A Comparative Analysis to estimate Odds Ratio for K 2 x 2 Sparse Data Sets -Bayesian Modelling

Abstract

An issue that exists during the meta-analysis process is that one or maybe more studies may have sparse data, such as no events in the treatment and control groups.Estimating Odds Ratio, one of the three association measures, will be challenging for such a dataset.In certain cases, it is standard procedure to either add a constant or remove the study to estimate the unknown values.This technique, however, is relied on the asymptotic characteristics of the estimates and may underperform for sample sizes. Another strategy would be to employ Bayesian approaches to have a better understanding of the problem. The main motive of this work is to perform a comparative study in analysing the robustness of highly imbalanced datasets with zero events for various values of K studies (0 < k ≤ 50) in the dataset between the Binomial-Normal and Normal-Normal models.A Bayesian method with an appropriate prior might be a feasible option for dealing with sparse data, according to the findings of a comparative research.Both methodologies performed better in all sparse datasets and are suggested to be used in meta-analyses of similar scope.