Rapid Review: systematic evidence on a deadline

Every rapid review makes trade-offs; the discipline is in choosing which ones the question can afford. Restricting to the last ten years is cheap for a fast-moving technology and expensive for a question about long-term outcomes. Single-screener screening risks missing studies; verification on a 20% sample bounds that risk and measures it.

The report therefore carries a limitations-by-design section: each abbreviation, why it was acceptable for this question, and what direction of bias it could introduce. That section is not an apology — it is the evidence that the review knew what it was doing.

The Cochrane Rapid Reviews Methods Group published interim guidance — Garritty and colleagues, Cochrane Rapid Reviews Methods Group offers evidence-informed guidance to conduct rapid reviews — which is the closest the field has to a standard: recommendations on database counts, screening verification, extraction checking, and appraisal focus.

Tricco and colleagues’ A scoping review of rapid review methods documents how variable practice actually is — dozens of distinct abbreviation patterns in the wild — which is itself the argument for stating yours explicitly.

For the reporting frame, PRISMA 2020 still applies; a rapid review reports against it and marks the departures.

• Right: a real deadline attached to a real decision. A ministry choosing between two service models in eight weeks is served better by a declared-shortcut review than by expert recollection.
• Right: an update of an existing systematic review. The heavy lifting — question, criteria, appraisal frame — is inherited; the rapid work is the new interval.
• Wrong: a dissertation or guideline that will stand for years. The time saved is trivial against the lifespan of the conclusion.
• Wrong: a contested field where selective inclusion will be litigated. Shortcuts, however honest, hand critics their opening; contested questions need the full apparatus.

A city commissioned a review of Housing First versus treatment-first models for its next budget, due in nine weeks. The team searched two databases and one homelessness research repository, English, 2010 onward — each restriction justified in the protocol (the model literature is recent, anglophone-dominated, and well-indexed).

From 612 records, 28 studies were included. Single screening with 20% verification found a 4% disagreement rate, reported as-is. Extraction was checked on all 28; appraisal used a brief checklist focused on comparison-group integrity, the dimension most likely to flip the conclusion.

The synthesis reached the commissioners in week nine: consistent evidence of higher housing retention under Housing First, thinner and more mixed evidence on substance-use outcomes, and an explicit note that the excluded non-English literature was concentrated in exactly the Scandinavian programmes the city most wanted to emulate — a shortcut cost the review named rather than buried, and the council funded a follow-up translation pass because of it.

• Undeclared shortcuts. The difference between a rapid review and a sloppy one is the declaration.
• Shortcuts chosen for convenience rather than risk. Dropping the second screener is cheap; dropping the grey literature on a policy question may erase the most relevant evidence.
• Letting the question grow. Scope creep converts an eight-week review into a late, thin systematic review.
• Presenting rapid findings with systematic-review confidence. The caveats belong in the executive summary, not only in the annex.
• No verification sample at all. Single-screener error rates are well documented; an unverified screen is an unmeasured risk.

The method’s validity depends on a judgement — that the shortcuts chosen do not change the answer — which can only be tested by doing the full review nobody had time for. Comparisons of rapid and full reviews on the same questions are mostly reassuring, but the guarantee is empirical, not logical.

Rapid reviews also concentrate their bias in one direction: toward the indexed, the English-language, and the recent. For globally distributed or older literatures, that is not noise but a systematic tilt.

And the format invites over-extraction by its commissioners: a bounded answer to a bounded question gets quoted without its bounds. Writing the caveats into the headline finding is the only real defence.

A systematic literature review is the reference design a rapid review abbreviates — when the deadline is soft, prefer it.

A scoping review suits a broad mapping question; rapid reviews suit narrow decision questions.

Narrative synthesis is usually the synthesis engine inside a rapid review, since pooling rarely fits the timeline.

Time is the whole constraint, so the leverage sits in screening and extraction. AI-assisted extraction against a defined set of fields, with each answer linked to its source passage, lets the second reviewer verify instead of re-extract — Evidano’s document analysis is built around exactly that check-the-quote workflow, and it compresses the weeks the deadline does not have.

What stays human is the risk calculus: which shortcuts this question can afford is a judgement about the decision, not the documents.