Objective: To establish a new method on stratification analysis when the stratification limits of confounding factors was not clear or contradictory.
Method: Data on a study of diabetes mellitus in Guangdong province collected in the year of 1997 and 1998 was analyzed using cluster-stratification analysis.
Results: The efficiency of stratification analysis was improved and the confounding bias was effectively controlled with information bias avoided when the clusters-stratification analysis was applied.
Conclusion: The problem was logically solved using cluster analysis as an assistant stratification means.