Automated methods to test connectedness and quantify indirectness of evidence in network meta-analysis

Area of Science:

• Statistics
• Network Meta-Analysis
• Graph Theory

Background:

• Network meta-analysis (NMA) integrates evidence from multiple studies to compare numerous treatments.
• Assessing the connectivity of evidence networks in NMA is crucial but challenging for complex networks.

Purpose of the Study:

• To introduce graph theory techniques for evaluating the connectedness of evidence networks in NMA.
• To provide methods for assessing the degree of indirectness in treatment comparisons within NMA.

Main Methods:

• Constructing an adjacency matrix representing direct comparisons between treatments.
• Deriving an indirect connection matrix to identify all possible direct or indirect comparisons.
• Calculating a distance matrix to quantify the minimum number of steps (assumptions) needed for indirect comparisons.

Main Results:

• The methods effectively determine if treatments can be compared directly or indirectly within the network.
• The distance matrix provides a quantitative measure of the assumptions required for indirect comparisons, considering network loops.
• Implemented methods are available in R code and a Stata package ('network').

Conclusions:

• Graph theory offers a fast and reliable approach to verify network connectivity in NMA.
• These techniques enhance the validity of NMA by ensuring comparisons are based on connected evidence and by assessing the level of indirectness.