The original Rivers trial [1] of early goal-directed therapy (EGDT) discovered that a process of EGDT that guided fluids, vasopressors, inotropic agents and transfusions in the first 6 h of sepsis dramatically reduced mortality (relative risk reduction, RRR of 42 %) compared to usual care in an emergency department (ED) setting. That discovery ushered in an exciting era of early identification, resuscitation and monitoring of patients with sepsis. Since 2003, EGDT has been recommended in international sepsis guidelines [2][3][4], and implementation in routine care has been associated with benefits [5]. Likewise, many observers suggested that prompt recognition and treatment of sepsis may partly explain declining mortality from sepsis [6,7].Some have suggested that EGDT as described in Rivers' trial may not apply to the intensive care setting, because it was a single-centre trial conducted in the emergency department, the control mortality rate was higher than expected, and baseline central venous oxygen saturation (ScvO 2 ) was remarkably low (about 50 %)-an observation thought to be rare in septic ICU patients. Accordingly, many argued that there was clinical equipoise about the role of EGDT in ICU patients with sepsis, and several subsequent multicentre RCTs [8-10] did not find survival benefits from EGDT (mortality was actually slightly higher in EGDT than the usual care group in ProMISe).In this issue of Intensive Care Medicine, Peake and colleagues [11] report a systematic review with metaanalysis of five RCTs [1,[8][9][10][11][12] showing no difference in mortality between ED-initiated EGDT versus usual care groups. However, the ED-initiated EGDT group had significantly more use of vasopressors and duration of stay in the ICU. Peake and colleagues [11] also reviewed six RCTs of non-ED-initiated EGDT, again finding no difference in mortality between EGDT versus usual care. The systematic review adheres to current standards for the conduct and reporting of a systematic review [13,14], including a pre-experimentally published protocol, a comprehensive search strategy, adequate conventional cumulative meta-analyses, relevant sensitivity and subgroup analyses, and adequate assessment of risk of bias. Conventional cumulative meta-analyses are at risk of producing random errors because of sparse data and repetitive testing of accumulating data [15][16][17]. The risk of random errors can be assessed by trial sequential analysis (TSA), where trial sequential monitoring boundaries are applied to the conventional meta-analysis. The underlying assumption of TSA is that significance testing and calculation of the 95 % confidence interval are performed each time a new trial result is published or available. TSA depends on the accrued information