Statistical analysis

The primary analysis investigated the incidence, relative risk and corresponding 95% CI of all-grade and high-grade hematologic toxicities in cancer patients treated by regorafenib. To calculate the incidence, the number of subjects receiving regorafenib alone and the number of subjects with hematologic toxicities (both all-grade and high-grade) were extracted from the eligible single-arm and randomized controlled trials. The proportion of patients with thrombocytopenia, anemia, neutropenia and leucopenia and 95% CIs were derived from every study. We calculated both RRs and CIs with data extracted only from randomized controlled trials, comparing the incidence of each adverse event in subjects assigned to regorafenib with subjects assigned to control treatment. To calculate 95% CIs, the variance of a log-transformed study specific RR was derived by the delta method. Statistical heterogeneity between different trials and subgroups was assessed by Cochrane’s Q statistic. The I² statistic was calculated to assess the extent of inconsistency contributable to the heterogeneity across different studies [49]. The assumption of homogeneity was considered invalid for I²> 25% or p<0.10. Summary RRs and incidences were calculated using fixed-effects or random-effects models depending on the heterogeneity of included trials. Potential publication bias was assessed by visual inspection of a funnel plot, and also evaluated using the tests of Egger et al. [50] and Begg et al. [51]. Two-sided p <0.05 were considered statistically significant. All analysis was performed using Stata version 12.0 (StataCorp, USA).