Skip to main content


Searching for indirect evidence: advantages and challenges of extending the network of studies

Date and Location


Presenting author and contact person

Presenting author

Pascale Dequen

Contact person

Pascale Dequen
Abstract text
Background: Methodological guidelines for conducting network meta-analysis (NMA) suggest that an iterative search of the literature, progressively including additional comparator treatments, may improve the identification of indirect evidence to be included for analysis (Hoaglin 2011). Objectives: Using the case study of apixaban (Eliquis®), we explore the potential advantages of searching for indirect evidence and evaluate the impact of study identification methods and network size on indirect and mixed treatment comparisons in venous thromboembolism (VTE) prophylaxis following major orthopaedic surgery. We also compare our methodology and summary findings to the previously published example by Hawkins et al. (2009a). Methods: We adapted the breadth-first search strategy from Hawkins et al. (2009b) to perform a stepwise systematic search in Medline®, Medline-in-Process®, and EMBASE for indirect evidence in VTE prevention. Table 1 details the multiple search orders included. Fixed-effects NMA models were run in WinBUGS (Lunn 2000) for clinical outcomes of interest for three different network sizes. Results: Figure 1 summarises the search results. Additional search orders and wider selection criteria maximised the number of indirect comparisons identified between existing VTE interventions. The NMA showed precision was increased from base case to first order as additional studies generally reduced the uncertainty around mean odds ratios for deep vein thrombosis, all VTE/death, and all bleeds. Estimates became more stable as fewer studies were included in the networks with each subsequent search order. No increase in inconsistency was noted across network sizes. Conclusions: Using a search strategy designed to optimise the identification of indirect evidence allowed us to extend the network of relevant studies for analysis. However, we find the methodology to be highly dependent on the definition of comparators in the first search order and suggest that the incremental value of subsequent network orders in NMA should be weighed against the associated additional search and analytical burden.