Dual screening systematic review protocol: workflow and guidance

A dual screening systematic review protocol specifies how two reviewers will independently assess the same records for eligibility, how they will handle disagreements, and how screening decisions will be reported. This is different from one reviewer screening records while another checks only the exclusions, so the protocol should name the actual workflow rather than describe both approaches as “dual screening.” [2]

Dual screening can help reduce the risk of missing eligible studies, but it does not guarantee accurate selection. Its value depends on clear eligibility rules, independent decisions, and a prespecified process for resolving disagreements. PRISMA 2020 asks authors to report how screening was conducted; it does not require two screeners for every review. [1][2]

What does dual screening mean in a systematic review protocol?

Dual screening means two reviewers independently make initial eligibility decisions on the same records. [2] Independence means each reviewer assesses a record without relying on the other reviewer’s decision before submitting their own.

A protocol should distinguish this process from verification, where one reviewer screens and a second checks some or all exclusions. These approaches differ in what the second reviewer assesses and should not be treated as interchangeable. [2] State which approach applies at each stage, such as title-and-abstract screening and full-text assessment.

It also helps to define the units being screened. A record is a database or other search result, a report is a document describing a study, and a study is the underlying investigation; one study can have multiple reports. [2][3] Keep deduplication separate from eligibility decisions, and preserve these distinctions in the protocol and study-selection reporting. [2][3]

What does the evidence say about missed studies and screening trade-offs?

The available estimates show that screening performance can differ between single and dual screening, but they are study-specific rather than universal error rates. In one randomized crowd-based study, single-reviewer abstract screening had 86.6% sensitivity, compared with 97.5% for dual screening; specificity was 79.2% for single screening and 68.7% for dual screening. [4]

Sensitivity describes how well the process identifies eligible studies; specificity describes how well it excludes ineligible ones. In that study, dual screening was associated with higher sensitivity and lower specificity. These results illustrate a trade-off, not a prediction for every review, reviewer team, or topic. [4]

A methodological review of four evaluations reported a median of 5% of studies missed by single screening, with results ranging from 0% to 58%. [5] The reported median also differed by reviewer experience: 3% for more experienced reviewers and 13% for less experienced reviewers. [5] These findings reinforce that screening accuracy depends on context and should not be summarized as a single expected miss rate.

Duplicate assessment can reduce missed eligible studies, but it cannot repair a weak search, unclear or biased eligibility criteria, or a shared misunderstanding of the criteria. Agreement between reviewers is not proof that their decisions are accurate. [2][5] A protocol should therefore address the quality of the search and eligibility rules as well as the number of reviewers.

When is dual screening required, recommended, or a justified shortcut?

Whether dual screening is required depends on the review guidance being followed; it is not a universal requirement for all systematic reviews. PRISMA 2020 is a reporting guideline: it asks authors to report how many reviewers screened each record and report, whether they worked independently, and how disagreements were handled. [1][2]

For Cochrane intervention reviews, two independent people are required for the final full-text eligibility decision. Cochrane describes duplicate title-and-abstract screening as ideal while allowing initial screening by one person. [6] Review teams should apply the requirements relevant to their review type and state any differences between stages.

Rapid-review guidance offers a separate, context-specific option: updated Cochrane guidance generally recommends independent dual screening of at least 20% of records, with agreement checked and discrepancies discussed. [7] This is guidance for rapid reviews, not a universal threshold for full systematic reviews. [7] If a team uses a partial or single-reviewer approach, it should explain the method and its rationale rather than imply that it conducted full independent duplicate screening.

How should you prespecify and calibrate the screening workflow?

A defensible protocol defines operational eligibility rules, pilots the process, screens independently where planned, and specifies how uncertainty and disagreements will be handled. [2][8] Operational rules turn broad criteria into instructions reviewers can apply consistently—for example, how to classify a record when its title and abstract do not provide enough information.

Before the main screening, reviewers can use a pilot to identify ambiguous criteria and discuss how to apply them. The available guidance supports calibration, but does not establish a universal pilot size, agreement threshold, or rule for changing the workflow if calibration is poor. [2][8] The protocol should describe the team’s planned approach without presenting an unsupported threshold as a general standard.

For title-and-abstract screening, prespecify that a record will move forward when either reviewer considers it potentially eligible or uncertain. [2][8] This approach avoids excluding a record solely because one reviewer judged it ineligible. At full-text stage, specify how each report will be assessed and how unresolved eligibility questions will be referred for resolution.

If the workflow changes after screening begins, report what changed and where the change applied. Transparent reporting lets readers distinguish planned methods from later decisions and assess how the selection process was conducted. [1][2]

How should disagreements, exclusions, and records be documented?

Document who screened at each stage, whether decisions were independent, how conflicts were resolved, and how many records and reports proceeded through selection. [1][2] A flow diagram can show selection counts, but it does not by itself explain the screening method. [1][2]

Set the disagreement-resolution process in advance. For example, specify whether reviewers will discuss conflicts and who will make a decision if they cannot reach agreement. The protocol should also say how uncertain records are treated at initial screening and how full-text exclusion reasons are recorded. These details make the process reproducible without assuming that agreement alone establishes accuracy. [2][5]

Track duplicates separately from screening decisions, and avoid collapsing records, reports, and studies into one count. One study may be represented by multiple reports, so the protocol and final reporting should make clear what each count represents. [2][3] This distinction helps readers follow how search results were deduplicated, assessed, and linked to included studies.

What can screening automation support—and what still needs human judgment?

Automation can be framed as support for an auditable, human-governed screening workflow, not as a substitute for prespecified eligibility rules or independent judgment. The available evidence here does not establish comparative performance for screening software or machine-assisted decisions, nor does it specify how to validate those decisions. Teams should not infer a validated automation method from the screening estimates cited above.

If a platform supports screening, document how its outputs relate to reviewer decisions and how disagreements are handled. Keep the protocol clear about who makes eligibility decisions and what process applies when a record is uncertain. Human oversight remains important because even two reviewers can share a misinterpretation, and agreement is not proof of accuracy. [2][5]

Key takeaways

• Dual screening means two reviewers independently assess the same records; having a second reviewer check exclusions is a different workflow. [2]
• PRISMA 2020 calls for transparent reporting of screening methods but does not mandate two screeners for every review. [1][2]
• Screening sensitivity and specificity estimates vary by context and should not be treated as universal error rates. [4][5]
• Prespecify eligibility rules, treatment of uncertain records, reviewer independence, and disagreement resolution. [2][8]
• Report selection counts while distinguishing records, reports, and studies. [1][2][3]
• Treat automation as workflow support, not a replacement for human eligibility decisions.

FAQ

Does reviewer agreement show that screening decisions are correct?

No. Agreement shows that reviewers made the same decision, not that the decision is accurate. Both reviewers may share a misunderstanding of the criteria, so clear rules and a process for resolving uncertainty remain important. [2][5]

Should every title-and-abstract decision be made by two reviewers?

Not necessarily. Cochrane describes duplicate title-and-abstract screening as ideal for intervention reviews while allowing initial screening by one person; its requirement for two independent people applies to the final full-text eligibility decision. [6] State the workflow used at each stage.

Can rapid-review screening recommendations be applied to a full systematic review?

Not as a universal rule. Updated Cochrane rapid-review guidance generally recommends independent dual screening of at least 20% of records, but that recommendation is specific to rapid reviews and is not a universal threshold for full systematic reviews. [7]

Why report both screening counts and the screening method?

Counts show how records and reports moved through selection, while the method explains who made eligibility decisions and how conflicts were handled. A flow diagram alone does not describe those procedures. [1][2]