Impact of antimicrobial de-escalation on mortality: a literature review of study methodology and recommendations for observational studies.

: The safety of de-escalation of empirical antimicrobial therapy is largely based on observational data, with many reporting protective effects on mortality. As there is no plausible biological explanation for this phenomenon, it is most probably caused by confounding by indication.: We evaluate the methodology used in observational studies on the effects of de-escalation of antimicrobial therapy on mortality. We extended the search for a recent systematic review and identified 52 observational studies. The heterogeneity in study populations was large. Only 19 (36.5%) studies adjusted for confounders and four (8%) adjusted for clinical stability during admission, all as a fixed variable. All studies had methodological limitations, most importantly the lack of adjustment for clinical stability, causing bias toward a protective effect.: The methodology used in studies evaluating the effects of de-escalation on mortality requires improvement. We depicted all potential confounders in a directed acyclic graph to illustrate all associations between exposure (de-escalation) and outcome (mortality). Clinical stability is an important confounder in this association and should be modeled as a time-varying variable. We recommend to include de-escalation as time-varying exposure and use inverse-probability-of-treatment weighted marginal structural models to properly adjust for time-varying confounders.