Author Affiliations: University of Louisville, Louisville, Kentucky (email@example.com).
To the Editor: Dr Ranpura and colleagues presented a meta-analysis of 16 randomized controlled trials to investigate the occurrence of FAEs with bevacizumab in the treatment of patients with cancer.1 There are several problems with the study. First, the model used is not clearly specified, and the input variables are not identified. The study only discussed the use of random effects vs fixed effects given the outcome of a test of heterogeneity. Second, subgroup analysis was used, which is often lacking in power. Third, and perhaps most important, the study does not indicate how the rare occurrence of mortality was managed in the model. A generalized linear model with the outcome of FAE or no FAE as a rare occurrence is required, and there should be a link function for a Poisson distribution.2
Cerrito PB. Bevacizumab and Cancer Treatment-Related Mortality. JAMA. 2011;305(22):2291-2293. doi:10.1001/jama.2011.759