Audit Language Restrictions in an Evidence Search

Locate where the restriction occurs

Language restrictions can enter through database filters, screening judgments, full-text retrieval or an inability to translate an otherwise eligible report. Audit each point separately. A search with no language filter can still become English-only during screening if unfamiliar titles are discarded without assessment.

Cochrane’s search guidance discusses language and other search restrictions as potential sources of bias. A practical audit records the actual restriction and its consequence rather than treating “no language limits” in the methods section as sufficient evidence.

Create an assessment route

Identify languages expected in the topic and available translation support before full screening. Specify which decisions need a qualified reader and how uncertain machine translations will be checked. Automatic translation can help orient a reviewer, but it does not guarantee correct interpretation of eligibility criteria, units or negation.

Keep the original text alongside translated extraction notes. Mark unresolved phrases and record who checked consequential interpretation. Never silently translate a technical label into a different outcome definition merely because it sounds more familiar.

Examine a hypothetical restriction

Suppose a broad search retrieves 500 unique records: 450 in the team’s working language and 50 in other languages. Initial translated screening identifies eight of those 50 as potentially eligible. Full-text assessment confirms three eligible studies, excludes four and leaves one unresolved.

An English-only filter would have removed all 50 records before their relevance was known. The audit cannot conclude that the three studies have different effects merely from their language, but it can show that the filter would have changed the eligible evidence base. The unresolved report should retain its own status, not be counted as an exclusion.

Document constraints honestly

If resources force a restriction, state it in the protocol, explain why and report the number and characteristics of affected records where available. Consider whether collaborators or a narrower scientific question would be more defensible than a broad review with concealed language exclusions.

Before publication, compare the executed searches, screening database and methods wording. Check that translated titles remain connected to their original identifiers and that the same study is not counted twice under translated and original citations. The deliverable is a visible account of how language influenced coverage, eligibility and extraction, with uncertainty retained wherever translation could not support a reliable decision.

Source verification: 6 November 2026. Examples are hypothetical.