Abstract: AIM: To find the method of analyzing data from multi-center cross-over clinical trails.METHODS: General linear models with mixed effects and generalized linear mixed models (GLMM) were adapted to investigate the difference between groups.RESULTS: When subjects were estimated as random effects, not only the accuracy of the estimation was improved, but also covariates, such as the center, age, gender and baseline could be taken into consideration.These methods could also deal with data with missing values and without deleting these observations.CONCLUSION: In multi-center cross-over clinical trails, general linear models with mixed effects should be applied when response variables are continuous.GLMM are suited for categorical data.

Key words: cross-over design, linear models with mixed effects, generalized linear mixed models, multi-center, clinical trails, missing data