Developing Methods to Improve Systematic Reviews Using Clinical Trial Registries

Bourgeois, Florence

Abstract

Systematic reviews are a critical information source supporting policy and clinical decision-making, and are expected to provide a comprehensive, current, and unbiased assessment of what is known about a clinical intervention. Traditional systematic reviews are resource-intensive endeavors, and with the rapid and increasing pace of evidence production, it is becoming increasingly difficult to ensure that systematic reviews are kept up-to-date. While innovations addressing this challenge have previously focused on automating the specific tasks such as screening and data extraction, innovative approaches are now needed that leverage new data sources and consider efficiencies across sets of reviews. This proposal uses a unique resource? ClinicalTrials.gov?as a data source to develop tools that automatically identify relevant clinical trials, track them as they are completed and reported, and signal when a systematic review requires updating. We propose to investigate a corpus of systematic reviews related to interventions targeting obesity and type 2 diabetes to address the two following aims: (1) To develop and evaluate graph-based semi-supervised learning methods for identifying and linking relevant clinical trials from ClinicalTrials.gov to systematic reviews; and (2) To create and populate a dynamically-updated database of systematic reviews with data that reflects the most up-to-date view of emerging trial evidence. We will make the tools and database developed in this proposal freely available for use by the systematic review community. Thus, our work will not only introduce new ways to identify relevant evidence but will also provide an ongoing resource to support systematic reviewers in prioritizing systematic review updates and ensuring that reviews are a comprehensive and timely summary of the scientific evidence.

Public Health Relevance

Systematic reviews represent a critical source of evidence to guide clinical decision-making and are the basis of national practice guidelines and recommendations. As such, clinicians rely on systematic reviews to be inclusive of all available knowledge and to be updated on a regular basis. This is challenging given the immense workload associated with scoping, analyzing, and synthesizing trial evidence for every intervention. We propose the development of tools that take advantage of data from a large clinical trials registry to support systematic reviewers by automating the processes of identifying relevant clinical trials, tracking them as they are completed, and generating the data required to determine which systematic reviews require updating.