A database to initiate methodological advances in the evaluation of transitivity assumption in network meta-analysis: qualitative features and limitations of the tracenma R package

• Published in: BMC medical research methodology Vol. 25; no. 1; pp. 183 - 13
• Main Authors: Spineli, Loukia M., García-Sierra, Andrés Mauricio, Yepes-Nuñez, Juan Jose
• Format: Journal Article
• Language: English
• Published: England BioMed Central Ltd 31.07.2025; BioMed Central; BMC
• Subjects: Analysis; Databases, Factual; Datasets; Decision theory; Humans; Meta-analysis; Methods; Network meta-analysis; Network Meta-Analysis as Topic; R package; Research Design; Software; Statistical analysis; Statistical methods; Statistical software; Systematic review; Transitivity assumption; United States; Database; Systematic review; Network meta-analysis; Transitivity assumption; R package
• ISSN: 1471-2288; 1471-2288
• DOI: 10.1186/s12874-025-02634-x

Transitivity assumption underlies the network meta-analysis framework and states that treatment comparisons are similar regarding the distribution of important characteristics that act as effect modifiers. Currently, there is a lack of methods to assess transitivity quantitatively. The methodological gap in assessing the transitivity assumption motivated the development of the tracenma database, aspiring to initiate methodological advances in transitivity assessment using empirical data. We used the nmadb R package to build a database of connected networks and consulted the relevant literature to determine the necessary characteristics comprising potential effect modifiers. We referred to the systematic review report and supplementary material for each eligible network to retrieve at least four extractable characteristics. The extracted information comprised the studies, corresponding treatment comparisons, and aggregated characteristics at the study level. We classified the characteristics into three types (clinical, demographic and methodological) and ten subtypes (participants, treatments, outcomes, age, sex, ethnicity, study design, study setting, risk of bias and withdrawals). We summarised the distribution of the characteristics and missing data across all datasets by type and subtype. Of the 453 networks in the nmadb database, 217 (48%) were eligible for the tracenma database, with each network comprising a dataset. The extracted characteristics ranged from 4 to 35 (median: 11), with the middle half comprising 8 to 15 characteristics. In most datasets, some characteristics were transformed for being reported inconsistently across the studies concerning the measure scale and summary statistics. Methodological characteristics dominated all datasets and were less prone to missing data. The study design was ubiquitous among the methodological subtypes, while withdrawals were hardly reported. Clinical characteristics were present in most datasets and were subject to most missing data; their characteristic subtypes received uneven attention, with participant characteristics reported more frequently. Demographic characteristics were the least frequently reported. Seventy per cent of the datasets included all types; however, the characteristic subtypes varied substantially in frequency. Tracenma is the first database built to motivate the development and empirical evaluation of novel methods for assessing transitivity. The database is hosted in the tracenma R package with functions that facilitate data access.