Generalized least squares for assessing trends in cumulative meta-analysis: applications in genetic epidemiology

Pantelis G. Bagos and Georgios Κ. Nikolopoulos

We propose a simple and formal regression-based approach for detecting trends in cumulative meta-analysis. By using the logarithm of the combined Odds Ratio (logOR) of studies published up to a particular time, as dependent variable and the rank of the published studies as independent variable in a linear regression with weights proportional to the inverse variance of logORs, we were able to detect and validate a possible trend over time for the speculated association. The correlation of successive logORs used in the regression, is been dealt by introducing a first order autoregressive (AR1) coefficient using Generalized Least Squares (GLS). Application of the method in several already published meta-analyses of genetic association studies provides very encouraging results, outperforming the commonly used method of comparing the results of first vs. subsequent studies. The particular method is intuitive, very easily implemented in nearly every statistical package, and allows drawing conclusions based on formal statistical tests.

The Stata program for fitting the models proposed in this work, is available here metatrend.ado (and a help file: metatrend.hlp)
The program was developed in Stata 8.0 although it is functional with older versions (6.0 and 7.0).


The method implemented here, is presented in:

Bagos PG, Nikolopoulos GK. Generalized least squares for assessing trends in cumulative meta-analysis with applications in genetic epidemiology. 2009, Journal of Clinical Epidemiology, 62 (10): 1037-1044 [PDF] [Pubmed] [Google Scholar]

The 4 datasets used in the paper for illustrating can be found here: CTLA4, GNB3, PAI1, ITA2

See also metatrend at Econpapers and Repec

From within Stata issue the command: