Application of Machine Learning Classifiers in Rapid Reviews for Health Research: A Case Example Using EPPI-Reviewer

Area of Science:

• Evidence synthesis
• Health technology assessment
• Systematic review methodology

Background:

• Rapid reviews are crucial for timely decision-making but face challenges in efficient study selection, especially for complex or diffuse evidence bases.
• Traditional systematic review screening can be time-consuming, posing a barrier to rapid evidence synthesis.

Purpose of the Study:

• To evaluate the effectiveness of a machine learning classifier within the EPPI-Reviewer platform for title and abstract screening in a rapid review.
• To assess the impact of machine learning-assisted screening on the efficiency and transparency of rapid evidence synthesis.

Main Methods:

• A machine learning classifier model was developed and implemented in EPPI-Reviewer for title and abstract screening.
• The model ranked records by predicted relevance based on predefined criteria to prioritize study selection.
• Real-time collaboration and audit trails were utilized to support consistent decision-making.

Main Results:

• The classifier model successfully concentrated relevant studies in higher probability bands, enabling early identification of eligible records.
• Screening progression showed declining new record identification in lower probability bands, indicating effective prioritization.
• Real-time collaboration and audit trails facilitated consistent reviewer decision-making.

Conclusions:

• Classifier-assisted screening in EPPI-Reviewer enhances the feasibility of conducting rapid reviews on complex topics within time constraints.
• While the risk of missed citations exists, this method improves efficiency and transparency in evidence synthesis.
• EPPI-Reviewer, with classifier models, offers a valuable tool for rapid evidence synthesis when appropriately trained and supported.