Article type
Year
Abstract
Background:
Network meta-analysis (NMA) allows synthesising the evidence simultaneously on multiple treatments. A key output of NMA is the relative ranking of the treatments; nevertheless, it has attracted a lot of criticism. This is mainly because ranking is a very influential output and, thus, prone to over-interpretations even when relative effects imply small differences between the alternative treatments. To date, common ranking methods rely on score metrics which are calculated based on the summary effects. Such metrics lack a straightforward interpretation, although it is still unclear how to measure their uncertainty.
Objectives:
To introduce a new modelling approach for networks of interventions that produces treatment hierarchies accounting for the clinical importance of the study-specific relative effects as well as for their uncertainty.
Methods:
We adapt methodology previously suggested for ranking in sports tournaments into the context of NMA. We use extensions of the so called ‘Bradley-Terry’ models, which are a family of probabilistic models that aim to predict the outcome of pairwise comparisons. We first translate the study-specific relative effects and their confidence intervals into wins, losses, and ties based on predefined minimally clinically important differences between treatments. Then, based on the number of wins, our model estimates the ‘worth’ of each treatment which is used as an intuitive measure to rank the treatments. This approach naturally captures the uncertainty of ranking because the estimates of treatment worth are accompanied by confidence intervals. The model also allows to consider simultaneously multiple outcomes in ranking by implementing a vector of several contrasts for each outcome. Finally, study precision and other important characteristics, such as risk of bias, can be incorporated by an additional parameter that represents the ‘importance’ of each study.
Results and Conclusions:
We illustrate our model using a Cochrane NMA comparing 7 treatment classes for chronic plaque psoriasis. Our model is able to clearly indicate the two classes with the largest worth but also classes with similar worth. The latter is a major strength of our approach as it preserves from exaggerating unimportant differences between treatments and drawing spurious conclusions.
Patient, public, and/or healthcare consumer involvement: None.
Network meta-analysis (NMA) allows synthesising the evidence simultaneously on multiple treatments. A key output of NMA is the relative ranking of the treatments; nevertheless, it has attracted a lot of criticism. This is mainly because ranking is a very influential output and, thus, prone to over-interpretations even when relative effects imply small differences between the alternative treatments. To date, common ranking methods rely on score metrics which are calculated based on the summary effects. Such metrics lack a straightforward interpretation, although it is still unclear how to measure their uncertainty.
Objectives:
To introduce a new modelling approach for networks of interventions that produces treatment hierarchies accounting for the clinical importance of the study-specific relative effects as well as for their uncertainty.
Methods:
We adapt methodology previously suggested for ranking in sports tournaments into the context of NMA. We use extensions of the so called ‘Bradley-Terry’ models, which are a family of probabilistic models that aim to predict the outcome of pairwise comparisons. We first translate the study-specific relative effects and their confidence intervals into wins, losses, and ties based on predefined minimally clinically important differences between treatments. Then, based on the number of wins, our model estimates the ‘worth’ of each treatment which is used as an intuitive measure to rank the treatments. This approach naturally captures the uncertainty of ranking because the estimates of treatment worth are accompanied by confidence intervals. The model also allows to consider simultaneously multiple outcomes in ranking by implementing a vector of several contrasts for each outcome. Finally, study precision and other important characteristics, such as risk of bias, can be incorporated by an additional parameter that represents the ‘importance’ of each study.
Results and Conclusions:
We illustrate our model using a Cochrane NMA comparing 7 treatment classes for chronic plaque psoriasis. Our model is able to clearly indicate the two classes with the largest worth but also classes with similar worth. The latter is a major strength of our approach as it preserves from exaggerating unimportant differences between treatments and drawing spurious conclusions.
Patient, public, and/or healthcare consumer involvement: None.