B0842
Title: Accounting for measurement errors in control risk regression through structural and functional approaches
Authors: Annamaria Guolo - University of Padova (Italy) [presenting]
Abstract: Detecting heterogeneity among studies about the same issue of interest is one of the main goals of meta-analysis. When studying the effectiveness of a treatment, between-study heterogeneity can be explained by including a measure of risk for subjects in the control condition, an approach giving rise to the so-called control risk regression. The measure of risk for the treatment group and the control group is a summary of information from each study. As a surrogate for the true unknown risk of outcome at the population level, it is prone to measurement error. Correcting for measurement errors has been recognized as a necessary step to provide reliable inference. A classical widespread solution considers a likelihood-based structural approach assuming specific distributions for all the involved variables, control risk measures included. A functional alternative - SIMEX - is examined to perform inference through a simulation-based approach without assuming the distribution of the true unobserved control risk. Such a robustness property and the feasibility of computation make SIMEX very attractive. Characteristics of the approaches, including accuracy of inference and computational performance, are illustrated through simulation and in a meta-analysis about the association between diabetes and the risk of Parkinson's disease.
