Granulocyte-colony stimulating factor (G-CSF) and granulocyte-macrophage colony stimulating factor (GM-CSF) for sepsis: a meta-analysis

Introduction To investigate the effects of G-CSF or GM-CSF therapy in non-neutropenic patients with sepsis. Methods A systematic literature search of Medline, Embase and Cochrane Central Register of Controlled Trials was conducted using specific search terms. A manual review of references was also performed. Eligible studies were randomized control trials (RCTs) that compared granulocyte-colony stimulating factor (G-CSF) or granulocyte-macrophage colony stimulating factor (GM-CSF) therapy with placebo for the treatment of sepsis in adults. Main outcome measures were all-cause mortality at 14 days and 28 days after initiation of G-CSF or GM-CSF therapy, in-hospital mortality, reversal rate from infection, and adverse events. Results Twelve RCTs with 2,380 patients were identified. In regard to 14-day mortality, a total of 9 death events occurred among 71 patients (12.7%) in the treatment group compared with 13 events among 67 patients (19.4%) in the placebo groups. Meta-analysis showed there was no significant difference in 28-day mortality when G-CSF or GM-CSF were compared with placebo (relative risks (RR) = 0.93, 95% confidence interval (CI): 0.79 to 1.11, P = 0.44; P for heterogeneity = 0.31, I2 = 15%). Compared with placebo, G-CSF or GM-CSF therapy did not significantly reduce in-hospital mortality (RR = 0.97, 95% CI: 0.69 to 1.36, P = 0.86; P for heterogeneity = 0.80, I2 = 0%). However, G-CSF or GM-CSF therapy significantly increased the reversal rate from infection (RR = 1.34, 95% CI: 1.11 to 1.62, P = 0.002; P for heterogeneity = 0.47, I2 = 0%). No significant difference was observed in adverse events between groups (RR = 0.93, 95% CI: 0.70 to 1.23, P = 0.62; P for heterogeneity = 0.03, I2 = 58%). Sensitivity analysis by excluding one trial did not significantly change the results of adverse events (RR = 1.05, 95% CI: 0.84 to 1.32, P = 0.44; P for heterogeneity = 0.17, I2 = 36%). Conclusions There is no current evidence supporting the routine use of G-CSF or GM-CSF in patients with sepsis. Large prospective multicenter clinical trials investigating monocytic HLA-DR (mHLA-DR)-guided G-CSF or GM-CSF therapy in patients with sepsis-associated immunosuppression are warranted.


Introduction
Despite improvements in antimicrobial therapy and supportive care, the incidence of sepsis continues to rise and sepsis is now the third leading cause of infectious deaths in the United States [1], with a mortality rate ranging from 20% for sepsis to 50% for septic shock [2,3]. During the past decades, many clinical trials testing anti-inflammatory therapies have been performed. However, the effects of these approaches on patient mortality were rather disappointing [4][5][6][7]. It is now generally agreed that patients with sepsis are more prone to die in a state of sepsis-induced immunosuppression, including reduced monocytic phagocytotic activity, changes in monocytic cytokine expression, diminished monocytic antigen presentation, lymphocytic dysfunction and apoptosis-induced loss of circulating T-and B-lymphocytes [4,[8][9][10][11][12][13][14][15][16][17][18][19][20][21]. Consequently, immunostimulatory therapies constitute an innovative strategy that deserves to be assessed for the treatment of sepsis [14,22,23].
To date, one approach is the use of granulocyte colony stimulating factor (G-CSF) or granulocyte-macrophage colony stimulating factor (GM-CSF), to augment myeloid cell functions in patients with sepsis. G-CSF, widely used in reducing the duration of febrile neutropenia following cytotoxic chemotherapy, has been shown to stimulate the production of neutrophils and modulate the function and activity of developing and mature neutrophils [24]. Compared to G-CSF, GM-CSF exhibits broader effects and induces proliferation and differentiation of neutrophils, monocytes, macrophages, and myeloid-derived dendritic cells. GM-CSF has been demonstrated to increase monocytic HLA-DR (mHLA-DR) expression and endotoxininduced proinflammatory cytokine production in ex vivo whole blood cultures of patients with severe sepsis [25,26].
So far, G-CSF and GM-CSF have shown promise in the treatment of infection in non-neutropenic hosts in many animal models [27][28][29][30]. Additionally, several clinical trials have been conducted to investigate the effect of G-CSF or GM-CSF treatment in neonates and adults with infection. Recently, a meta-analysis investigating the effect of G-CSF and GM-CSF for treating neonatal infection showed no significant reduction in 14-day mortality [31]. To our best knowledge, no previous systematic review had been conducted to define the efficacy and safety of G-CSF and GM-CSF in patients with sepsis. Therefore, we attempted to summarize the available randomized control trials (RCTs) to determine whether G-CSF or GM-CSF therapy significantly reduced all-cause mortality at 14 days and 28 days, inhospital mortality and occurrence of adverse events, and increased reversal rate from infection in patients with sepsis.

Materials and methods
We followed the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analyses for reporting our meta-analysis and results [32].

Search strategy
We searched Pubmed, Embase, and the Cochrane Central Register of Controlled Trials (to 25 October 2010) to identify potentially relevant trials. We restricted the search to trials on adults and used the search terms "granulocyte colony-stimulating factor", "granulocyte colony stimulating factor, recombinant", "GCSF", "filgrastim", "lenograstim", "sargramostim", "pegfilgrastim", "granulocyte-macrophage colony-stimulating factor", "granulocyte-macrophage colony stimulating factor, recombinant", "GMCSF", "molgramostim", AND "sepsis", "septicemia", "septicaemia", "septic shock". We restricted the findings of this search with a highly sensitive search strategy recommended by the Cochrane Collaboration for identifying all randomized controlled trials [33]. In addition, we checked the reference lists of identified trials and previous relevant meta-analyses identified by the electronic search to find other potentially eligible trials. There was no language restriction for the search. Authors of papers were contacted when results were unclear or when relevant data were not reported.

Study selection
We considered trials that investigated the therapeutic effects of G-CSF or GM-CSF administered intravenously or subcutaneously in adults with sepsis. Sepsis was defined according to the American College of Chest Physicians/Society of Critical Care Medicine consensus criteria [34] or was extrapolated to these criteria if not provided. Trials that allowed concurrent use of other therapies, including antibiotics, mechanical ventilation, steroids, bronchodilators and so on were included if they allowed equal access to such medications for patients in both arms of the trial. Randomized controlled trials specifically involving neutropenic patients or patients following chemotherapy were excluded. Agreement between reviewers regarding trial inclusion was assessed using the Cohen К statistic [35].

Data extraction
Full text versions of all eligible trials were obtained for quality assessment and data extraction independently by two reviewers. Extracted data were entered into Microsoft Excel 2007 and were checked by a third reviewer. Disagreement or doubt was resolved by discussion. Abstracted data included study design (for example, date of conduct and sample size), patient characteristics, study methodology (for example, eligibility criteria, method of randomization, and blinding), intervention (for example, G-CSF and GM-CSF dosage, duration and route of administration) and main outcomes. The quality of trials was assessed with the methods recommended by the Cochrane Collaboration for assessing risk of bias [36]. The criteria used for quality assessment were sequence generation of allocation, allocation concealment, blinding, selective outcome reporting and other sources of bias. Each criterion was categorized as 'yes', 'no', or 'unclear', and the summary assessments of the risk of bias for each important outcome within and across studies was categorized as 'low risk of bias', 'unclear risk of bias' and 'high risk of bias'.
The primary outcomes of this meta-analysis were allcause mortality at 14 days and 28 days after initiation of G-CSF or GM-CSF therapy. Secondary outcomes included in-hospital mortality, reversal rate from infection, and adverse events. The definition of reversal of infection referred to resolution of all signs, symptoms and laboratory assessment of infection or recovery from sepsis, which varied among trials due to different origins of sepsis. Adverse events were defined as organ dysfunction that was life threatening, required treatment and prolongation hospitalization, or was associated with death of the patient.

