Objective: To introduce two comprehensive methods of analysis, Meta and pooled, frequently used in cancer epidemiology, and to compare them with authentic examples.
Methods: From data of a multicenter case-control study of lung cancer risk factors among non-smoking females, "family history of lung cancer" and "smoking history of parents" were the two variables selected for analysis. Based on the data from individual centers, Meta and pooled methods were employed to perform comprehensive analysis and the results compared. In the former, both fixed and random effect models were considered.
Results: When the data from individual study centers were analyzed, odds ratios (OR) differed. The OR of variable "family history of lung history" was statistically significant in Shanghai, while that of variable "smoking history of parents" was statistically significant in Harbin. Given the fixed effect model, the results of analysis were identical under different conditions of analysis (non-adjusted, adjusted with Mantel-Haenszel methods and multivariate logistic analysis). The results were similar when random effect model was used. In case the fixed effect model was not valid and there were some differences obtained by the two methods, combined parameters calculated from random effect model would be used. Stratified logistic model seemed better than basic logistic model in the pooled analysis.
Conclusion: The results from Meta and pooled comprehensive analyses are essentially consistent.