Meta-analysis for trend estimation.

Grouped dose measures, heterogeneity and publication bias are three major problems for meta-analysis in trend estimation. In this paper, we propose a model that allows for arbitrarily aggregated dose levels, and show that the resulting estimates and standard errors can be quite different from those given by the usual assigned value method. Based on fitting a model to the funnel plot, we discuss a method for random-effects sensitivity analysis that deals with the problems of heterogeneity and publication bias. A meta-analysis of epidemiological studies on the effect of alcohol on the risk of breast cancer is used to illustrate the method. Our analysis suggests that the rate of increase in risk with alcohol consumption is substantially less than has been previously suggested. Copyright 2004 John Wiley & Sons, Ltd.