Selection of right studies criteria to foster conclusions. Caveats for metanalysis conducted in sepsis.

Abstract

We have read with interest the recent manuscript published in Heart Lung journal by Guo et al. Guo et al conducted a systematic review and meta-analysis evaluating the impact of de-escalation therapy on clinical outcomes in patients with severe sepsis and/or septic shock.1 To date, de-escalation represents a very attractive strategy for antibiotic stewardship programs based on the high antibiotic consumption and development of antibiotic resistances. Numerous studies have been conducted to evaluate the clinical impact and safety of de-escalation therapy in patients with severe sepsis and/or septic shock. Whilst the author have performed a deep review in order to include all the originals studies to date, we have found that the studies search and inclusion criteria implemented for their study has a substantial bias within the current evidence available. Indeed, inclusion of studies should be based on their design. As the authors reported, the majority of studies come from retrospective design publications and abstracts. Only well-designed prospectively randomized trials should be included for a well design systematic review and meta-analysis, otherwise the results might be invalidated. There are two studies that should have only be included such as Leone M et al2 and Garnacho-Montero et al3, because of their randomized design in the first and the prospective design in the second. In this study, de-escalation of empirical therapy was found to be feasible in critical care settings and even as a protective factor for mortality in patients with severe sepsis and septic shock as Garnacho et al could demonstrate. It could be possible, and as acknowledged by the authors, due to the study design, that causal inference remains somewhat unclear. However, the authors used a statistical approach that includes multivariate regression and propensity score analysis, which minimizes far the residual confounding, although it cannot be absolutely excluded