An Improved Model for Addressing Confounders Based on Logistic Regression

In response to the problems of class imbalance and interference from confounders faced by current medical statistical analysis, this article proposes a new parameter fusion framework, which fuses the parameters of conditional logistic regression and logistic regression to construct a prediction model that is still effective under the influence of confounders and can handle confounders. At present, traditional logistic regression cannot effectively handle confounders, and conditional logistic regression cannot directly establish a prediction model. This paper aims to systematically solve these challenges. At the same time, numerical experimental results of simulation and real dataset prove the feasibility and superiority of our proposed method. Our proposed method shows strong performance in handling confounders. Whether it is parameter estimation in the simulation stage or prediction evaluation of real data, most of its metrics are better than two baseline methods based on propensity score (PS) to control confounding.