TRIAL SEQUENTIAL ANALYSIS: THE EVALUATION OF THE ROBUSTNESS OF META-ANALYSES FINDINGS AND THE NEED FOR FURTHER RESEARCH

Abstract

Randomized controlled trials (RCTs) may sometimes be underpowered to detect differences between interventions. Meta-analyses may reduce such risk by pooling together the available evidence from studies with similar inclusion criteria. However, this approach does not entirely rule out the risks of type-1 and type-2 statistical errors. Trial sequential analysis (TSA) is a statistical method developed for the assessment of the robustness of meta-analyses findings. Similar to the sample size calculation for a RCT, the calculation of the required “information size” for any TSA requires the outcome estimation (incidence or value) for the control and intervention groups, and the acceptable risks for type-1 and type-2 statistical errors (significance threshold and statistical power, respectively). Once performed, TSA graphically allows evaluation of whether the meta-analysis findings (cumulative effect size described in terms of Z-curve) are robust enough, or if there is need for further research. We briefly discuss the rationale and interpretation of TSA, identifying three main areas of the graph if the “information size” has not yet been reached. No further research is needed when the Z-curve crosses the adjusted significance thresholds or stands in the futility area. Conversely, if the Z-curve sits between adjusted significance thresholds and the futility boundary, more research is needed to establish differences between the interventions. In summary, similar to the interim analyses performed during any RCTs, the TSA graph can be helpful for scientists to understand if a meta-analysis finding is robust, and if further research would be desirable or futile on that topic.

Keywords

• rct
• statistical error
• systematic review
• futility