Is the minimum inhibitory concentration of vancomycin an infallible predictor of the clinical outcome of Staphylococcus aureus bacteremia treated with vancomycin?

other b-lactam antibiotic” was not all correct. Of the 4 subgroup analyses (ie, cefepime vs. ceftazidime, cefepime vs. piperacillin-tazobactam, cefepime vs. imipenem-meropenem, and cefepime vs. ceftriaxone-cefotaxime), only the cefepime versus piperacillin-tazobactam subgroup, which included 3 studies [5–7], was statistically significant. Cases of neurotoxicity, which included encephalopathy and nonconvulsive status epilepticus, were used to explain the increased mortality in the cefepime group [2]. We could not find any description of a suspected case of neurotoxicity in the studies analyzed that could have contributed to the death of a patient in the cefepime group. However, Biron et al [6] reported 6 patients from the cefepime group who died of extensive cancer. Chandrasekar and Arnow [9] reported that 24 (75%) of 32 patients who died did so 11 weeks after the completion of the study. Attributing the difference in the deaths of those patients with cancer to rare case reports of toxicity, which have been reported among patients receiving b-lactams and/or carbapenems [11–13], did not make much sense, especially when the reported rates of adverse effects were similar between the 2 treatment groups in many studies analyzed [3–9]. Yahav et al [2] attributed inadequate antimicrobial efficacy in vivo to the increase in mortality. Their meta-analysis [2] found no difference in clinical failure between treatment with cefepime and treatment with the comparator drugs. Although their analyses “did not reveal a specific cause for the increased mortality” [2, p. 344], they insisted on finding reasons to explain the statistically significant difference in the mortality rates without providing much clinical evidence. Therefore, we raise the same question that Machtay et al [14] did 10 years ago: is meta-analysis really meta-physics? Metaanalyses should not be considered a substitute for well-designed trials [14]. What Yahav et al [2] concluded should not be taken as the final word. Acknowledgments