Why this is a stage, not just another screen
An examiner or reviewer asks why studies were excluded, not only how many. This stage makes you inspect that list before proceeding, and you must approve it before full text opens.
How reasons are attributed
The table counts excluded studies grouped by criterion. Attribution follows two rules:
- Automatic exclusions: assigned to the first criterion that fails. If a study fails two, it counts once under the first.
- Human exclusions: assigned to the criterion selected by the reviewer when excluding it from the queue.
Manual overrides always prevail. If you rescue a study, it stops counting as excluded and the table updates automatically.
The table groups by required abstract criteria. Reasons not tied to a criterion have their own row: uncertainty-band exclusions and manually written reasons. Optional criteria do not appear because they cannot exclude by themselves.
What to review here
- Find any criterion that swallowed the review. If one explains almost every exclusion, it is usually unclear or fails to say what to do when information is missing. Catch that now.
- Inspect unusual reasons. A criterion with two or three exclusions may be correct, or the AI may read it differently from you. Open one and check the quotation.
- Rescue anything excluded incorrectly. An override requires a reason, recorded in the audit log with your name and the date.
- Export the table. It is available as CSV alongside the override log, ready for your appendix.
PRISMA uses these counts
The reason counts in this table appear in the PRISMA 2020 diagram at the export stage. You never enter them manually.
Troubleshooting
- The numbers do not match screening. Each excluded study counts once under the first failed criterion, so criterion totals are not totals of every failure. Also check non-criterion rows: band and manual reasons are separate.
- One criterion excludes almost everything. Return to Criteria, check what it says about unreported data, then rerun screening.
- I want to rescue several studies at once. Overrides are deliberately individual. Each rescue needs a reason to keep the audit log useful.