OBJECTIVES:

Indirect treatment comparison (ITC) methods are required to assess the relative efficacy between treatments when no head-to-head clinical trials are available. However, trials will frequently include a combination of different follow-up lengths, making it challenging to perform long-term comparisons. The aim of this study is to validate a mixed model for repeated measures (MMRM) approach to include trials with various follow-up times in unanchored ITCs of long-term outcomes.

METHODS:

The proposed approach uses a multivariate normal likelihood with unstructured variance covariance matrix which assumes that missing timepoints are missing at random and can be considered similar to an aggregate version of MMRM. This MMRM approach is compared to a simpler model-based network meta-analysis modeling approach with a spline smoothed trend across time and an assumed constant correlation for variance inflation. Models are conducted within a Bayesian framework using aggregate level inputs.

RESULTS:

If there are many timepoints with missing data, applying an MMRM approach can lead to significant improvements in the precision of estimated effects. In situations where final models are combined in random effect meta-analysis, the MMRM approach allows for a more reliable estimate of between trial heterogeneity than is otherwise possible. The MMRM approach can be challenging to implement computationally and in settings where the distance between timepoint measurements or their correlation over time differ.

CONCLUSIONS:

Applying an MMRM approach may be valuable in situations where unanchored comparisons conducted at earlier timepoints are forced to use limited trials at later timepoints due to a lack of long-term data.