Title and abstract screening is the first selection stage of a systematic review, where two reviewers read the title and abstract of every record returned by the search and decide whether each one is plausibly relevant to the review question. It is a deliberately generous filter: the goal is to discard only records that clearly fail the eligibility criteria and to keep everything that might qualify for the closer look that comes at full text.
Why the first pass is intentionally lenient
An abstract rarely contains every detail you need to make a final inclusion decision. Sample sizes, comparator arms, and outcome definitions are often missing or vague, so a strict reading at this stage would wrongly exclude eligible studies you can never recover. The working rule is simple: when in doubt, keep it in. A false exclusion here is invisible and permanent, while a false inclusion is caught and corrected at full-text screening. This asymmetry is why the first pass prioritises sensitivity over precision.
A worked case shows the rule in action. Suppose the review studies a drug in adults, and an abstract describes the right drug and the right outcome but never states participant age. A strict reading would exclude it for an unconfirmed population; the lenient rule keeps it, because the full text will reveal the age range and the cost of being wrong is asymmetric. Excluding it now, if it turns out to enrol adults, loses an eligible study forever, whereas keeping it costs only one extra full text to read. The same logic applies to abstracts that omit the comparator or report only a composite outcome: ambiguity at this stage resolves toward inclusion, and the closer reading at full text does the discriminating work that thin abstract metadata cannot support.
Setting up the screen before you start
Lock the eligibility criteria first
Screening decisions are only reproducible when the rules are fixed in advance. Your inclusion and exclusion criteria should map directly onto your PICO question, so each record can be judged against population, intervention, comparator, outcome, and study design without on-the-fly interpretation. Criteria invented during screening are a classic source of bias and a red flag for peer reviewers.
De-duplicate, then divide the work
Run de-duplication of your search results before any human reads a record, so the same study is never screened twice under two database identifiers. Then load the unique records into a dedicated platform. Working in a structured tool keeps the audit trail intact and produces the counts you will need later for your flow diagram.
Screening in duplicate and piloting your judgement
The standard is double screening: two reviewers assess each record independently and blind to each other’s votes. The question of how many reviewers you need for screening has a clear default answer of two, with a third resolving disagreements. Before screening the full set, pilot on a sample of 50 to 100 records and compare votes. Where the two reviewers diverge, the criteria are usually ambiguous rather than the reviewers careless, so refine the wording until agreement is strong.
Quantifying agreement
Piloting is also where you should measure inter-rater reliability in screening with a chance-corrected statistic such as Cohen’s kappa. A low value tells you the criteria are not yet operational; you fix the rules before, not after, screening thousands of records. You can compute the statistic quickly with our Cohen’s kappa calculator.
Handling conflicts and recording the count
Every disagreement between the two reviewers is a conflict that must be settled, never silently dropped. The approaches to resolving screening conflicts range from discussion between the pair to adjudication by a third reviewer. At this stage you do not record reasons for exclusion; title-and-abstract exclusions are reported only as a single number in the PRISMA flow diagram. Detailed exclusion reasons are required only at the full-text stage.
Speeding it up without cutting corners
Large reviews can return many thousands of abstracts, and reading them twice is the most labour-intensive part of study selection. Purpose-built screening tools for systematic reviews keep the two reviewers blind to each other, surface conflicts automatically, and offer machine-learning prioritisation that pushes the likely-relevant records to the top of the queue. Used carefully, that ordering shortens the screen; it must never be used to stop reading records early without a validated stopping rule, which would reintroduce the very selection bias the duplicate screen is meant to remove.
A practical first-pass workflow
The stage runs cleanly when it follows a fixed order rather than ad hoc reading:
- Confirm the de-duplicated record count against your search log, so the number entering screening is the one that will anchor the flow diagram.
- Load the unique records into a screening platform and enter the locked eligibility rules where both reviewers can see them.
- Pilot 50 to 100 records, compute agreement, and refine any criterion that drove disagreement before touching the rest.
- Screen the full set independently and blind, applying the keep-if-in-doubt rule so borderline records pass to full text.
- Resolve every conflict, then move the survivors forward and record the single excluded count for reporting.
Common first-pass mistakes
A few errors recur often enough to be worth naming. Strict reading too early excludes eligible studies whose abstracts simply lacked detail, and those losses are invisible and permanent. Inventing criteria mid-screen to dispatch an awkward record breaks reproducibility and is a red flag for peer reviewers. Recording exclusion reasons here wastes effort on information too thin to attribute, since itemised reasons belong only at full text. And screening solo at this volume lets one tired reviewer’s misses go uncaught. A team that pilots its rules, keeps the filter generous, and screens in duplicate avoids all four, because the discipline lives in the question and the criteria rather than in heroic effort at the abstract stage. If the volume is more than your team can read twice, our managed screening service can run the pass in duplicate to a documented standard.
Where the first pass sits in the wider process
Title and abstract screening is one stage in a longer pipeline. It follows the search and feeds full-text assessment, after which surviving studies move on to data extraction and risk of bias assessment. Getting this first filter right protects everything downstream: include the wrong records and you waste effort, exclude the right ones and your synthesis is quietly incomplete.