Background: Healthcare providers, policy-makers and patients are often interested in the effect of specific treatment dosages or categories of dosage (e.g. low, moderate, high), and the treatment-effects overall. Decisions regarding network geometry (e.g. lump related dosages together into one treatment node or maintain their separation) may impact the network meta-analysis (NMA) results, and influence decision-making.
Objectives: To present hierarchical NMA models accounting for effects of treatments, dosage-categories, and single dosages, providing additional insight on different heterogeneity levels.
Methods: We developed three approaches accounting for the relationship between treatment and dosage, and including up to four heterogeneity levels. The first approach assumes the dosage-effects equal to their parent treatment-effects, and involves within-study and between-study heterogeneity levels within dosage. The second approach assumes exchangeable dosage-effects from the same distribution with a common mean, and incorporates an additional heterogeneity level, the between-dosage within-treatment. The third approach assumes exchangeable dosage-effects within a specific dosage-category with either fixed or random mean dosage-category-effects across treatments. The third model still considers the first two heterogeneity levels, yet the between-dosage is evaluated within dosage-category, and an extra heterogeneity level is added, the between-dosage-category within treatment. Consistency constraints on dosage-effects can also be placed in each model.
Results: Results from the application of the different approaches to empirical examples, will be ready by September 2015, and will be presented at the Colloquium along with strengths and limitations of the models.
Conclusions: Different approaches regarding the classification of treatments in a network may result in important variations in interpretations drawn from NMA. We suggest that researchers account for different treatment dosages in NMA, providing additional insight on heterogeneity, inconsistency, intervention ranking, and hence decision-making.