Systematic Review Assistant - a prototype

Louise AC Millard^*, Peter A Flach, Julian PT Higgins

In this work we are investigating the use of text mining methods to automate risk of bias assessments in systematic reviews.

This work is reported in: Millard LAC, Flach PA, Higgins PT. Machine learning to assist risk of bias assessments in systematic reviews. 2015 (submitted)

A prototype demonstrating our methods can be found here.

The annotated figure below shows the interface and functionality of this prototype.

Prototype summary

• Users are required to register / login. This is so that reviews can be saved and returned to later.
• Reviewers can create a review and upload PDF articles, each describing a clinical trial.
• The predict button can then be used to make predictions for each article. This is completed when the articles have been processed message is shown, highlighted in green.
• The reviewer can then go to the article viewer, where the model predictions are shown.
• The sentences are highlighted by relevance with respect to each risk of bias property. A stronger highlighting colour (higher opacity) indicates higher relevance.
• The articles are assigned scores denoting the risk of bias as described in this article, for each risk of bias property. A high score denotes a low risk of bias, and a low score denotes a high risk of bias.