Hu Chunyan,Hu Liangping,Reasonably conduct the multiple Logistic regression analysis combined with the multilevel model analysis[J].SICHUAN MENTAL HEALTH,2022,35(6):500-505
Reasonably conduct the multiple Logistic regression analysis combined with the multilevel model analysis
Hu Chunyan1, Hu Liangping1,2
1.Graduate School, Academy of Military Sciences PLA China, Beijing 100850, China;2.Specialty Committee of Clinical Scientific Research Statistics of World Federation of Chinese Medicine Societies, Beijing 100029, China
Abstract:The purpose of this paper was to introduce how to reasonably analyze the multiple Logistic regression models in combination with the multilevel model analysis. Firstly, four basic concepts related to the multilevel model analysis were introduced. Secondly, three steps for building a multilevel model were given. Thirdly, through an example of a multicenter drug clinical trial, the whole process of how to use SAS software for the analysis was presented. The contests were as follows: ① testing whether the odds ratios of each center were homogenous; ② building the multiple Logistic regression model after generating dummy variables for the trial center; ③ constructing a multiple Logistic regression model with the trial center as a stratified variable; ④ building a random intercept multilevel multiple Logistic regression model; ⑤ constructing a random intercept and random slope multilevel multiple Logistic regression model. The conclusion was that when there were differences among the data at different hierarchies with binary outcome variables, the most appropriate approach was to build a multilevel multiple Logistic regression model.
Key words:  Hierarchical structure  Hierarchical variable  Multilevel model  Random intercept  Random slope