Network meta-analysis

A network meta-analysis compares three or more treatments at once, including pairs that were never tested head to head. If some trials compared A with B and others compared B with C, the shared comparator B allows an indirect estimate of A against C. Bucher and colleagues formalised the adjusted indirect comparison in 1997, and Lumley named the method in 2002. Treatments form nodes, direct comparisons form edges, and the weight of an edge reflects how much evidence supports it. Where direct and indirect evidence both exist they are combined, and the difference between them is tested as inconsistency. Results are often ranked with SUCRA values or P-scores, though rankings from sparse networks are unstable and should not be read as a league table. The load-bearing assumption is transitivity: trials must be similar enough in patients, dose and outcome definition that the shared comparator means the same thing everywhere. When it fails, indirect estimates are biased and no statistic flags it. Reporting follows the PRISMA extension of 2015, and confidence in each estimate can be graded with CINeMA. In longevity research the format allows comparisons single trials avoid, such as the 2013 analysis of 305 trials and 339,274 participants that found no detectable mortality difference between exercise and drugs in secondary prevention of coronary heart disease.

Last reviewed: August 25, 2026

This definition is educational and is not medical advice, a diagnosis, or treatment. Talk to a doctor about any health decisions. Read our full medical disclaimer