Methods for ranking competing treatments in network meta-analysis.

Chiocchia, Virginia (2022). Methods for ranking competing treatments in network meta-analysis. (Unpublished). (Dissertation, University of Bern, the Faculty of Medicine and the Faculty of Human Sciences)

Full text not available from this repository.

Background
One of the main objectives of comparative effectiveness research is to identify the most preferable treatments for a specific condition. Network meta-analysis (NMA) has been increasingly used for this purpose as it enables synthesis of data about competing interventions compared directly and indirectly in many studies, that form a network of evidence. NMA is used to estimate the relative treatment effect of each intervention versus all the others, and can produce statistical ranking metrics that lead to a treatment hierarchy from the least preferable to the most preferable option. Treatment hierarchies have been increasingly reported in published NMAs, and their use and reporting are recommended by international guidance and reporting guidelines. However, several methodological issues have been debated. For instance, the agreement between hierarchies from different ranking metrics has not been explored empirically. Methods to rank treatments for multiple outcomes, accounting for both efficacy and safety as well as individual preferences simultaneously, are underdeveloped. Also, it is unclear how to critically appraise a treatment hierarchy, since a rigorous framework to assess reporting bias in NMA is lacking.

Aim
The aim of this thesis is to fill some of these methodological gaps on the topic of ranking metrics and reporting bias in NMA. The first objective is to study the agreement between different rankings from an empirical perspective and to aid the interpretation and use of existing ranking metrics. The second objective is to extend the existing ranking methodology to account for multiple efficacy and safety outcomes, as well as specific preferences and trade-offs between benefits and harms. The third objective is to develop a methodological framework to evaluate the risk of reporting bias in network meta-analysis.

Methods
An empirical evaluation of the level of agreement between hierarchies obtained from existing ranking metrics is carried out by re-analysing over 200 previously published networks of four or more interventions. We explore how agreement is affected by the amount of information present in a network in terms of average variance, differences in the variance estimates, and the total sample size over the number of interventions of a network. To expand on the existing ranking methodology, we combine a recently developed ranking metric, accounting for both multiple outcomes and individual preferences, with a trade-off value defining the compromise between positive and negative outcomes. For evaluating the risk of reporting bias in NMA we combine the risk of bias due to missing evidence in pairwise comparisons with that of the network estimates. For the latter, we consider the contribution matrix, the unobserved comparisons, and the presence of small study effects as evaluated by network meta-regression. We also develop an online web application to facilitate this evaluation.

Results
The level of agreement between treatment hierarchies obtained by different ranking metrics can be affected by the amount of information present in a network. Differences in level of agreement become more evident when there are large imbalances in the precision of the estimates, though we find that such imbalances are rare in practice. We also developed rankings based on relative treatment effects against a fictional treatment of average performance, which are useful in networks of interventions where a natural reference treatment does not exist. We provide recommendations for reporting the treatment hierarchies obtained from different ranking metrics, avoiding misinterpretation, and properly addressing "treatment hierarchy questions" in the decision-making context. We extended the existing ranking methodology by combining the standardised area within spie charts with different trade-offs between benefits and harms. The obtained quantity is useful to show variation in the ranking for a whole range of trade-off values and a specific set of individual preferences. We developed the first risk of bias tool to evaluate the risk of bias due to missing evidence in NMA and we facilitate its use with a user-friendly web application that automates some of the required steps.

Conclusions
In this thesis we made significant contributions to the evidence synthesis field, providing knowledge and tools that assist clinicians, policy makers and patients in choosing the most preferable treatment for a specific condition. Our results are a step forward in the direction of actively translating knowledge into practical use, although more implementation research in clinical practice is still needed to guide decision-making processes.

Item Type:	Thesis (Dissertation)
Division/Institute:	04 Faculty of Medicine > Pre-clinic Human Medicine > Institute of Social and Preventive Medicine (ISPM)
Graduate School:	Graduate School for Health Sciences (GHS)
UniBE Contributor:	Chiocchia, Virginia, Salanti, Georgia
Subjects:	600 Technology > 610 Medicine & health 300 Social sciences, sociology & anthropology > 360 Social problems & social services
Language:	English
Submitter:	Doris Kopp Heim
Date Deposited:	23 Mar 2023 15:11
Last Modified:	23 Mar 2023 23:27
Additional Information:	PhD in Health Sciences (Biostatistics)
URI:	https://boris.unibe.ch/id/eprint/180563