← allmeta

Network Meta-Analysis

Frequentist network meta-analysis of study contrasts, as the R package netmeta: common- and random-effects (DerSimonian–Laird) estimates with multi-arm studies handled, league table, prediction intervals, heterogeneity and the design-based decomposition of Q, and P-scores. Checked against netmeta on every dataset it ships.

Test consistency before interpreting. The league table and P-scores assume the network is consistent (direct and indirect evidence agree). Run a node-split / design-by-treatment check in NMA Inconsistency first; if inconsistency is present, the pooled estimates and rankings can be misleading.

Contrasts

Data CSV: study, treatment1, treatment2, estimate, SE Each row is one pairwise contrast. Effect is treatment2 − treatment1 on the analysis scale (e.g. log OR). A multi-arm study needs a row for every pair of its arms. Example data (datasets shipped with the R package netmeta) Choose a dataset, then press Load example.

Options Reference treatment Set to the comparator for the league table; all effects are re-expressed relative to it. Model Random-effects confidence intervals Ranking: which direction is better? Show estimates Plot title

Network graph

Heterogeneity and inconsistency

Estimates versus the reference treatment

Ranking: P-scores

League table: row treatment versus column treatment

Network meta-analysis of study contrasts by generalised least squares, as the R package netmeta (Rücker 2012; Balduzzi et al. 2023): a multi-arm study's correlated contrasts are handled through its arm variances; random effects use a common τ² by the generalised DerSimonian–Laird estimator, with τ²/2 covariance between a multi-arm study's contrasts. Studies with an incomplete, duplicated or internally inconsistent set of comparisons are refused, as in netmeta. Prediction intervals use t with the degrees of freedom of Q (when there are at least 2). The decomposition of Q into within-design heterogeneity and between-design inconsistency follows Krahn et al. (2013). P-scores: Rücker & Schwarzer (2015). Checked against netmeta 3.7.0 on all its example datasets (hub/shared/tests/nma-netmeta-parity.spec.mjs). Node-splitting is in NMA Inconsistency.