Efficacy and tolerability of granulocyte colony-stimulating factors in cancer patients after chemotherapy: A systematic review and Bayesian network meta-analysis

The optimum granulocyte colony-stimulating factor (G-CSF) treatment for cancer patients after being treated with cytotoxic chemotherapy remains unknown. Therefore, a systematic review and Bayesian network meta-analysis were performed to assess the efficacy and tolerability of 11 G-CSF drugs on patients after chemotherapy. A total of 73 randomized controlled trials (RCTs) containing 15,124 cancer patients were included for the final network meta-analysis. Compared with pegfilgrastim, there were a higher risk with filgrastim for incidence of febrile neutropenia (FN) (OR [95% CI]: 1.63 [1.07, 2.46]), and a higher risk with short-acting G-CSF (S-G-CSF) biosimilar and lenograstim for incidence of bone pain (BP) (OR [95% CI]: 6.45 [1.10, 65.73], 5.12 [1.14, 26.12], respectively). Mecapegfilgrastim, lipegfilgrastim and balugrastim were best G-CSF drugs in reducing FN (cumulative probabilities: 58%, 15%, 11%, respectively). S-G-CSF biosimilar, empegfilgrastim, and long-acting G-CSF (L-G-CSF) biosimilar were best G-CSF drugs in reducing severe neutropenia (SN) (cumulative probabilities: 21%, 20%, 15%, respectively). Mecapegfilgrastim, balugrastim, lipegfilgrastim and L-G-CSF biosimilar were best G-CSF drugs in reducing BP (cumulative probabilities: 20%, 14%, 8%, 8%, respectively). Mecapegfilgrastim, lipegfilgrastim and balugrastim might be the most appreciate G-CSF drugs with both good efficacy and tolerability when treating cancer patients after cytotoxic chemotherapy.

www.nature.com/scientificreports www.nature.com/scientificreports/ Efficacy and tolerability of G-CSF drugs from pair-wise meta-analysis. A traditional direct pair-wise meta-analysis was performed, as shown in . The heterogeneity of these meta-analyses was mostly low or moderate. In the meta-analysis of RCTs that compared pegfilgrastim with placebo, a high heterogeneity was observed with FN (I 2 = 89%), SN (I 2 = 91%), and BP (I 2 = 56%). This heterogeneity might have been introduced by the variation that resulted from the multiple types of tumors, since there were approximately five kinds of tumors in these seven trials. Since the sample size for every specific kind of tumor in these trials containing multiple types of tumors that was too small, it was difficult to implement an effective subgroup analysis. In the sensitivity analysis, no significant heterogeneity change was observed after removing studies from the analysis.
Efficacy and tolerability of G-CSF drugs from network meta-analysis. Figure 3 summarizes the results of the random-effects network meta-analysis for the efficacy of G-CSF drugs based on FN and SN and acceptability, in terms of BP. There was no direct comparison trial of pegteograstim (transverse line indicate no comparison in Fig. 3)    www.nature.com/scientificreports www.nature.com/scientificreports/ the difference was not statistically significant (95% CI contains 1), a reduction in the incidence of FN and SN was observed when empegfilgrastim, lenograstim, leridistim, mecapegfilgrastim, and pegteograstim were compared with placebo. The reason may be because the number of trials included was too small. In terms of the incidence of BP, S-G-CSF biosimilar and lenograstim significantly led to more than pegfilgrastim (OR [95% CI]: 6.45 [1.10-65.73]; 5.12 [1.14-26.12]). The incidence of BP by filgrastim, S-G-CSF biosimilar, lenograstim, and pegfilgrastim was significantly higher than placebo. However, there was no difference between other G-CSF drugs in the incidence of BP. By contrasting direct with indirect evidence using the node-split method, the network analysis did not reveal any statistical inconsistency with regards to FN, SN and BP.
Comparison of the possibility of efficacy and tolerability of G-CSF drugs. Figure 4 shows the distribution of possibility rank of the 12 treatments in terms of FN, SN, and BP. The higher the probability rank of the 12 treatment, the lower the probability of FN, SN and BP. Mecapegfilgrastim, lipegfilgrastim and balugrastim may be among the three best effective G-CSF drugs that could prevent the incidence of FN (cumulative probabilities: 58%, 15%, and 11%, respectively). S-G-CSF biosimilar, empegfilgrastim, and L-G-CSF biosimilar are possibly among the three more favorable G-CSF drugs that could prevent the occurrence of SN (cumulative probabilities: 21%, 20%, and 15%, respectively). Mecapegfilgrastim, balugrastim, lipegfilgrastim and L-G-CSF biosimilar were ranked as the lowest G-CSF drugs on incidence of BP (cumulative probabilities: 20%, 14%, 8%, and 8%, respectively).

Discussion
In the present network meta-analysis, the efficacy and tolerability of 11 different G-CSF drugs for cancer patients after chemotherapy in 73 RCTs containing 15,124 patients were evaluated using FN, SN and BP as indicators. It was found that pegfilgrastim was better than filgrastim in reducing FN, and more tolerable than S-G-CSF biosimilar and lenograstim in terms of the incidence of BP. In terms of both efficacy and tolerance, mecapegfilgrastim, lipegfilgrastim and balugrastim might be the most efficacious and tolerable among G-CSF drugs.
