Matching-adjusted indirect comparisons: identifying method variations and implementing models in R

Abstract

In the framework of evidence-based medicine, comparative effectiveness research is a fundamental activity to the development of pharmaceutical products and medical treatments. For a given medical condition, several competing treatments may exist, however, there may be no multi-arm study available comparing all the relevant treatments simultaneously. In the absence of such direct evidence, healthcare decision-makers can examine indirect comparisons of treatments generated using outcome data from separate clinical trials. In order to theoretically reduce the possibility of error in the estimates of treatment effect for each treatment obtained from the indirect comparison, adjustments can be made to account for differences between trials. Matching-Adjusted Indirect Comparisons (MAICs) have been devised as a means of accounting for the heterogeneity that can arise between different patient populations in separate trials. Using individual patient data (IPD) from one trial, baseline patient characteristics and patient outcomes are weighted to match to published aggregate summary data available for a competing treatment. This statistical technique is relatively new and the various implications for its use have not been fully explored. In addition, while the technique has been employed in numerous published applications, the finer details of its implementation are rarely fully reported. The work in this thesis aims to articulate a greater understanding of the MAIC method and standardise its implementation through means of a package for the R programming language, which will allow for analyses to be reproducible.