Statistical analysis
Analyses were on an intention-to-treat basis. We calculated a weighted treatment effect across trials using fixed-effect model. The results were expressed as relative risks (RRs) with 95% confidence intervals (CIs) for dichotomous outcomes. We considered using randomeffects model only in case of heterogeneity (P-value of χ 2 test less than 0.10 and I 2 greater than 50%). Potential sources of heterogeneity were identified by subgroup analysis on the basis of G-CSF and GM-CSF or/and by sensitivity analyses performed by omitting one study in each turn and investigating the influence of a single study on the overall meta-analysis estimate. Publication bias was assessed using funnel plots. A P-value of less than 0.05 was considered statistically significant. All statistical analyses were performed using Review Manager, version 5.0 (RevMan, The Cochrane Collaboration, Oxford, UK).
Patients' baseline characteristics in comparative groups were well balanced, such as the Acute Physiology and Chronic Health Evaluation (APACHE) II score. Only one trial investigating GM-CSF versus placebo had discrepant baseline characteristics, in which the mean ages were 62 and 46.5 years old, respectively [41]. Details of the included studies are summarized in Table 1.
Randomized allocation sequence was adequately generated in six trials [41][42][43][46][47][48], for the other six trials it was judged to be unclear based on the available documents [37][38][39][40]44,45]. Allocation sequences concealment was adequately reported in six trials [41,43,44,[46][47][48] and was judged to be unclear in the other six trials [37][38][39][40]42,45]. It was clearly stated that blinded fashion was conducted in all but one trial [44] and the outcome measurements were not likely to be influenced by lack of blinding. The numbers and reasons for withdrawal/ dropout were detailed reported in all but one trial [42]. None had stopped early due to data-dependent process or other problems, so free of other sources of bias were defined across trials. Therefore, five trials [41,43,[46][47][48] were determined as low risk of bias (plausible bias unlikely to seriously alter the results), and six trials [37][38][39][40]42,45] were at unclear risk of bias (plausible bias that rises up to some doubt about the results), while one trial [44] was at high risk of bias (plausible bias that seriously weakens confidence in the results). An overview of the quality appraisal was shown in Table 2. For the meta-analysis of G-CSF or GM-CSF therapy on 28day mortality, there was evidence of significant funnel plot asymmetry ( Figure 2).

All-cause mortality at 14 days
In regard to 14-day mortality, only four trials [40,[43][44][45] consisting of a total of 138 patients evaluated the shortterm outcome. The average sample size for one trial was 34 patients (sample sizes ranged from 18 to 58). Although one [43] of the four trials had calculated sample size according to 14-day mortality in the protocol, fewer patients were enrolled due to lower mortality and recruitment frequency than anticipated eventually. Hence, the meta-analysis of these four trials might be inappropriate to reveal the mortality benefit due to the limited numbers of patients. Moreover, none of the four trials reported a significant benefit in 14-day mortality following GCSF or GM-CSF administration. A total of 9 death events occurred among 71 patients (12.7%) in the treatment group compared with 13 events among 67 patients (19.4%) in the placebo group.

Adverse events
Overall, there were no significant differences in adverse events between treatment group and placebo group according to the data from seven trials (RR = 0.93, 95% CI: 0.70 to 1.23, P = 0.62) [37,38,40,42,43,45,47], with heterogeneity among the trials (P for heterogeneity = 0.03, On the basis of the results of the sensitivity analysis, one study was excluded [37]. Exclusion of the study did not significantly change the results of adverse events (RR = 1.05, 95% CI: 0.84 to 1.32, P = 0.44; P for heterogeneity = 0.17, I 2 = 36%; Figure 6).

