The hierarchical metaregression approach and learning from clinical evidence

Abstract

The hierarchical metaregression (HMR) approach is a multiparameter Bayesian approach for meta-analysis, which generalizes the standard mixed effects models by explicitly modeling the data collection process in the meta-analysis. The HMR allows to investigate the potential external validity of experimental results as well as to assess the internal validity of the studies included in a systematic review. The HMR automatically identifies studies presenting conflicting evidence and it downweights their influence in the meta-analysis. In addition, the HMR allows to perform cross-evidence synthesis, which combines aggregated results from randomized controlled trials to predict effectiveness in a single-arm observational study with individual participant data (IPD). In this paper, we evaluate the HMR approach using simulated data examples. We present a new real case study in diabetes research, along with a new R package called jarbes (just a rather Bayesian evidence synthesis), which automatizes the complex computations involved in the HMR.

Keywords: Bayesian hierarchical models; comparative effectiveness; conflict of evidence; cross-design synthesis; individual participant data; meta-analysis; personalized medicine.