow‐up times in the studies analysed in this review, the mortality was low as discussed above. Longer follow‐up periods could determine differences in remission, relapse and mortality rates. 
 
 Adverse effects of CSF use seem to be more frequent and severe in adult populations (Biesma 1990; Garcia‐Carbonero 2001; Maher 1994; Mayordomo 1995; Riikonen 1994). RCTs in paediatric populations do not demonstrate great concern about adverse effects of CSF administration. Safety could be better determined after longer follow up and administration of CSF in future RCTs. 
 
 The small number of included studies and great differences in the reporting of results between them prevented more meaningful subgroup analysis. We did not perform subgroup analysis comparing G‐CSF and GM‐CSF as only one included study evaluated GM‐CSF (Calderwood 1994). Even though some RCTs do not show differences in effectiveness and safety between G‐CSF and GM‐CSF in various clinical situations (Alvarado 1999; Beveridge 1998), some others report that G‐CSF has greater benefit compared to GM‐CSF (Palmeri 1999; Pierelli 2001Weaver 2001). This could explain the widespread tendency to use G‐CSF in clinical practice and research. 
 
 Another subgroup we considered is that of primary and secondary prophylaxis studies. Dibenedetto et al (Dibenedetto 1995) was the only included study that used CSF as secondary prevention. We analysed data from this study for most of the outcomes and identified no heterogeneity. It would be interesting to see if new RCTs evaluating the use of CSF as secondary prophylaxis are planned, as this approach could lead to different outcomes. 
 
 The empirical doses of 5 mcg/kg body weight per day or 10 mcg/kg body weight per day of G‐CSF used in the included studies warrant additional studies. Different doses could have a bearing on all outcomes evaluated in the studies, particularly if the optimal dose is yet to be found. The importance of trials to establish the optimal dose of CSF in children cannot be overstated. 
 
 Currently there is no evidence to support the routine use of CSF in children with ALL undergoing myelosuppressive chemotherapy. In developing countries, where medication costs are greater than hospitalisation costs, the routine use of CSF may not financially viable (Bennett 2000; Elting 2002). 
 
 This systematic review is based on short‐duration RCTs without consensus about CSF dose in children. Future RCTs should focus on long‐term follow up and determination of ideal CSF dosage in order to evaluate the real role of CSF to prevent febrile neutropenia and improve outcomes in paediatric patients with ALL.

Authors' conclusions

Implications for practice.

There is evidence that the prophylactic administration of CSF reduces length of hospitalisation and rate of infection during treatment. There is no evidence that the use of CSF reduces the length of neutropenia episodes or diminishes delay of chemotherapy courses in paediatric patients with ALL undergoing myelosuppressive chemotherapy. Although there were statistically significant fewer febrile neutropenia episodes in the CSF group, substantial heterogeneity between trials prevents us from drawing conclusions, and no useful information is available on survival.

Implications for research.

Future RCTs should focus on determination of optimal CSF dosage as well as the use of CSF during all cycles of the most intensive phases of treatment and long‐term follow up in order to determine its real role in the prevention of febrile neutropenia in paediatric patients with ALL undergoing myelosuppressive chemotherapy. There must be an effort to conduct high‐quality trials with proper evaluation and reporting of the main outcomes including mortality. Cost‐effectiveness analysis of the implication of routine use of CSF in ALL therapy should be performed with consideration of different economic scenarios.

What's new

Date	Event	Description
13 March 2012	Amended	Additional tables linked to text.

History

Protocol first published: Issue 2, 2003
 Review first published: Issue 3, 2005

Date	Event	Description
14 October 2008	New search has been performed	Review updated with no change in conclusions.
18 September 2008	New search has been performed	New search for clinical trials completed.
15 August 2008	Amended	Converted to new review format.

Appendices

Appendix 1. Search strategy

#1 Methodological search strategies 
 #2 explode COLONY‐STIMULATING‐FACTORS / all subheadings 
 #3 CSF 
 #4 #2 or #3 
 #5 explode FEVER / all subheadings 
 #6 FEVER* or FEBR* 
 #7 #5 or #6 
 #8 #4 and #7 
 #9 #1 and #8

Data and analyses

Comparison 1. Febrile neutropenia episodes.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Febrile neutropenia episodes	2		Rate Ratio (Fixed, 95% CI)	0.63 [0.46, 0.85]

1.1. Analysis.

Comparison 1 Febrile neutropenia episodes, Outcome 1 Febrile neutropenia episodes.

Comparison 2. Length of neutropenia.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Length of neutropenia	3	134	Mean Difference (IV, Fixed, 95% CI)	‐3.44 [‐4.76, ‐2.12]
2 Length of neutropenia excluding Michel 2000	2	67	Mean Difference (IV, Fixed, 95% CI)	‐1.11 [‐3.55, 1.32]

2.1. Analysis.

Comparison 2 Length of neutropenia, Outcome 1 Length of neutropenia.

2.2. Analysis.

Comparison 2 Length of neutropenia, Outcome 2 Length of neutropenia excluding Michel 2000.

Comparison 3. Length of hospitalisation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Length of hospitalisation	3	134	Mean Difference (IV, Fixed, 95% CI)	‐1.58 [‐3.00, ‐0.15]

3.1. Analysis.

Comparison 3 Length of hospitalisation, Outcome 1 Length of hospitalisation.