Discussion
During the past few decades, there were a variety of trials investigating the effect of G-CSF or GM-CSF therapy in patients with sepsis. However, consistent results have not been reported and no individual study has definitively established whether G-CSF or GM-CSF bring clinically important benefits to septic patients. In the present meta-analysis, we found no significant differences in all-cause mortality at 14 days or 28 days, inhospital mortality, or adverse events between the G-CSF or GM-CSF group and placebo group in adult patients with sepsis. However, our result indicated that G-CSF or GM-CSF therapy was associated with a significant increase in reversal rate from infection. With respect to mortality, a previous meta-analysis suggested that addition of G-CSF or GM-CSF to antibiotic therapy in preterm infants with suspected systemic infection did not significantly reduce all cause mortality at 14 days or in-hospital mortality [31]. Recently, another meta-analysis showed that administration of G-CSF was not associated with improved 28-day mortality in adults with pneumonia [49]. Our meta-analysis, which included 12 relevant RCTs comparing G-CSF or GM-CSF with placebo, demonstrated that G-CSF or GM-CSF therapy did not significantly reduce all cause mortality at 14 days or 28 days or in-hospital mortality in patients with sepsis. However, a previous trial showed that receipt of G-CSF was associated with a longer duration of survival (P = 0.05) in severe septic patients due to melioidosis [46]. Meanwhile, another study demonstrated that GM-CSF significantly reduced the length of in-hospital stay in patients with nontraumatic abdominal sepsis (P <0.001) [45]. However, it should be noted that except one trial by Meisel et al. [48], none of the included trials in this meta-analysis were designed with patient stratification, whereas drug efficacy should be assessed only in patients with beforehand established impairment in monocytic functions [14]. Immunological biomarkers were thus needed that allow guidance of immunotherapy, risk stratification, and determination of which individuals might benefit from a given intervention [7]. As opposed to circulating cytokines, the major advantage of measuring mHLA-DR was that its level of expression was resultant of the sum of the effects of multiple mediators during septic shock [14]. At present, there was a general consensus that a diminished mHLA-DR expression may be a reliable marker for immunosuppression in critically ill patients [50,51]. Recently, Meisel and colleagues [48] explored the GM-CSF therapy versus placebo in sepsis, the first mHLA-DR-guided immunostimulatory treatment, indicating that administration of GM-CSF to patients in the immunosuppressive phase of sepsis reversed the characteristic monocyte deactivation as demonstrated by an increase in mHLA-DR levels and Toll-like receptor-4and Toll-like receptor-2-induced cytokine production, and reduced time of mechanical ventilation as well as length of hospital and intensive care unit stay. However, their study was not designed to explore the survival benefits of GM-CSF in sepsis so that it was insufficiently powered to evaluate mortality. Therefore, more RCTs with large number of patients to evaluate clinical parameters and mortality as primary endpoints were needed to investigate the effects of mHLA-DR-guided G-CSF or GM-CSF therapy in patients with sepsis-associated immunosuppression. Treatment with G-CSF or GM-CSF enhanced cellular functions that were critical in infectious diseases, such as neutrophil, monocyte, and macrophage activation and increased circulating white blood cells [52][53][54]. In a series of nine consecutive septic patients with immunosuppression, the administration of GM-CSF induced a sustained mHLA-DR recovery, which was accompanied by a restoration in ex vivo tumor necrosis factor (TNF)-α production after LPS challenge [55]. Moreover, in surgical patients, Schneider et al. observed that G-CSF-induced restoration of mHLA-DR was not only accompanied by an increased lymphocyte proliferation and Th1 cytokine production (interleukine (IL)-2 and interferon (IFN)-γ) in response to PHA but also by a better capacity to release inflammatory cytokines in a whole blood model after LPS challenge [56]. Theoretically, G-CSF and GM-CSF might benefit patients with sepsis-associated immunosuppression and result in markedly increase in reversal rate from infection. Recently, Orozco et al. demonstrated that addition of GM-CSF to the standard treatment of patients with nontraumatic abdominal sepsis reduced the rate of infectious complications, shorten the duration of antibiotic therapy and the length of hospital stay [45]. Rosen and colleagues [ [44] observed a higher leukocyte count, increased mHLA-DR, and better cure/improvement of infection in GM-CSF group. Results from our present meta-analysis were in line with the above results and revealed a significant increase in reversal rate from infection with G-CSF or GM-CSF therapy verse placebo (RR = 1.34, 95% CI: 1.11 to 1.62, P = 0.002). However, it did not bring a significant benefit in 28-day mortality (RR = 0.93, 95% CI: 0.79 to 1.11, P = 0.44). Due to the complexity of sepsis and associated complications, the severity of sepsis varied largely. Therefore, G-CSF or GM-CSF therapy might not be able to bring a significant mortality benefit if administrated regardless of the specific situation in individual patient, especially the immunological state of enrolled patients.
Administration of G-CSF or GM-CSF theoretically may increase the prevalence of adverse events which include immunologically mediated organ dysfunction, such as acute respiratory distress syndrome (ARDS), to which patients with severe sepsis are particularly prone [49]. However, Tanaka et al demonstrated that G-CSF caused leukocyte stiffness but attenuated inflammatory response without inducing lung injury in septic patients [39]. Another two studies showed that administration of G-CSF appeared to be safe in patients with pneumonia and severe sepsis [40,42]. Recently, Presneill et al. demonstrated that low-dose GM-CSF was associated with improved gas exchange without pulmonary neutrophil infiltration and was not associated with worsened acute respiratory distress syndrome or the multiple organ dysfunction syndromes in patients with sepsis-associated respiratory dysfunction [41]. Consistent with these studies, our meta-analysis suggested that G-CSF or GM-CSF therapy did not significantly increase the rate of adverse events (RR = 0.93, 95% CI: 0.70 to 1.23, P = 0.62; P for heterogeneity = 0.03, I 2 = 58%). Definitions of adverse events in included studies were heterogeneous and sometimes absent, which might bring about heterogeneity. Sensitivity analysis did not significantly change the results of adverse events (RR = 1.05, 95% CI: 0.84 to 1.32, P = 0.44; P for heterogeneity = 0.17, I 2 = 36%).
There are several limitations that need to be considered in the present study. First, the geographic regions covered in this meta-analysis included North America (United States, Mexico and Canada), Europe (Belgium, Germany and Spain), Asia (Japan and Thailand) and Oceania (Australia). Therefore, our results limited generalizability to other regions (for example, Africa and Latin America). Second, considerable heterogeneity existed in the type and dosage of G-CSF and GM-CSF. Different baseline characteristics such as age and APACHE II score might have affected the outcome of patients' response to medical management and might have produced possible clinical heterogeneity. Finally, the sample sizes were highly variable across trials and the trial enrolling the maximum number of patients [37] contained 42 times as many subjects as the two trials enrolling the minimum number of patients [40,41].

