To assist researchers in selecting appropriate tools and reporting them transparently, we provide the technical specifications for all automation features currently available in Covidence.
This page serves as a central directory for these features. Use the table below to see which tools are available for your current review stage, understand how they assist your workflow, and find links to detailed instructions and validation metrics for each tool.
Feature | Review Stage | Model Type & Developer | Training / Validation Basis | Intended Use & Safety Measures |
Screening | Active Learning (Developed by EPPI-Centre; endorsed by Cochrane) | Dynamic: Trains solely on the reviewer’s own screening decisions within the current review. Predictions improve iteratively. | Prioritization: Reorders list to show relevant studies first. | |
Screening | Classifier | Biomedical: Evaluated on English-language biomedical records (Thomas et al., 2021). | Tagging: Applies "Possible RCT" or "Not RCT" tags. | |
Screening | Classifier (Developed by EPPI-Centre; endorsed by Cochrane) | Biomedical: Uses the same high-sensitivity model (>99.5% recall) as the tagging feature above. | Filtering: Moves non-RCTs to "Irrelevant." | |
Extraction | Generative AI - LLM (Feature designed and implemented by Covidence) | Pre-trained LLM: Uses a Large Language Model to analyze the text within your uploaded PDF. | Populating: Fills data tables with suggestions from the full text. | |
Extraction | Generative AI - LLM (Feature designed and implemented by Covidence) | Pre-trained LLM: Uses a Large Language Model to analyze the text within your uploaded PDF. | Populating: Fills intervention fields with suggestions from the full text. |
Detailed Feature Guides
For specific instructions, validation data, and copy/paste reporting templates for each tool, please refer to the dedicated articles as linked in the table above.
📝 How do you feel about Covidence’s approach to automation? Share your perspective in this 2-minute survey (anonymous)