Economic evaluation of factorial randomised controlled trials: challenges, methods and recommendations

Abstract

Increasing numbers of economic evaluations are conducted alongside randomised controlled trials. Such studies include factorial trials, which randomise patients to different levels of two or more factors and can therefore evaluate the effect of multiple treatments alone and in combination. Factorial trials can provide increased statistical power or assess interactions between treatments, but raise additional challenges for trial-based economic evaluations: interactions may occur more commonly for costs and quality-adjusted life-years (QALYs) than for clinical endpoints; economic endpoints raise challenges for transformation and regression analysis; and both factors must be considered simultaneously to assess which treatment combination represents best value for money. This article aims to examine issues associated with factorial trials that include assessment of costs and/or cost-effectiveness, describe the methods that can be used to analyse such studies and make recommendations for health economists, statisticians and trialists. A hypothetical worked example is used to illustrate the challenges and demonstrate ways in which economic evaluations of factorial trials may be conducted, and how these methods affect the results and conclusions. Ignoring interactions introduces bias that could result in adopting a treatment that does not make best use of healthcare resources, while considering all interactions avoids bias but reduces statistical power. We also introduce the concept of the opportunity cost of ignoring interactions as a measure of the bias introduced by not taking account of all interactions. We conclude by offering recommendations for planning, analysing and reporting economic evaluations based on factorial trials, taking increased analysis costs into account. © 2017 The Authors. Statistics in Medicine published by John Wiley & Sons Ltd.

Keywords: cost-utility analysis; factorial design; guidelines; randomised controlled trial; trial-based economic evaluation.

Figure 1

CEACs for multiple comparisons using A regression with interaction term and B regression without an interaction term. The lines for 0, a, b and ab show the proportion of bootstrap replicates where each of the four treatment‐combinations had highest NMB. The dotted line that generally follows the top‐most curve shows the cost‐effectiveness frontier, i.e. the probability that the treatment with highest expected NMB is cost‐effective. [Colour figure can be viewed at wileyonlinelibrary.com]

Figure 2

Opportunity cost of ignoring interactions given both current and perfect information.