Title: Efficient implementation of Copas selection model for publication bias in meta-analysis using clinical trial registry
Authors: Ao Huang - Department of Biomedical Statistics, Graduate School of Medicine, Osaka University (China) [presenting]
Sho Komukai - Osaka University (Japan)
Satoshi Hattori - Osaka University (Japan)
Abstract: Copas selection model is one of the useful sensitivity analysis methods for publication bias in the standard meta-analyses. Despite its usefulness to quantify potential impacts of publication bias, it has some undesirable features. In conducting the sensitivity analysis, one needs to make inference with sets of fixed sensitivity parameters. This may lead us difficulty in interpreting the results of the sensitivity analysis. A method to estimate all the unknown parameters has been proposed based on data with an EM-type algorithm. However, this method is constructed under a strong assumption on funnel-plot symmetry. We extend the inference procedure for the Copas selection model by utilizing information from the clinicalTrials.gov, and propose two strategies in estimating the parameters of interest: one is the two stage method which estimates the parameters in the selection model first using a probit model, then with the parameters fixed in the likelihood, we estimate the bias-adjusted treatment effects; the other is utilizing the full likelihood function with all the information to estimate the parameters simultaneously. Through applications to real datasets and simulation studies, we show that our methods enable us to conduct the sensitivity analysis more stably and have more interpretable insights on publication bias.