Abstract: AIM: To find the method of analyzing counting data from multi-center crossover clinical trails. METHODS: Multilevel Poisson regression model was fitted with the repeated-measurements as the first level,stage as the second level and patient as the third level. RESULTS: Multilevel model is not only convenient to handle data with hierarchical structures,but also more flexible than traditional methods such as repeated-measurements ANOVA. CONCLUSION: Multilevel Poisson regression model could be used in the analysis of counting data from crossover clinical designs.

Key words: crossover design, non-independent, multilevel model, multilevel Poisson regression