I have found in the surgical/ medical literature that 5 to 12% of records identified through literature searching are included in the review. There should be a clear audit trial, through PRISMA, of articles identified/ de-duplicated and excluded. Sometimes a very small proportion are retained because of language limitations, inclusion and exclusion criteria, non-human studies or just plain poor quality studies.
As Paul answered, it is usually quite low. Another really important factor to be considered is how specific your search strategy is - if you use a strategy that is very general, you will get a lot of results that are not relevant to your research question. If you use a strategy that is very specific AND is tailored to the database(s) you are using, you will return less results, but they are likely to be of higher relevance.
In systematic reviews, only a small percentage of the initially retrieved records from databases are typically included in the final review. Common estimates suggest that 1-3% of the retrieved records are usually included.
The percentage will be small if the research question is very specific, but if the question is general, you will get a lot of studies, and there is an important matter, which is the eligibility criteria.
Thank you, Britt Fedor, for your thoughtful response. I have one follow-up question, if you don't mind:
1. Given the trade-off between sensitivity and specificity in search strategies, how do you recommend balancing the risk of missing relevant studies versus the burden of screening large numbers of irrelevant ones? Are there any tools or methods you find particularly effective for optimizing this balance?
Dewan Md. Sumsuzzman it would depend on the specificity of your research question. If you have a very specific PICOS statement, then having a very tailored research strategy will be helpful, particularly if the population of interest is quite niche or you are looking to compare a specific endpoint.
The two best strategies I would suggest are 1) work with a research librarian to build your search strategy and 2) import your results into a program like covidence that automatically removes duplicate records and allows you to track your screening