Performance of Model-based Network Meta-Analysis of Time-Course Relationships for Continuous Outcomes: A simulation study

Research output: Contribution to conferenceConference Abstractpeer-review

Abstract

Context: Time-course Model-Based Network Meta-Analysis (MBNMA) is a new framework for evidence synthesis of multiple treatments when studies report outcomes at multiple follow-up times. Time-course MBNMA pools relative effects on time-course function model parameters. However, the performance of the method is unknown, and may depend on model specification, model selection, and the number, timing and location of observations in the network. We aim to explore these aspects of model performance using simulation.
Methods: Data from two-arm RCTs forming a 3-treatment network with repeated observations were simulated from an Emax time-course function. The number and location of observations was varied between generated datasets, as was the covariance structure, correlation and between-study heterogeneity. Several time-course MBNMA models were fitted and different model selection strategies compared.
Results: A staged model selection strategy successfully identified a well-performing model, even in the presence of high correlation between observations. Bias was low for both time-course parameters (range: -0.3 to 5.6%) and predicted responses at different follow-up times (range: -0.01 to 0.44%). Time-course parameter estimates were more biased when very limited direct evidence was synthesised with indirect evidence, although predictions remained stable. Error in the precision of estimates was low in all datasets for selected models but substantially increased if correlation/heterogeneity was not properly accounted for.
Conclusions: Time-course MBNMA performs well and is statistically robust under a range of different dataset characteristics. The validity of conclusions drawn from time-course MBNMAs should be considered in light of the number/location of observations in the data, and the purpose for which the model will be used (i.e. whether time-course parameters/predictions are of most interest). This framework provides a method for the inclusion of multiple time points in meta-analysis, whilst delivering the statistical rigour required in the reimbursement decision-making process, thereby presenting an opportunity for the incorporation of early-phase clinical trials into HTA.
Original languageEnglish
Publication statusUnpublished - Jul 2019
Event40th Annual Conference of the International Society for Clinical Biostatistics - KU Leuven, Leuven, Belgium
Duration: 14 Jul 201918 Jul 2019
Conference number: 40
http://www.iscb2019.info

Conference

Conference40th Annual Conference of the International Society for Clinical Biostatistics
Abbreviated titleISCB
CountryBelgium
CityLeuven
Period14/07/1918/07/19
Internet address

Fingerprint

Dive into the research topics of 'Performance of Model-based Network Meta-Analysis of Time-Course Relationships for Continuous Outcomes: A simulation study'. Together they form a unique fingerprint.

Cite this