B2017
Title: Likelihood-based inference in control risk regression with study-specific covariates
Authors: Phuc Tran - University of Padova (Italy) [presenting]
Annamaria Guolo - University of Padova (Italy)
Annamaria Guolo - University of Padova (Italy)
Abstract: Control risk regression is a meta-analysis approach to investigate the effectiveness of a treatment in clinical studies. It is characterized by the inclusion of a risk measure for the subjects in the control condition as a way to explain between-study heterogeneity. The measures of risk for the treatment group and the control group are summary information of each included study in the meta-analysis, and thus they are affected by an error. Properly accounting for measurement error is a necessary step for inference to be reliable. We investigate likelihood-based inference in control risk regression models in case additional error-affected covariates, aggregated at the study level, are taken into account. In such a situation, when the within-study covariances are unknown, we show how to compute them using Taylor expansions and empirical-type estimation. As an alternative, we develop a pseudo-likelihood approach, under a working conditional independence assumption, which sets unknown covariances to zero. The performance of the proposed solutions is examined in an extensive simulation study. Moreover, the methods are applied to a meta-analysis of the efficacy of convalescent plasma in the treatment of COVID-2019.
