Skip to main content


How to design future studies based on conditional power of network meta-analysis

Date and Location




Monday 23 September 2013 - 13:30 - 15:00


Presenting author and contact person

Presenting author

Georgia Salanti

Contact person

Georgia Salanti
Abstract text
Background: It has been suggested that future studies should not be considered in isolation but designed with an aim to inform a meta-analysis. However, a trial between the interventions of interest (say AvsB) might be expensive or difficult and investigators are interested to know whether another comparison (e.g. BvsC) could add power to the evaluation of AvsB when included in an ABC network of studies. Objectives: To provide guidance for the design of futures study based on the results of Network Meta-Analysis (NMA): which treatments to compare, in how many studies and with what sample size in order to achieve a pre-defined level of power for a specific comparison. Methods: We extend the idea of Conditional Power (CP) in the presence of indirect evidence and we derive a formula for the sample size under fixed and random effects. We illustrate the method in a network that compares resynchronization devices (RD), RD with defibrillation (RDD) and pharmaceutical therapy (P) for all-cause mortality. We plot CP versus sample size under various hypotheses for the ratio of direct to indirect evidence and the amount of heterogeneity. Results: We found that CP is highly dependent on the magnitude of heterogeneity and on the ratio of direct to indirect evidence. We demonstrate that future direct evidence increases CP more than indirect evidence in some, but not all cases, depending on the amount of indirect evidence already included in the network. In our example, to detect a 20% reduction in mortality in RDvsRDD with 80% power we need either 3300 patients in direct studies or 7070 patients distributed in indirect studies (4700 in RDDvsP and 2370 in RDvsP). Conclusions: CP based on updated NMA can help investigators planning a future study, decide which treatments to compare and with what sample size.