Combining Information from Diverse Sources

Research synthesis plays a central role in the process of scientific discovery, providing a formal methodology for the systematic accumulation and evaluation of scientific evidence. There are many situations in which research synthesis is required because obtaining the required information from an individual trial is not possible or practical. This thesis proposes a general method to synthesize information from multiple sources. The study of the relationship between suicide and antidepressant use in children and adolescents is one such case. To shed light on this issue, the FDA combined data from 24 randomized controlled trials using standard frequentist fixed and random effects models. However, the diversity of the trials suggested the presence of systematic effect variation that was not incorporated into these models. We applied a Bayesian hierarchical model to the data to include more appropriate variance structure and answer scientific questions regarding subsets of the data. While this technique was successful in solving the variance and subsetting issues, the collection of clinical trials share qualities that limit the generalizability of the meta-analysis. Several available administrative databases do not suffer from these same limitations of generalizability. Combining administrative databases and clinical trials in one analysis would improve the reliability and generalizability of the analysis. Several methods have been proposed to accomplish this goal, but to our knowledge, only one has been implemented and so many practical problems related to these methods (including exchangeability, bias, prior specification and model selection) have not been addressed. This thesis proposes a general method to synthesize information from diverse sources that has two of the already proposed methods as special cases and is also consistent with the third proposed method.