How to Model Different Dose-Effects in Networks of Interventions
Presenting author: Dr. Areti-Angeliki Veroniki, Li Ka Shing Knowledge Institute, St Michael's Hospital
Co-authors: Dr. Andrea C. Tricco, Scientist and Dr. Sharon Straus, Clinical Epidemiologist and Director, Li Ka Shing Knowledge Institute, St Michael's Hospital
Background: Network meta-analysis (NMA) is a powerful tool that combines different sources of evidence in a network for a particular clinical topic. To ensure the validity of NMA findings, the structure of the network should carefully be designed. However, a common difficulty investigators encounter is that the intervention doses may vary across the eligible trials, and different node assumptions (i.e., about how doses are lumped together) might affect the network structure. Decision-makers are often interested in the relative efficacy of different dosages; in particular, whether the same treatment is more effective at a higher dose.
Objective: To provide a description of the available approaches to model dose-effects of interventions along with illustrative examples.
Methods/Results: A common practice is to choose between "lumping" (i.e., different doses of the same treatment are considered the same intervention) and "splitting" (i.e., different doses of the same treatment are considered separate interventions) the treatment nodes in the network. The splitting approach ignores the relationship that exists between doses that belong to the same intervention, whereas the lumping approach ignores the different doses for each intervention. We will present the approaches that account for dose in NMA, providing additional insight on heterogeneity and inconsistency. The strengths and limitations of these approaches will be described using empirical examples.
Conclusion: Different approaches and assumptions on dose-effects may result in discrepant NMA findings because of the differences in the structure or the extent of the network. This might also impact the heterogeneity and inconsistency estimation, and hence the intervention ranking.
An Alternative Parameterization of Bayesian Logistic Hierarchical Models for Mixed Treatment Comparisons
Presenting author: Dr. Petros Pechlivanoglou, Health Economist, THETA — Toronto Health Economics and Technology Assessment — Collaborative
Co-authors: Dr. Fentaw Yazew, Post-doctoral Fellow; Maarten Postma, Professor; and Ernst Wit, Professor; University of Groningen
Mixed treatment comparison models (MTCs) rely on estimates of relative effectiveness from randomized clinical trials (RCTs) so as to respect randomization across treatment arms. This approach could potentially be simplified by an alternative parameterization of the way effectiveness is modelled. We introduce a treatment-based parameterization of the MTC model that estimates outcomes on both the study and treatment levels. We compare the proposed model to the commonly-used MTC models using a simulation study, as well as three RCT datasets from published systematic reviews comparing treatments on bleeding after cirrhosis, the impact of antihypertensive drugs in diabetes mellitus, and smoking cessation. The simulation results suggest similar or sometimes better performance of the treatment-based MTC model. Moreover, from the real data analyses, little differences were observed on the inference extracted from both models. Overall, our proposed MTC approach performed as good, or better, than the commonly-applied indirect and mixed treatment comparison models and is simpler, faster, and easier to implement in standard statistical software.
Incorporating Data From Single-Arm Studies Into Network Meta-Analyses
Presenting author: Steve Kanters, Graduate Student, University of British Columbia (UBC)
Co-authors: Dr. Kristian Thorlund, Assistant Professor, McMaster University; Dr. Edward Mills, Associate Professor, Stanford University; Dr. Jeroen Jansen, Associate Professor, Tufts University; Dr. Nick Bansback, Assistant Professor, UBC
Network meta-analyses (NMA)are primarily restricted to evidence provided through randomized controlled trials (RCTs). The advantage of RCT evidence is that the treatment effect can be disentangled from the study effect. Nonetheless, there are areas of research where observational studies may be required, such as networks that would be disconnected otherwise, in rare diseases where RCTs are uncommon because of logistical constraints, and to incorporate evidence from non-randomized extension studies.
Methods have been developed to incorporate comparative observational studies within NMA, such as the use of Bayesian hierarchical modelling. However, no methods have been discussed regarding the incorporation of endonodal, or single-arm, data. We propose two approaches by which to incorporate such data. First, constructing informative priors based on the endonodal data. In this manner, priors for the study effects μi would remain non-informative, but priors for the treatment effects dij would be constructed using single-arm evidence at each pair of nodes. Using this approach, evidence from single-arm evidence could be down-weighted in order to determine the influence of each data source. The second manner by which to incorporate such data is through the creation of a pseudo-reference group. By using characteristics of the single-arm trials, an imputation of the result for a pseudo-reference category could be created — hence, allowing for the single-arm data to connect to the network.