Reply to N. Lathia et al

Abstract

Lathia et al comment on our meta-analysis of granulocyte colony-stimulating factor (G-CSF) primary prophylaxis reported in Journal of Clinical Oncology in 2007, demonstrating significant reductions in infection-related mortality, all-cause early mortality, and febrile neutropenia in patients randomly assigned to receive primary G-CSF prophylaxis compared with those receiving secondary G-CSF prophylaxis or untreated controls. Among the 17 eligible randomized controlled trials, relative risk for febrile neutropenia was 0.54 (95% CI, 0.43 to 0.67) overall and 0.61 (95% CI, 0.53 to 0.72) for filgrastim. Despite the use of different statistical programs and exclusion of certain trials, Lathia et al arrived at estimates of treatment effect nearly identical to ours. To consider all relevant studies, we included the study by Zinzani et al, because it reported clinically relevant infections during the period of neutropenia, a better approximation for febrile neutropenia than the alternative of including all neutropenic events. As Lathia et al confirm, removal of the Zinzani et al study from the meta-analysis would not only reduce heterogeneity but also have no clinically meaningful impact on estimated treatment effect, nor would it alter any study conclusions. No significant heterogeneity was observed in filgrastim studies, regardless of the inclusion of the study by Zinzani et al. In addition, because Zinzani et al reported no meaningful beneficial effect of G-CSF on early mortality or infection-related mortality, inclusion of this study in the meta-analysis is the more conservative approach. For each trial using a 2 2 factorial design, rather than assessing the visual effect of point estimates as Lathia et al suggest, we formally assessed interaction between randomization strata. Statistically significant interaction between subgroups was determined by comparing the ratio of the difference in the natural logarithm of relative risks and the SE of the difference in log relative risks with the standard normal distribution. Significant interaction was observed for survival between chemotherapy and G-CSF assignments in the Osby et al study but not in the Fossa et al study. Therefore, separate comparisons of the results for the Osby et al study were included. Although we recognize that investigators may differ in their approaches to the analysis in this circumstance, no meaningful differences in effect estimates were observed with either approach. The minimal differences in estimated rates for the two studies noted by Lathia et al are most likely attributable to rounding differences, use of differing statistical programs, and varying presence of updated study information. Overall, it is gratifying that Lathia et al have essentially replicated and confirmed the results of our meta-analysis.