Comparison of the Mantel-Haenszel and Peto Methods Used inMeta-Analysis: Methodological Study

Abstract

Objective: Meta-analysis methods aim to achieve asingle common summary statistic for the parameter estimation bycombining homogeneous statistics from different studies. In thisstudy, the performances of two of the most preferred meta-analysisapproach used for combining summary statistics calculated frombinary data sets, the Mantel-Haenszel (MH) and Peto methods, areexamined. Material and Methods: In the study, the performancesof the MH and Peto methods, were examined by means of a simula tion study. Hypothetical populations formed from 1,000,000 unitswith different disease-cause rates (P(E+ \\P + )=0.50, 0.60, 0.70, 0.80,0.90) were created. Both methods were applied by generating oddsratios with data obtained from samples taken from each hypothet ical population having different disease-cause rates (P), in differentsample sizes (n), and with different numbers of studies (k). Tocompare the performance of the methods, relative bias (RB) andrelative mean squared error scales were used. Results: Consideringthat the studies taken for meta-analysis are both homogeneous andheterogeneous, the data obtained from the simulation study wereanalyzed and the results obtained from the analysis were presentedthrough tables. Evaluation of the performance of the 2 methodsaccording to RB and relative mean squared error criteria accordingto (n) and (k) are presented with graphics. Conclusion: For both thefixed effects model and the random effects model, the Peto methodprovides more coherent estimates for the population parameter thanthe MH method.