Absolute or relative effects? Arm‐based synthesis of trial data

We congratulate Hwanhee Hong and colleagues on another fascinating paper (Hong et al., 2015a) arguing the case for arm-based models for meta-analysis. The standard approach to meta-analysis is the contrast-based model where the information that is pooled over trials is the information of the trial-specific relative treatment effect, expressed for example as a log relative risk, log odds ratio, or as a mean treatment difference. In an arm-based model, it is the absolute log risk, log odds, or mean outcome on each arm that are pooled. There is no doubt that arm-based models are an intriguing alternative to the accepted understanding of metaanalysis, and that they provide a very elegant alternative approach to network meta-analysis (NMA). However, readers of Research Synthesis Methods (RSM) will have no difficulty in recognising that arm-based models represent a radical – even revolutionary – departure from current meta-analytic practice. In this commentary, we begin by outlining the key differences between armand contrast-based meta-analysis to make clear the true extent of the implications of the claims being made. We then argue that contrast based models are to be preferred on both theoretical and practical grounds. Hong and colleagues present a number of arguments about the core assumptions of contrast-based models, both in their RSM paper (Hong et al., 2015a) and previously (Zhang et al., 2014, Ohlssen et al., 2014, Zhang et al., 2015, Hong et al., 2015b), which we believe are mistaken. We offer some counter-arguments and comment on the simulation study presented by Hong et al. (2015a).