How to Conduct a Rapid Evidence Review Without Cutting the Wrong Corners
A rapid review done poorly is worse than no review at all. It gives false confidence in conclusions drawn from incomplete evidence. Here is how to do it properly.
Decision-makers have always needed evidence faster than systematic reviewers could produce it. The 67-week average completion time for a Cochrane review is not a failure of the process, it reflects the genuine difficulty of doing the job properly. But it does mean that by the time a formal review answers the question, the decision has often already been made on the basis of something else.
This is where rapid evidence reviews come in. A rapid review is explicitly designed to trade some methodological rigour for speed, producing a synthesis of the available evidence in weeks rather than months. Done well, they are a legitimate and valuable tool. Done poorly, they are a way of giving false confidence to conclusions that would not survive a more complete search.
The challenge is that rapid reviews and low-quality reviews can look identical from the outside. Both produce a document describing the evidence in a short time. What distinguishes them is invisible unless you know what to look for, which is why the methodological choices matter and why they need to be reported transparently.
What you are allowed to shortcut, and what you are not
Cochrane's guidance on rapid reviews identifies several dimensions where methodological shortcuts are acceptable without fundamentally compromising the review's utility:
Limiting the date range of the search is usually acceptable, with appropriate caveats about what older literature might contain. Restricting to English-language publications is a common practical constraint, though it introduces language bias that should be acknowledged. Searching fewer databases than a full systematic review, while increasing the risk of missing relevant studies, is a recognised trade-off when time is the binding constraint.
What you cannot shortcut without compromising the review's integrity is the transparency of what you did and what you did not do. A rapid review that does not clearly report its search strategy, its inclusion and exclusion criteria, and the limitations introduced by its time constraints is not a rapid review. It is a selective literature summary, which is a fundamentally different thing.
The other thing that cannot be eliminated is quality assessment. Every included study needs at least a basic evaluation of its risk of bias, even if that evaluation is simpler than the full Cochrane Risk of Bias 2 tool. A rapid review that presents study findings without any assessment of study quality invites readers to weight all evidence equally, which produces conclusions that are systematically misleading.
The search is where most rapid reviews go wrong
The temptation in a rapid review is to run a quick PubMed search, retrieve a manageable number of results, and work from there. The problem is that a quick search is often a narrow search, and a narrow search has a specific failure mode: it surfaces the evidence that is easy to find and misses the evidence that requires effort to locate.
The evidence that is hard to find is disproportionately likely to be important. Null results, published in smaller journals or in non-English-language sources, are systematically underrepresented in the easy-to-find literature. Findings from adjacent specialties using different terminology may not surface under your search terms at all. Conference abstracts and registry data that contain relevant preliminary evidence will not appear in a PubMed search that only retrieves indexed journal articles.
A rapid review with a poor search strategy does not just miss some evidence. It systematically biases its conclusions in the direction of the published positive results that are easiest to find. The reviewer reports what the easy-to-find evidence says, which is a reliably optimistic picture of the intervention's effectiveness.
This is where AI-assisted literature search has the most to offer in a rapid review context. Semantic search across larger corpora, including preprints and grey literature, can significantly improve recall within a constrained time budget. The time saved on the search can be reinvested in the quality assessment and synthesis, which is where the irreducible human judgment sits. NousLab's evidence mapping tools are built specifically for this kind of time-constrained but rigour-conscious work.
Single screening is acceptable; single extraction is not
Full systematic reviews require two independent reviewers for both title and abstract screening and full-text review, with a reconciliation process for disagreements. This is methodologically correct and practically impossible in most rapid review timelines.
The acceptable compromise, supported by empirical evidence on screening reliability, is single screening with verification. One reviewer conducts the screening, and a second reviewer checks a random sample of exclusions. The sample size needed to detect systematic errors is smaller than most people assume, around 10 to 15 percent of exclusions is usually sufficient to identify any consistent misapplication of criteria.
What should not be reduced to a single reviewer is data extraction, specifically for the outcomes that will drive the review's conclusions. Extraction errors are common, subtle, and consequential. A misread confidence interval or an incorrectly transcribed outcome measure can invert a conclusion. If resource constraints mean only one person is doing extraction, at minimum the critical data points, the ones driving the main finding, should be verified against the source paper by a second person.
Communicating uncertainty is non-negotiable
A rapid review produced under time pressure, with a restricted search, single screening, and simplified quality assessment, will reach conclusions with more uncertainty than a full systematic review would. This needs to be said explicitly in the report, not buried in a limitations section that nobody reads.
Decision-makers commissioning rapid reviews often want certainty on a timeline that certainty is not available. The appropriate response is to be clear about what the evidence supports and at what level of confidence, and to identify the specific questions that remain uncertain and would require a more complete review to address. A rapid review that overstates the strength of its conclusions does more harm than a longer review that is honest about the limits of the evidence.
The WHO, the UK's NICE, and the US Agency for Healthcare Research and Quality (AHRQ) all have guidance on rapid review methodology. The Cochrane Rapid Reviews Methods Group has published detailed recommendations that are worth reading before starting any rapid review intended for policy or clinical guidance use. They are not prescriptive constraints, they are hard-won methodological knowledge about where the shortcuts are safe and where they are not.
