Bayesian analysis of multidrug resistance tuberculosis from Amravati Region using non-informative priors

Abstract

This study is an attempt to fit a binary logistic model on the data of TB- patients registered under DOTS from Amravati region, with the aim to determine predictors (risk factors) of MDR-TB, under Bayesian framework. Drug resistant tuberculosis is a serious public health problem in India and worldwide. Detection and treatment of MDR‑TB is a priority in National Tuberculosis program in India. Bayesian approach with Non-informative prior is employed for data analysis in this study. MDR-TB presence is taken as the response variable in this study, with 18 explanatory variables related to clinical and treatment details of present and past history of the patients. Odds ratios for the Bayesian estimates of parameters are calculated using Gibbs Sampling procedure. It is found in the study that probability of developing MDR-Tb increases with increase in the number of previous TB treatment. Out of 18, eight variables are found to be potentially effective in the development of MDR-TB among TB patients.