Background: Cross-design synthesis usually refers to the inclusion in a meta-analysis of studies addressing the same question but using different designs, for example, combining results from randomised trials with those from case-control studies. Here we describe a procedure for combining information from studies addressing different but clinically related questions, referred to, for brevity, as cross-issue synthesis.
Methods: Surveys have measured the oestrogen receptor (ER) status of invasive breast cancer in women with mutations in the BRCA1 or BRCA2 gene. These mutations confer a substantially increased risk of breast cancer. These are also published randomised trials of tamoxifen administered for at least 3 years, either as an adjuvant therapy or for primary prevention, which record whether breast cancer recurred or occured respectively. These studies also give results by ER status. There are biological reasons to suppose that tamoxifen is more effective at preventing ER-positive cancers, and may have little or no effect at preventing ER-negative cancers. Women with BRCA mutations are more likely to develop ER-negative cancers. Combining meta analyses for these two types of studies supplies an estimate of the effectiveness of tamoxifen in preventing breast cancer in women with BRCA mutation. Hierarchical models were developed for this purpose. Estimation was by Markov chain Monte Carlo (MCMC).
Results: A range of models were fitted by MCMC. Using these, the effect of tamoxifen on the relative risk of developing breast cancer in women with a mutation in the BRCA1 gene is estimated to be 0.90 95% confidence interval (CI) (0.52, 1.61), and for the BRCA2 gene 0.71 95% CI (0.45, 1.21).
Discussion: This procedure can be generalised to combine information from two sets of studies addressing different, but clinically related questions.