Metaanalizde Kullanılan Mantel-Haenszel ve Peto Yöntemlerinin Karşılaştırılması: Metodolojik Çalışma

Abstract

Objective: Meta-analysis methods aim to achieve a single common summary statistic for the parameter estimation by combining homogeneous statistics from different studies. In this study, the performances of two of the most preferred meta-analysis approach used for combining summary statistics calculated from binary data sets, the Mantel-Haenszel (MH) and Peto methods, are examined. Material and Methods: In the study, the performances of the MH and Peto methods, were examined by means of a simulation study. Hypothetical populations formed from 1,000,000 units with 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 odds ratios with data obtained from samples taken from each hypothetical population having different disease-cause rates (P), in different sample sizes (n), and with different numbers of studies (k). To compare the performance of the methods, relative bias (RB) and relative mean squared error scales were used. Results: Considering that the studies taken for meta-analysis are both homogeneous and heterogeneous, the data obtained from the simulation study were analyzed and the results obtained from the analysis were presented through tables. Evaluation of the performance of the 2 methods according to RB and relative mean squared error criteria according to (n) and (k) are presented with graphics. Conclusion: For both the fixed effects model and the random effects model, the Peto method provides more coherent estimates for the population parameter than the MH method.