Rapid review

What a rapid review is

Decision makers often need evidence faster than a full systematic review can supply. A health service has to decide whether to adopt a technology, a government faces a new public health threat, a guideline panel has weeks, not a year. A rapid review responds by simplifying the review process while keeping its systematic character: a stated question, explicit methods, a search that can be repeated, and a transparent report. The review is fast because particular steps are shortened, narrowed or omitted, and those choices are part of the method and must be reported.

There is no single definition. A recent analysis of definitions found that rapid reviews are characterized by a short timeframe, commonly under about five months, a streamlining of methods, and an intended use in decision making, with the specific shortcuts varying widely. Because the term covers many approaches, a rapid review should say precisely what was done. It is not a lower category of poor review, and it is not equal to a full systematic review either. See systematic review for the full method.

When a rapid review is appropriate

A rapid review is appropriate when there is a genuine, near deadline for a decision, when the question is well defined, and when users understand and accept the trade-off between speed and certainty. It suits questions about whether a new intervention is likely to be effective and safe in a setting, what the evidence says about an emerging problem, or what the current state of the evidence is before commissioning a full review. It is less suitable when the question is broad or ill-defined, when the topic has a vast literature that cannot be handled in the time, or when the stakes of an error in the conclusion are very high and cannot be mitigated. A rapid review may be followed by a full systematic review, or be updated as new evidence emerges, as in a living review, which is described under living systematic review.

Common simplifications

Rapid reviews simplify in various ways. The most common are listed with the main risk of each.

What the evidence says about shortcuts

Studies of how simplifications affect results give a mixed picture, and the effect depends on the shortcut. Experiments on single-reviewer title and abstract screening have found that a single screener misses a notable share of relevant records, with figures around one in ten in some studies, which is a larger loss than most reviews would want, although the effect on the final conclusions is often small when the missed studies are not decisive. Methodological reviews of single against double screening reach similar conclusions: missing studies is the main risk. Limiting databases tends to reduce the number of studies found, though the loss in results depends on the topic. Comparisons of rapid reviews with full reviews on the same question often find similar conclusions, but not always, and the differences are hard to predict.

The practical lesson is that shortcuts are not equivalent, and the less harmful ones should be chosen. Many teams prefer to retain a comprehensive search and complete dual screening and instead narrow the question or limit the appraisal, or to use technology to speed up screening. Whatever is done, the effect on the conclusions should be discussed in the report.

Guidance for conduct

The Cochrane Rapid Reviews Methods Group has published recommendations that give a common basis. They advise discussing the question and the methods with the end users at the start, so that the review fits the decision; developing a protocol, even a brief one; limiting the scope in a way that serves the decision; and, where studies are searched, using more than one database and checking additional sources if feasible. They suggest that screening and extraction can be done by one reviewer with a second checking, with that approach stated, that risk-of-bias assessment can be streamlined to the most important outcomes, and that certainty of evidence be assessed for the main outcomes. They advise reporting the methods and the limitations fully and framing conclusions to the level of certainty. These are recommendations and not a fixed recipe, and teams are expected to choose their approach with reasons.

Working with the people who will use the review

Because a rapid review exists to serve a decision, the people who will make it should be involved from the beginning. The first conversation establishes what decision is being made, by when, what kind of evidence would change it, and what level of uncertainty can be tolerated. It also settles the question and the outcomes that matter, which prevents effort being spent on those that will not affect the decision. Regular contact during the review keeps it on course, and the findings are presented in a form that users can act on, usually a short report with a summary of the main findings and a plain statement of the certainty. Agreeing this at the start protects against the review being reshaped, mid-way, to support a preferred answer.

Quality checks that save time

Some safeguards cost little and protect much. A protocol of even a page, written before the search, limits post hoc changes. A search strategy checked by a second searcher catches errors that would otherwise lose studies. Verifying a sample of single-reviewer screening decisions measures how many are being missed, so that the team knows whether to add a second reviewer. Checking all extracted data that feed the main conclusions, even when other data are single-extracted, protects the numbers that matter. A brief peer review by someone outside the team, before release, can detect over-statement. These checks make the difference between a rapid review that can be trusted for its purpose and one that cannot.

When a rapid review should become a full review

Sometimes a rapid review shows that the question needs more than it can give. Signs include a literature so large and varied that the shortcuts would remove too much, conflicting findings that depend on details the review cannot examine in the time, a decision whose consequences are serious enough that the extra certainty is worth waiting for, and evidence that is changing so fast that a one-time answer would soon be out of date. In such cases the options are to extend the review, to follow the rapid review with a full systematic review, or to maintain it as a living review. The rapid review can still be useful as a first stage, since the protocol, the search and the screening can be built on. The honest course is to tell the users what the rapid review cannot answer.

Updating

Evidence changes, so a rapid review is a snapshot. For topics where decisions will be revisited, it should state the date of the last search and the conditions under which an update would be needed, such as the publication of a major new trial. Updating is faster than the first review because the strategy, the criteria and the forms exist. Fields in which evidence accumulates quickly, such as emerging infectious diseases, have used continually updated reviews, with rules about when new studies are incorporated and when the conclusions are revised, which is the approach described for living systematic reviews.

Technology and speed

Software can speed up parts of a review without changing its logic. Screening tools prioritize records most likely to be relevant, so that the relevant ones are found earlier and screening can stop when few remain. Text mining and classification help to find randomized trials, and automation can assist in de-duplication, retrieving full texts and extracting some data items. These tools are evolving, and their performance depends on the topic. Used well, they reduce the workload without removing human judgment, but their use must be reported, with what was automated and how it was checked. Where such tools are used, the review states how decisions were verified, and our approach to automated tools is explained in the AI-use policy.

Reporting

The report should state that the review is a rapid review, give the timeframe and the reason for it, and list every simplification, with a statement of its likely effect. It should follow PRISMA 2020 where applicable, state the deviations from it, give the full search strategies and the flow of studies, and give conclusions that match the certainty of the evidence. Users of a rapid review need to know what it can and cannot show: conclusions should avoid overstatement, and the report should say what a full review might add. Where a protocol was written, it should be registered or deposited. The transparency about the shortcuts is what distinguishes a legitimate rapid review from an unexplained incomplete one. See PRISMA 2020.

Limitations

The central limitation is that speed is bought at some cost to completeness and certainty. Studies may be missed, errors may be more likely, and the conclusions are more uncertain than those of a full review. Decisions made on rapid reviews can be wrong, and the strength of the evidence should be conveyed clearly to decision makers. A rapid review done to confirm a decision already made is not a systematic assessment of the evidence. The label should not be used to cover poor methods, and there is no accepted minimum standard, so the quality of rapid reviews varies widely. The results are not advice about the care of individual patients.

How we can help

Frequently asked questions