Comparison 4. Delays in chemotherapy courses.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Number of delays in chemotherapy	4		Rate Ratio (Fixed, 95% CI)	0.75 [0.47, 1.20]

4.1. Analysis.

Comparison 4 Delays in chemotherapy courses, Outcome 1 Number of delays in chemotherapy.

Comparison 5. Number of infectious disease episodes.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Number of infectious disease episodes	5		Rate Ratio (Fixed, 95% CI)	0.56 [0.39, 0.80]

5.1. Analysis.

Comparison 5 Number of infectious disease episodes, Outcome 1 Number of infectious disease episodes.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Calderwood 1994.

Methods	Randomised Blind: yes Withdrawals: 5 (4 of treatment group and 1 in placebo group) Size: not pre‐determined; 20 individuals for each arm (N=40) ITT: no Placebo: yes Multicentre: no Funding: pharmaceutical
Participants	Children with ALL; ANC < 0.5 X 109/l Time period: November 1989 through September 1992
Interventions	GM‐CSF (Sandoz) 5.5mcg/kg SC
Outcomes	Overall mortality, time of neutropenia, time of hospitalisation, infection episodes, delays of chemotherapy, side effects and rate of remission.
Notes	2 cycles CSF co‐administered with chemo
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Low risk	A ‐ Adequate

Clarke 1999.

Methods	Randomised Blind: no Withdrawals: 1 patient did not receive G‐CSF Size: not determined previously. N = 17 (8 randomised to receive G‐CSF after first block and 9 following the second block) ITT: yes Placebo: no Multicentre: no Funding: pharmaceutical
Participants	Children with ALL; ANC < 1 X 109/l Time period: January 1995 through April 1996
Interventions	G‐CSF (filgrastim) 5mcg/kg SC
Outcomes	Delays of chemotherapy
Notes	1 cycle Cross‐over study: patients received G‐CSF starting 4 days following completion of either the first or the second block of intensification therapy CSF in interval of chemo
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Unclear risk	B ‐ Unclear

Dibenedetto 1995.

Methods	Randomised Blind: no Withdrawals: no Size: N = 32 (14 received G‐CSF and 18 did not) ITT: no Placebo: no Multicentre: no Funding: unclear
Participants	Children with ALL; ANC < 0.5 X 109/l Time period: March 1991 through November 1993
Interventions	G‐CSF 10mcg/kg SC
Outcomes	Episodes of febrile neutropenia, time of neutropenia, time of hospitalisation, infection episodes, delays of chemotherapy and side effects (bone pain).
Notes	1 cycle CSF in interval of chemo
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Unclear risk	B ‐ Unclear

Michel 2000.

Methods	Randomised Blind: no Withdrawals: no Size: N = 67 (34 with G‐CSF and 33 without) ITT: no Placebo: no Funding: academic
Participants	Children with ALL; ANC < 0.5 X 109/l Time period: June 1993 through January 1998
Interventions	G‐CSF 5mcg/kg SC
Outcomes	Length of febrile neutropenia, chemotherapy delays, incidence of infections
Notes	6 cycles CSF in interval of chemo
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Unclear risk	B ‐ Unclear

Pui 1997.

Methods	Randomised Blind: yes Withdrawals: 16 because received parenteral antibiotics before G‐CSF was scheduled to begin Size: N = 164, but only 148 evaluated (73 in G‐CSF group and 75 in placebo group) ITT: no Placebo: yes Multicentre: no Funding: unclear
Participants	Children with ALL; ANC < 0.5 X 109/l Time period: December 1991 through August 1994
Interventions	G‐CSF 10mcg/kg SC
Outcomes	Episodes of febrile neutropenia and infection episodes
Notes	1 cycle CSF in interval of chemo
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Unclear risk	B ‐ Unclear

Welte 1996.

Methods	Randomised Blind: no Withdrawals: no Size: N = 34 (17 to each arm) ITT: yes Placebo: no Multicenter: yes Funding: pharmaceutical
Participants	Children with ALL; ANC < 0.5 X 109/l Time period: January 1991 through December 1992
Interventions	G‐CSF 5mcg/kg SC
Outcomes	Episodes of febrile neutropenia, infection and delays of chemotherapy
Notes	>9 cycles evaluation
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Unclear risk	B ‐ Unclear

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Chen 1998	Not randomised
Heath 2003	Including patients older than 18 years of age (up to 21 years)
Laver 1998	Included older patients (22 years) and reported insufficient data to perform meta‐analysis
Little 2002	Cross‐over study without separate analysis of the period before and after cross‐over
Ohno 1993	Inclusion of adult subjects and data not extractable for children
Saarinen‐Pihkala2000	Not randomised Use GM‐CSF after ANC < 0.5 x 109/L (treatment of neutropenia)

Contributions of authors

• ECS wrote the protocol and performed the search for articles, selected articles, extracted and analysed data and wrote the manuscript.

• OC contributed to the protocol elaboration, performed the manual search for articles, selected articles, extracted and analysed data.

• ADS contributed to the protocol elaboration, performed the manual search for articles, selected articles, extracted and analysed data and also served as a third opinion to solve disagreements.

• SRB conceptualised the study and contributed to the protocol elaboration. She contacted the authors of selected articles and wrote the manuscript.

• SR extracted and analysed data and amended the review text.

All authors contributed to the analysis and interpretation of data and results.

Declarations of interest

None known.

Edited (no change to conclusions)