Introducing CiteSource: A new R package to explore source-level contributions to the stages of a systematic review

Article type
Authors
Young S1, Riley T2, Hair K3, Wallrich L4, Grainger M5, Pritchard C6, Haddaway N7
1Carnegie Mellon University
2National Oceanic and Atmospheric Administration
3University of Edinburgh
4University of London
5Norwegian Institute for Nature Research
6NTU University Centre
7Stockholm Environment Institute
Abstract
Background:
Determining the contribution of different databases to a systematic review search is challenging and tedious with currently available tools like Endnote. Yet, understanding the value that a source brings to a review in terms of unique and relevant records can greatly improve and optimize an overall search strategy and guide decisions about sources to search in the review updating process.

Objectives:
We aimed to build a new R package that maintains source or other user-defined metadata during the deduplication process, opening up many possibilities for the analysis of source contribution, including contributions at different screening and inclusion stages of the review process.

Methods:
The R package, called CiteSource, was developed collaboratively by a team of interdisciplinary researchers, developers, and information specialists and emerged from a hackathon at the Evidence Synthesis and Meta-Analysis in R Conference. The package builds on existing evidence synthesis R packages like ASySD. Unlike standard deduplication processes in most tools, which merge records into a single record while losing original metadata and source information, CiteSource was designed to maintain user-defined metadata at the imported file level, such as source, database, search string, or review stage. Various plots and tables were tested to determine the best visualizations to describe overlap between sources and various other use cases were considered for the tool.

Results:
CiteSource is now available along with an accompanying Shiny app. It includes three customizable plots to visualize overlap, as well as a search summary table with sensitivity and precision values and a record-level table that indicates the databases in which individual records were found. It also includes various export formats that can facilitate further analysis and comparison.

Conclusions:
CiteSource can be used for source contribution comparisons and to generate search summary tables. However, there may be many other unexplored use cases for this new tool. For example, CiteSource can be used pre-screening to evaluate the usefulness of databases to a search and their contribution to a set of benchmark studies, or to examine differences in search methods.

Patient, public, and/or healthcare consumer involvement:
None.