Part 1 · The review, stage by stage
1.11Living reviews
A living review re-runs its locked search on a schedule, screens what is new, and tells you when the pooled results move. Nothing it finds changes a reviewed conclusion until you accept it.
What it does
Once a review's search strategy is locked and living updates are on, an update runs daily, weekly or monthly. It searches the saved PubMed and ClinicalTrials.gov queries again over the dates since the previous update, with a day of overlap so no study falls between two updates, screens the new records against the eligibility criteria, and puts them in the review's inbox with the AI's verdict. Where you have switched them on, it also retrieves full texts, extracts data as drafts, assesses risk of bias and re-runs the pairwise analysis, then compares each pooled result with the previous cycle's. When the change crosses the thresholds you set, you receive an e-mail; otherwise a digest says how many new studies were found.
It proposes screening verdicts, draft extractions and the comparison. It never includes a study, validates a flagged value or overwrites a reviewed conclusion by itself: new evidence returns through screening, full text, extraction, analysis and your approval. A network meta-analysis is not re-run by an update.
What you do
Turn updates on
Lock the search strategy first (Search and sources): an update re-runs the saved search unattended, so the dialog refuses to switch updates on until the queries are locked, and offers the lock control. Then open Living review on Configure, switch updates on and choose the frequency. Save and run now starts the first update at once; otherwise the first runs on the scheduler's next hourly tick after the chosen interval. The dialog shows when the last update ran and an estimate of the next.
What an update searches
Living review settings say what each update searches, among the databases the review uses. Searched by date: PubMed, ClinicalTrials.gov, Europe PMC and OpenAlex, each update searching the records added since the last one, with a day of overlap so no study falls between two updates, and reading every page for those dates, up to 2,000 studies per database. For OpenAlex the date is the publication date, looking back 30 days, so records it adds late are still found. Searched again in full: Cochrane CENTRAL, read to its own limit every update, adding only records not already in the review. After the search: citation chasing, where the strategy has it on. Each update appears on the Search page under Recent search activity as an automatic update search, with the dates it covered, how many studies it found and how many were new. An update that could not search a database, or could not search it to the end, is marked incomplete and names the database. When a database matches more than an update takes in, narrow its search: the next update starts again from the same day, and CENTRAL from its first results. Update searches are not held to a review's 5,000-record limit, and the studies they add do not count toward it, so a full living review keeps taking in new evidence.
Work the inbox
New candidates wait in the inbox on the review's Overview, which tops the to-do list whenever it holds something, each with the AI's screening verdict. Select studies and choose Accept, Reject or Import data. Import data includes the selected studies and validates only the values nothing flags, as validating a whole study does; the rest wait for a reviewer in Extraction. When any study in the batch goes against the AI's verdict, or was not screened by it yet — rejecting a study the AI included, or accepting one it was unsure about — the action asks for one reason for the batch, kept in each of those studies' history; a batch that agrees with the AI saves at once. A study already decided at full text is refused rather than moved.
Read an alert
With Re-run the analysis after each cycle on, every update pools each validated outcome again and compares the result with the previous cycle's. A result is flagged as a significant change when, with at least the minimum number of newly included trials, one of the rules fires: the pooled estimate moved by more than the effect-size threshold (on the log scale for ratio measures); the confidence interval brackets the null value in one cycle but not the other; between-study inconsistency rose by at least the set number of percentage points; or, where the update assesses certainty, the GRADE certainty was downgraded. Magnitude is measured against the last result you were alerted on, so a slow drift is not lost between cycles. The e-mail, “Evidence change detected”, names the review and the outcomes affected and links to it; an update with nothing significant sends “N new studies found” instead.
Switch updates off
Switch updates off in the same dialog. Start the search over on the Search page also switches them off, so an update cannot bring the hidden records back as new ones; switch them on again once you have searched again and locked the queries.
Settings
The schedule — whether updates run and how often — is kept with the review but outside its settings file; what an update does is a review setting (what the settings cover; Review settings and history). An Admin or a Lead Reviewer changes either; the Configure assistant can change them too, with your agreement. A change applies from the next update; results already produced are kept.
- Frequency — daily, weekly or monthly. Switching updates on starts the clock now; changing the frequency while on leaves the current cycle in place.
- Most studies per run and Minimum AI confidence — a study below the confidence is marked Unsure and not extracted automatically.
- Extract automatically, as drafts and Assess risk of bias automatically — what each cycle does with a newly included study.
- Re-run the analysis after each cycle — the pairwise comparison and its thresholds: effect-size change (10% by default), alert when the CI crosses null (on), heterogeneity jump (20 percentage points of I²), and minimum new trials (1, up to 20).
Who is e-mailed, and whether at once or in a digest, is each person's choice under Settings › Notifications; a significant change and a failed update are sent at once by default, the count of new studies in your digest.
What is recorded
| Record | Where you see it | Exported in |
|---|---|---|
| Each update: when it started, whether the scheduler or a person started it, how it ended and how many new studies it added | Configure › Living review (last and next update) | — |
| The PRISMA flow as each update left it, with the records in each box, to open under As of and compare with now | PRISMA flow › As of | The PRISMA flow’s files |
| The update search: the dates it covered, what it sent to each database, how many studies it found and how many were new, and whether a database was left incomplete | Search › Recent search activity | — |
| Each new record, marked as found by an update, with its screening verdict and the reasons | Inbox; Screening; the record | RIS/BibTeX of the records |
| Each inbox decision: Accept, Reject or Import data, who made it, when, and the reason when the batch went against the AI | The study’s history | — |
| For each pooled outcome, each cycle’s estimate, interval, heterogeneity, number of studies and certainty, and whether it raised an alert | Analysis (the current result); the alert e-mail (the change) | — |
| Each notification sent: the significant change, the digest of new studies, a failed update | The notification bell; your e-mail | — |
Limits
ERIC and MAUDE are not searched again. A network meta-analysis is not re-run automatically: re-run it from Analysis after new evidence arrives, and the pairwise comparison stays paused for network reviews so results never mix treatment comparisons.
The significance rule compares pooled results; it does not judge whether a change matters clinically, and a first cycle has nothing to compare with. A review whose search was started over must be searched and locked again before updates resume.
What the AI does by itself in an update, and the consent it runs on: What the AI does by itself.