• Susanne Unverzagt, Roland Prondzinsky, and Frank Peinemann.
• Institute of Medical Epidemiology, Biostatistics and Informatics, Martin Luther University of Halle-Wittenberg, 06097 Halle (Saale), Germany. Electronic address: susanne.unverzagt@medizin.uni-halle.de.
• J Clin Epidemiol. 2013 Nov 1;66(11):1271-80.

ObjectivesTo assess whether the reported trial characteristics are associated with treatment effects on all-cause mortality within critical care medicine.Study Design And SettingWe identified all eligible randomized controlled trials (RCTs) from Cochrane Reviews on patients with sepsis, septic shock, and cardiogenic shock. Risk of bias was judged on 12 trial characteristics, including the differentiation between single-center and multicenter trials. Hierarchical random-effects models quantified the impact of the risk of bias items on the reported effect estimates of mortality.ResultsTwelve meta-analyses that involved 82 RCTs were selected and judged. Single-center trials estimated a significant larger treatment effect compared with multicenter trials (ratio of odds ratios, 0.64; 95% confidence interval: 0.47, 0.87). Treatment effect tended to be overestimated with selective reporting of preplanned end points. Biases in different trial characteristics are unlikely to operate independently and may have modified these associations.ConclusionThe results of this study highlight a substantial difference in treatment effect estimates between single-center and multicenter trials. Therefore, we recommend that results from single-center trials should be cautiously used for decision making.Copyright © 2013 Elsevier Inc. All rights reserved.

23 keywords...

hide…