Data science for Evidence

Metaverse #

Metaverse is “An R ecosystem for meta-research”. It is a meta package which installs:

  • synthsisr
    • Import, assemble and de-duplicate bibliographic data
  • litsearchr
    • Quasi-automatic search strategy development for systematic reviews
  • revtools
    • Article screening for evidence synthesis using manual or visual methods
  • metaDigitise
    • High-throughput, reproducible extraction of data from figures
  • robvis
    • Publication quality risk-of-bias figures

synthsisr #

This package is designed mainly to simplify download and export of bibliographic information:

  • Read and assemble bibliographic files
    • synthesisr can read any BibTex or RIS formatted bibliographic data files. It detects whether files are more bib-like or ris-like and imports them accordingly
  • Deduplicate bibliographic data
    • Many journals are indexed in multiple databases, so searching across databases will retrieve duplicates. After import, synthesisr can detect duplicates and retain only unique bibliographic records using a variety of methods such as string distance or fuzzy matching records
  • Write bibliographic files
    • To facilitate exporting results to other platforms after assembly and deduplication, synthesisr can write bibliographic data to .ris or .bib files

litsearchr #

Litsearchr has a GUI which is in early development

The litsearchr package for R is designed to partially automate search term selection and writing search strategies for systematic reviews

Litsearchr identifies keywords by looking at an existing set of search results using algorithms such as Rapid Automatic Keyword Extraction (RAKE) (or a simple ngram search). The package can then group the keywords into concepts and then write a boolean search strategy. This can then be checked for precision and recall against a “Gold standard” of papers which should be returned by the search.

revtools #

Revtools is a package which allows the user to interact with their literature search using code as well as with a GUI. It includes:

  • Locating duplicates
  • Screening duplicates
  • Screening of titles
  • Screening of abstracts
  • Creation of topic models to cluster and show similarity between papers

metaDigitise #

This package can extract data points from graphics, as well as computing statistics for data extracted from a graphic. It includes GUI elements to assist with the extraction.

robvis #

The robvis package takes the summary table from risk-of-bias assessments and produces plots formatted according to the assessment tool used

Resources for automation in systematic reviews #

A variety of point and click tools (with fairly restricted licences, on the whole)

Litstudy #

This is a Python package to help with literature reviews. It includes:

  • Data import and merge
  • Statistics (publication year, country of origin, number of authors etc.)
  • Network analysis (co-citation network)
  • Topic modelling (automatic topic discovery based on the words used in documents abstracts)

Appendix #

Cochrane guidelines on the use of automated tools in paper selection #

https://training.cochrane.org/handbook/current/chapter-04#section-4-6-6-2