Conclusions
While this meta-analysis demonstrated that G-CSF or GM-CSF therapy significantly increased the reversal rate from infection, it was not associated with a significant reduction in all cause mortality at 14 days or 28 days, in-hospital mortality or adverse events in patients with sepsis. Therefore, our present meta-analysis did not suggest routine use of G-CSF or GM-CSF in patients with Figure 3 28-day mortality of G-CSF or GM-CSF therapy versus placebo. Fixed-effect model of risk ratio (95% confidence interval) of 28-day mortality associated with G-CSF or GM-CSF therapy compared with placebo. Figure 4 In-hospital mortality of G-CSF or GM-CSF therapy versus placebo. Fixed-effect model of risk ratio (95% confidence interval) of inhospital mortality associated with G-CSF or GM-CSF therapy compared with placebo.

Key messages
• The literature shows that G-CSF or GM-CSF result in no differential effectiveness in treating sepsis, in terms of all-cause mortality at 14 days or 28 days, in-hospital mortality or adverse events. However, G-CSF or GM-CSF therapy significantly increased the reversal rate from infection in patients with sepsis.
• There was a lack of consensus in the literature in regard to the effect of G-CSF or GM-CSF versus placebo in treating sepsis in adults.
• Larger prospective randomized controlled trials investigating monocytic HLA-DR (mHLA-DR)guided G-CSF or GM-CSF therapy in sepsis are warranted.