Article type
Year
Abstract
Background: Mixed treatment comparison (MTC) meta-analysis allows several treatment options to be compared simultaneously in a single analysis while utilising both direct and indirect evidence. Three key assumptions underlie the methodology: similarity; homogeneity; and consistency. Assessment of the assumptions is vital to ensure the results of the MTC meta-analysis are interpreted appropriately. A previously proposed assessment method is the inclusion of potential treatment effect modifying covariates in the MTC meta-analysis models. For conventional pair-wise meta-analysis, important benefits regarding the investigation of treatment by covariate interactions, gained from using individual patient data rather than aggregate data for meta-analysis, have previously been described.
Objectives: We aim to compare individual patient data MTC meta-analysis models that include patient-level covariates with aggregate data MTC meta-analysis models that include study-level covariates.
Methods: Two types of random effects MTC models for dichotomous outcomes were applied: models based on the individual patient data from the trials' original datasets and models based on the 'aggregate’ event rates from each trial. For each type of MTC model, three different model specifications were applied that made increasingly stronger assumptions regarding the treatment by covariate interactions. We compared the two types of models through application to a real dataset. To ensure that any differences in the results would be due to the model specifications, we used the real individual patient dataset to generate the aggregate data.
Results: We found that the treatment effects and drug rankings from MTC models that include patient-level covariates and those from models that included study-level covariates differed. The inclusion of patient-level rather than study-level covariates produced more precise results.
Conclusions: When assessing the similarity assumption of a MTC model, including patient-level covariates is more favourable than including study-level data when the covariate distributions vary within trials.
Objectives: We aim to compare individual patient data MTC meta-analysis models that include patient-level covariates with aggregate data MTC meta-analysis models that include study-level covariates.
Methods: Two types of random effects MTC models for dichotomous outcomes were applied: models based on the individual patient data from the trials' original datasets and models based on the 'aggregate’ event rates from each trial. For each type of MTC model, three different model specifications were applied that made increasingly stronger assumptions regarding the treatment by covariate interactions. We compared the two types of models through application to a real dataset. To ensure that any differences in the results would be due to the model specifications, we used the real individual patient dataset to generate the aggregate data.
Results: We found that the treatment effects and drug rankings from MTC models that include patient-level covariates and those from models that included study-level covariates differed. The inclusion of patient-level rather than study-level covariates produced more precise results.
Conclusions: When assessing the similarity assumption of a MTC model, including patient-level covariates is more favourable than including study-level data when the covariate distributions vary within trials.