Interpretation of logistic regression and propensity score analysis

Abstract

 One of the multivariate analyses commonly used in medical research is logistic regression. It is designed to predict from explanatory variables (covariates) the probability of binary (nominal) outcomes (P), such as survival or death and positive or negative response to a test drug, using a logistic regression model, where log (P /(1-P)) is assumed to be a linear combination of covariates (b₀ + b ₁ x₁ + b ₂ x₂ ...).
  In non-randomized retrospective studies, the results of these analyses may be biased by confounding factors. To adjust for the confounders, a propensity score analysis using logistic regression has been proposed, permitting us to abridge multiple confounders to a single composite variable. This concept and technique are herein explained.