Meta-analysis of heterogeneous clinical trials: an empirical example

Meta-analysis of heterogeneous clinical trials is currently sub-optimal. This is because there has been no improvement in the method of weighted averaging for such studies since the DL method in 1986. This article presents the argument for the use of situation specific weights to integrate results from such trials. An empirical example is given with data from a meta-analysis done 10 years earlier. Previously reported data on 21 studies that looked at the effect of working conditions on preterm births were re-analyzed. Several methods were used to estimate the overall effect sizes. Study specific scores were included in the weighting process when combining studies and it was shown that this model not only was more conservative than the model of DL but also retains the legitimacy of the pooled effect size. The inclusion of appropriate study specific scores in an appropriate meta-analysis model permits the quantification of the variation between studies based on something tangible as opposed to the random adjustments made by the random effects model to the pooled effect size. It is important that such differences are recognized by the wider research community so that meta-analyses remain a valid tool for synthesizing research.