Since FN is the main and severe adverse event for many chemotherapy regimens, and is intimately associated with chemotherapy-related mortality 83 , FN was chosen as the primary outcome of the G-CSF drug treatment and a crucial indicator to evaluate the efficacy of G-CSF drugs. In the present study, it was found that compared with placebo, most of the G-CSF drugs could reduce the risk of the incidence of FN, except for empegfilgrastim,  www.nature.com/scientificreports www.nature.com/scientificreports/ leridistim, and pegteograstim. While leridistim might have an opposite effect, although the effect was not statistically significant. The network meta-analysis revealed that there was no difference or inferiority among the tested G-CSF drugs, except for filgrastim and pegfilgrastim in FN (filgrastim vs. pegfilgrastim OR [95% CI]: 1.63 [1.07-2.46]). Filgrastim, pegfilgrastim, lipegfilgrastim and lenograstim reduced the incidence of FN in cancer patients undergoing chemotherapy compared with placebo. Lipegfilgrastim appeared to lead to a greater reduction in the incidence of FN, when compared to pegfilgrastim and filgrastim, although the difference was not statistically significant. These findings were consistent with the previous observations 10, 84 . In accordance with previous reports, pegfilgrastim was more effective than filgrastim in reducing the incidence of FN 10,[85][86][87][88] . SN is also another important evaluation indicator of G-CSF drug efficacy. Filgrastim, pegfilgrastim, lipegfilgrastim, S-G-CSF biosimilar, mecapegfilgrastim, and lenograstim reduced the incidence of SN in patients undergoing myelosuppressive chemotherapy based on direct and indirect evidence. All these results indicate that compared with placebo, most of the tested G-CSF drugs were effective to prevent the incidence of FN and SN.
BP is one of the most common adverse events associated with G-CSF drug treatment 89 [1.10, 65.73)]) led to a much higher incidence of BP than pegfilgrastim. However, the level of incidence of BP widely varied among the RCTs of G-CSF drugs, which might have resulted from the differences in race of patients, stage and type of tumors, chemotherapy regimens, and definition of BP. These results suggest that patients might have different tolerances to different G-CSF drugs.
Even though there was no difference in efficacy among the tested G-CSF drugs and tolerability among patients to these G-CSF drugs in the pair-wise meta-analysis, the comparative ranking of these 12 G-CSF drug treatments suggest that mecapegfilgrastim, lipegfilgrastim and balugrastim might be more effective than leridistim, filgrastim and S-G-CSF biosimilar in preventing the incidence of FN, and S-G-CSF biosimilar, empegfilgrastim and L-G-CSF biosimilar might be more effective than filgrastim and pegfilgrastim in preventing the incidence of SN. In terms of BP, mecapegfilgrastim, balugrastim, lipegfilgrastim and L-G-CSF biosimilar might be more tolerable for patients, when compared to other G-CSF drugs. Those results indicate that mecapegfilgrastim, lipegfilgrastim and balugrastim might be the most efficacious and tolerable G-CSF drugs, and might provide a guideline for the selection of G-CSF-drugs for patients after chemotherapy.
Caution should be taken in interpreting the results, since there might be inconsistencies between the direct and indirect comparisons. These inconsistencies might have resulted from the different characteristics of trials, such as the study design, definition of indicators, inclusion criteria of subjects, and method of implementation, as well as the difference in identifying the external effect on the mean effect of the specific comparison between the network meta-analysis and pair-wise meta-analysis methods 90 . Although no inconsistency was found in FN, SN and BP through the node-split method in the main network analysis, the direct and indirect meta-analyses revealed contradictory results in terms of the comparisons between filgrastim vs. S-G-CSF biosimilar and filgrastim vs. L-G-CSF biosimilar. This mutually exclusive result could be explained as follows 90 : (1) if the direct evidence of the pair-wise meta-analysis was true, the comparison between other G-CSF drugs in indirect evidence of the network meta-analysis might overstate or understate the efficacy and tolerance; (2) if the indirect evidence was true, significant intrinsic heterogeneity might exist in the comparison among filgrastim, S-G-CSF biosimilar and L-G-CSF biosimilar. A low or moderate heterogeneity was observed in the pair-wise meta-analysis, indicating that the direct pair-wise meta-analysis was true.
Although the present study is the first network meta-analysis to comprehensively assess clinically and commonly used G-CSF drugs, it should be acknowledged that there were some limitations with the present analysis. First, many factors correlated with neutropenia after chemotherapy were not analyzed, such as the duration of neutropenia, duration of SN, depth of the absolute neutrophil count (ANC) nadir, time to recovery of ANC, FN-related hospitalization, and other toxic or side effects of G-CSF drugs. Second, in most of the included trials, the report for FN, SN and BP was incomplete, which caused some of the G-CSF drugs to be ruled out for comparison in terms of SN and BP. Third, trials on some G-CSF drugs were too few to be assessed. For example, merely one trial on mecapegfilgrastim has been reported to date. Fourth, the definition of BP and other indicators varied among these studies. Furthermore, the dose of G-CSF drugs also varied across the studies. These might be the source of heterogeneity and inconsistency. Fifth, the outcomes might only apply to developed countries, since some G-CSF drugs are not available on the market in many developing countries.
In summary, based on the present network meta-analysis, evidence suggests that compared with placebo, most of the tested G-CSF drugs are not different in terms of efficacy and tolerability, except for pegfilgrastim, which is more effective than filgrastim in reducing FN. Furthermore, pegfilgrastim is more tolerable for patients, when compared to S-G-CSF biosimilar and lenograstim, in terms of BP. Mecapegfilgrastim, lipegfilgrastim and balugrastim might be the most appreciate G-CSF drugs, which have both better efficacy and tolerance. It is noteworthy that more large-scale RCTs would be required to further confirm the efficacy and tolerance of the G-CSF drugs observed in the present study. The benefit-risk ratio of these G-CSF drugs still deserves to be further explored.
RCTs that compared at least two different G-CSF drugs (placebo-controlled included) in all kinds of cancer after chemotherapy were included. These trials should report the data on FN, SN, and/or BP in cancer patients after the use of G-CSF drugs. Non-randomized controlled trials, non-interventional studies, retrospective studies, or trials that contained only one treatment (single-arm) were excluded. Furthermore, studies that included healthy volunteers, but not cancer patients who received chemotherapy, were also excluded.
Study selection and data extraction. Study selection, data extraction and review, and quality assessment were independently performed by two authors (Y. Wang and L. Chen), according to the predefined criteria from eligible studies. The Cochrane Collaboration's tool for assessing risk of bias 92 was independently used for the quality assessment and evaluation of risk of bias by the same authors. The key characteristics of each study were recorded, which included: the first author's name and year of publication, country, study design, patient characteristics, chemotherapy regimens, dose and protocol of treatment, and outcomes (FN, SN and BP). All data for the study characteristics and clinical responses were summarized in a structured table to ensure consistency. All the disagreements were resolved by discussion and consensus with a third author (Y. Li).
Outcome measurements. The incidence of FN after cytotoxic chemotherapy within two weeks was taken as the primary indicator of efficacy of G-CSF drugs, the incidence of SN was taken as the secondary indicator of G-CSF drug efficacy, and BP was taken as the primary indicator for the tolerability of G-CSF drugs. FN was defined as an absolute neutrophil count (ANC) of <0.5 or 1.0 × 10 9 /L, with an oral temperature of ≥38.0 °C. SN was defined as ANC < 0.5 or 1.0 × 10 9 /L. If both data of both grade 3 and 4 bone marrow suppression (ANC < 0.5 and 1.0 × 10 9 /L) were reported in a study, the data of the ANC < 0.5 × 10 9 /L was used with priority for analysis, because grade 3 had lesser clinical significance, and was not always reported in the included studies.
Statistical analyses. Pair-wise meta-analysis was carried out for FN, SN and BP to compare the corresponding interventions. The random effects model for pair-wise meta-analysis was used to account for the heterogeneity. The heterogeneity among different trials was estimated by Cochran's Q-test (P < 0.05 indicated significant heterogeneity) and I 2 statistic. If I 2 = 0-25%, it is designated as low heterogeneity, if I 2 = 25-50%, this was designated as moderate heterogeneity, if I 2 = 50-75%, this was designated as high heterogeneity, and if I 2 = 75-100%, this was designated as extremely high heterogeneity. According to the Cochrane handbook, heterogeneity can be accepted when I 2 ≤ 50% 93 . Pair-wise meta-analysis was performed using Review Manager 5.3 (Cochrane Collaboration, Oxford, UK) and STATA 12.0 (Stata Corporation, TX, USA) statistical software.
Random-effects models were applied for the network meta-analysis. Bayesian network meta-analysis was used to combine the collected data. The Bayesian network meta-analysis was performed with WinBUGS version 1.4.3 (MRC Biostatistics Unit, Cambridge, UK). Random effects models were used to incorporate the effects from different studies, while heterogeneity within the comparison was evaluated in a relatively conservative and appropriate manner 94 . The models were performed using the Markov Chain Monte Carlo simulation. The initial values were set for three different chains, 150,000 interactions with 5,000 burn-in samples were produced to obtain the model parameters from the posterior distributions, and 50 thinning rates were adopted for each chain. The odds ratios (ORs) were collected or calculated from combing the direct evidence, and the significance was assessed by P < 0.05, or the 95% confidence interval (CI) did not contain 1. The best efficacious and tolerant regimen was confirmed by ranking the included G-CSF drugs according to the OR for each G-CSF drug compared with placebo, and assessing the probability. Inconsistencies in the present study were assessed by comparing the direct evidence with indirect evidence from the network meta-analysis using the node-split method 95 .
A sensitivity analysis was performed by determining whether there was statistically significant heterogeneity in the meta-analysis after studies were randomly removed from the others.