Objectives: We aimed to identify all treatments that affect mortality in adult critically ill patients in multicenter randomized controlled trials. We also evaluated the methodological aspects of these studies, and we surveyed clinicians' opinion and usual practice for the selected interventions.
Data sources: MEDLINE/PubMed, Scopus, and Embase were searched. Further articles were suggested for inclusion from experts and cross-check of references.
Study selection: We selected the articles that fulfilled the following criteria: publication in a peer-reviewed journal; multicenter randomized controlled trial design; dealing with nonsurgical interventions in adult critically ill patients; and statistically significant effect in unadjusted landmark mortality. A consensus conference assessed all interventions and excluded those with lack of reproducibility, lack of generalizability, high probability of type I error, major baseline imbalances between intervention and control groups, major design flaws, contradiction by subsequent larger higher quality trials, modified intention to treat analysis, effect found only after adjustments, and lack of biological plausibility.
Data extraction: For all selected studies, we recorded the intervention and its comparator, the setting, the sample size, whether enrollment was completed or interrupted, the presence of blinding, the effect size, and the duration of follow-up.
Data synthesis: We found 15 interventions that affected mortality in 24 multicenter randomized controlled trials. Median sample size was small (199 patients) as was median centers number (10). Blinded trials enrolled significantly more patients and involved more centers. Multicenter randomized controlled trials showing harm also involved significantly more centers and more patients (p = 0.016 and p = 0.04, respectively). Five hundred fifty-five clinicians from 61 countries showed variable agreement on perceived validity of such interventions.
Conclusions: We identified 15 treatments that decreased/increased mortality in critically ill patients in 24 multicenter randomized controlled trials. However, design affected trial size and larger trials were more likely to show harm. Finally, clinicians view of such trials and their translation into practice varied.