Improving the transparency of meta-analyses with interactive web applications

Increased transparency in study design and analysis is one proposed solution to the perceived reproducibility crisis facing science. Systematic review and meta-analysis—through which individual studies on a specific association are ascertained, assessed for quality and quantitatively combined—is a critical process for building consensus in medical research. However, the conventional publication model creates static evidence summaries that force the quality assessment criteria and analytical choices of a small number of authors onto all stakeholders, some of whom will have different views on the quality assessment and key features of the analysis. This leads to discordant inferences from meta-analysis results and delayed arrival at consensus. We propose a shift to interactive meta-analysis, through which stakeholders can take control of the evidence synthesis using their own quality criteria and preferred analytic approach—including the option to incorporate prior information on the association in question—to reveal how their summary estimate differs from that reported by the original analysts. We demonstrate this concept using a web-based meta-analysis of the association between genetic variation in a key tamoxifen-metabolising enzyme and breast cancer recurrence in tamoxifen-treated women. We argue that interactive meta-analyses would speed consensus-building to the degree that they reveal invariance of inferences to different study selection and analysis criteria. On the other hand, when inferences are found to differ substantially as a function of these choices, the disparities highlight where future research resources should be invested to resolve lingering sources of disagreement.