Connecting the dots: Linking disconnected networks using dose-response Model-Based Network Meta-Analysis

Research output: Contribution to conferenceConference Abstractpeer-review


Background: Network Meta-Analysis (NMA) is often used as a source of clinical evidence with which to make cost-effectiveness decisions in Health Technology Appraisals (HTA). However, NMA relies on a connected network of treatments, which may not be available. Comparisons between disconnected treatments are not possible using NMA without making strong, untestable assumptions. In some circumstances, modellinga parametric dose-response function using Model-Based NMA (MBNMA) has the potential to re-connect the network via the dose-response relationship enabling comparisons between otherwise disconnected treatments.
Objectives(s): To identify scenarios in which MBNMA allows evidence synthesis in disconnected networks that could otherwise not be obtained using standard NMA, and to evaluate the performance of the method.
Method(s): Using an example dataset of triptans for migraine relief which included 70 studies investigating 6 agents and placebo, we removed agents and doses to generate different scenarios that illustrated disconnectedness in all possible combinations of pairs of agents. To assess performance, we examined agreement between MBNMA estimates from the disconnected network and NMA estimates from an “augmented” network connected by adding agents/doses back into the dataset.
Results: Within augmented networks, estimates of relative efficacy were more precise from MBNMA than from NMA models (ratio of posterior SDs for NMA vs MBNMA: median=1.13; range=1.03-1.20). In disconnected networks MBNMA was able to provide estimates for all treatment comparisons where NMA could not. MBNMA relative effects in disconnected networks were entirely in agreement with NMA estimates in augmented networks for 15 out of 18 treatment comparisons.
Conclusions: By modelling the dose-response in a statistically robust manner, MBNMA can provide some increase in precision over NMA in networks that are already connected. In disconnected networks, it can allow for estimation of treatment effects that cannot be estimated using NMA. MBNMA relies on the dose-response relationship being characterised correctly, but this assumption can be tested providing sufficient evidence is available at different doses of each agent to allow reliable estimation. MBNMA enables earlier phase evidence to strengthen recommendations based on phase-II studies in HTA, and we argue that earlier phase evidence should be included in systematic reviews to support HTA.
Original languageEnglish
Publication statusUnpublished - 25 Aug 2020
Event41st Annual Conference of the International Society for Clinical Biostatistics - Online, Krakow, Poland
Duration: 23 Aug 202027 Aug 2020


Conference41st Annual Conference of the International Society for Clinical Biostatistics
Internet address


Dive into the research topics of 'Connecting the dots: Linking disconnected networks using dose-response Model-Based Network Meta-Analysis'. Together they form a unique fingerprint.

Cite this