updating-literature-reviews

Keeping a Literature Review Current: A Practical Guide to Review Updates

A literature review starts ageing the day its search is run. In fast-moving fields a review can be noticeably out of date within a year, and by the time it is published the newest included study may already be eighteen months old. Updating a review — whether as a one-off update, a scheduled update or a "living" review — is often treated as a smaller version of the original project. In practice it has its own pitfalls. This guide covers when to update, how to re-run searches without losing reproducibility, how to screen only what is new, and how an AI research agent such as Yila can take over the repetitive parts while you keep control of the decisions.

Decide whether an update is worth it

Not every review needs updating. Before starting, ask three questions:

1. Is the question still relevant? Practice, policy or technology may have moved on so far that a new question would serve readers better than an update of the old one.
2. Is there likely to be new evidence? A quick scoping search — the original strategy limited to the period since the last search — gives a rough count of new records. If there are only a handful and none look like large studies, the conclusions are unlikely to change.
3. Would new evidence change the conclusions? If the original review found strong, consistent evidence from many studies, one or two new trials probably will not alter it. If the evidence was sparse or conflicting, even a single good new study could matter.

Write down the answers. They justify the decision either way and form part of the update's rationale.

Reuse the original protocol, and record every change

An update should normally use the original protocol: the same question, the same eligibility criteria, the same outcomes and the same methods. That is what makes the old and new results comparable. Changes are sometimes necessary — a new outcome becomes important, a new study design becomes common, a tool such as a risk-of-bias instrument is replaced by a newer version — but each change should be listed with its reason in a "differences from the previous version" section.

A frequent mistake is to "improve" the eligibility criteria quietly during an update. Even a sensible refinement changes which old studies would qualify, so it must be applied to the original set as well, not only to new records.

Re-run the searches carefully

Search interfaces change. Databases add fields, retire syntax, change their thesaurus and alter their default behaviour. Before re-running a search:

• Check the original strategy still parses. Run it unchanged and compare the number of results for the original date range with the number reported in the first review. A large difference suggests the database has changed how it interprets the query.
• Update controlled vocabulary. Thesauri such as MeSH and Emtree add new terms every year. A new subject heading may now describe your topic more precisely than the free-text terms you relied on before.
• Set the date limit with overlap. Do not start exactly on the day of the last search. Indexing delays mean that records published before the last search may have been added to the database since. An overlap of one to three months is common, with duplicates removed afterwards.
• Add sources that did not exist before. New preprint servers, registries or discipline-specific indexes may now be important.

Log the date, the exact strategy and the number of records for each source, just as in the original review.

An AI research agent is particularly helpful here. Yila can take the original strategy, highlight syntax that may now behave differently, suggest new controlled-vocabulary terms that have appeared since the first search, and keep a precise log of what was run and when. The researcher reviews and approves each change, but the comparison work is done in minutes instead of hours.

Deduplicate against everything you have already seen

New search results must be checked against the complete library from the original review — including records that were excluded at title and abstract stage — not only against each other. Otherwise previously excluded papers reappear as "new" and reviewers waste time re-screening them, sometimes reaching different decisions the second time. Keep the original library, with its screening decisions, as the reference set, and remove matches before new screening begins.

Screen only what is new, with the same rules

New records are screened with the original criteria, by reviewers who have read the original guidance notes. If the team has changed, run a short calibration exercise using a sample of records from the original review where the decisions are already known. This shows quickly whether the new reviewers interpret the criteria in the same way.

Watch for new reports of old studies: extended follow-up, secondary analyses or the final publication of a study that was previously included from a conference abstract. These are not new studies. They update the data for studies already in the review and must be linked to them.

Update the analysis transparently

When new studies are added, re-run the meta-analysis or synthesis with all eligible studies. Report:

• the number of new studies and participants,
• whether the direction or size of the effect changed,
• whether the certainty of evidence changed and why,
• whether any previously included study has been retracted, corrected or superseded.

Retractions are easily overlooked. Check every previously included study against retraction databases and journal notices as part of the update; a retracted study that remains in a pooled estimate is a serious error.

Consider a living review

If evidence is accumulating rapidly and the question is important, a living systematic review may be appropriate. Searches are run frequently, perhaps monthly, new evidence is incorporated as it appears and the review is republished whenever conclusions change. Living reviews need a clear plan for search frequency, criteria for when to update the analysis, and a team that can sustain the workload. They also benefit most from automation: routine searches, deduplication against the existing library and first-pass screening suggestions are exactly the steps where an agent like Yila reduces effort, while human reviewers make the inclusion decisions and interpret the results.

Summary

A good update is not simply the original review with a newer date. It starts with a justified decision to update, reuses the protocol and documents every change, re-runs the searches with care for changed interfaces and indexing delays, deduplicates against everything already seen, screens with the same rules, links new reports to existing studies and checks for retractions. Much of this is careful administration, which an AI research agent can handle reliably, leaving researchers to focus on whether the new evidence changes what the review should tell its readers.