Abstract Screening vs. Full-Text Screening: Where Should the Decision Happen?

Screening decisions belong at two different points, and the right point depends on what the record can actually tell you. Abstract screening is the fast, broad filter: you decide from titles and abstracts whether a record is plausibly eligible. Full-text screening is the slower, narrower confirmation: you decide from the complete report whether the study meets every eligibility criterion. Cochrane's guidance describes screening in stages, and PRISMA 2020 asks reviews to report search and selection methods transparently. Put together, those two expectations point to a simple rule: make the exclusion decision as early as the evidence allows, and make the inclusion decision only when you have seen enough of the study to defend it.

This article compares the two stages on the same task — deciding whether a record enters your review — using the same criteria: speed, available evidence, uncertainty, reason codes, and escalation rules for borderline records. It then gives a decision table and a combined workflow you can adapt to a manuscript, a peer-review response, or a reproducible search.

Set up one task so the comparison is fair. Your review asks: does this record meet the eligibility criteria? At abstract screening you have a title, an abstract, and perhaps keywords. At full-text screening you have the methods, the population, the outcomes, and the limitations. The question is identical; the evidence behind the answer is not.

That asymmetry is the whole design. Cochrane recommends planning searches from eligibility criteria and screening in stages, which implies the criteria should be written so that some can be judged from an abstract and others cannot. If a criterion can only be assessed from a methods section, it does not belong in the abstract stage as a pass/fail test.

Abstract screening is the throughput stage. Because the evidence is thin, a decision is fast, and a large deduplicated record set can be reduced to a manageable pool. Cochrane's process includes deduplicating records before screening, which keeps the fast stage from being spent on duplicates.

Full-text screening is the depth stage. It is slower per record by design, because you are reading methods and results rather than a summary. The practical consequence is that you want the full-text stage to receive only records that survived a defensible abstract filter — otherwise the slow stage absorbs work the fast stage could have removed.

Where each stage fails on speed:

• Abstract screening fails when it is asked to resolve questions the abstract cannot answer, producing slow deliberation over thin text.
• Full-text screening fails when it is used as the first filter, because the per-record cost is paid on records that a title-and-abstract check would have excluded.

Sort your eligibility criteria by where the evidence lives.

Cochrane's guidance on linking multiple reports of one study is a useful signal: study-level identity is often a full-text judgment, not an abstract one. If two records may describe the same study, the abstract stage should not be the place where that is finally settled.

Uncertainty is not a reason to skip a stage; it is a reason to route a record forward. The cleanest rule is:

• At abstract screening, uncertainty means include. A record you cannot confidently exclude should advance, because the full text is where the missing evidence lives.
• At full-text screening, uncertainty means document and resolve. Record which criterion was unclear and what evidence would settle it, then resolve it against the full report.

This keeps the fast stage from making irreversible exclusions on thin evidence and keeps the slow stage from becoming a dumping ground for records that never had a chance.

Reason codes are what turn a screening decision into an auditable method trail. PRISMA 2020's flow-diagram templates and selection reporting make the expectation concrete: a reader should be able to see how records moved from identification to inclusion.

Use a small, stable code set at each stage.

Abstract-stage codes (exclusion-oriented):

• ABS-TOPIC — outside the review question
• ABS-DESIGN — design clearly not eligible
• ABS-DUPLICATE — duplicate of an already-handled record
• ABS-ADVANCE — uncertain; sent to full text

Full-text-stage codes (criterion-specific):

• FT-POP — population ineligible
• FT-DESIGN — design ineligible on full reading
• FT-OUTCOME — required outcome absent
• FT-LINK — same study as another report; linked, not excluded
• FT-INCLUDE — meets all criteria

Two habits keep the trail reproducible. First, record the code, not just the decision, so a later reader can see why. Second, keep the codes stable across the project so counts in your flow diagram reconcile with your screening log.

Escalation is the mechanism that keeps borderline records from being decided by fatigue. Write the rules before screening starts.

1. Advance on abstract uncertainty. If the abstract does not clearly trigger an exclusion code, send the record to full text.
2. Escalate on full-text ambiguity. If a full-text record is unclear on a single criterion, flag it rather than forcing a code.
3. Link before excluding. If a record may be a second report of an included study, apply FT-LINK and resolve at the study level.
4. Re-check the search before finalizing. Cochrane recommends rerunning searches before publication; a late-arriving record should pass through the same staged rules.

These rules are where the two stages connect. The abstract stage's job is to protect the full-text stage from noise; the full-text stage's job is to protect the review from premature inclusion.

Abstract screening fails when:

• Eligibility criteria include details that abstracts rarely report, so exclusions are made on absence of information rather than evidence of ineligibility.
• Duplicates are not removed first, so the fast stage is spent on repeated records.
• Uncertainty is treated as exclusion, silently dropping records that would have qualified.

Full-text screening fails when:

• It is the only stage, so per-record cost is paid on records a title check would have removed.
• Multiple reports of one study are counted as separate studies because linking was not part of the stage.
• Reason codes are not criterion-specific, so the flow diagram cannot explain why records left the review.

Use one worked example: a research team preparing a systematic review manuscript and expecting a methods-focused peer review.

1. Write criteria with stage tags. Mark each criterion as abstract-visible or full-text-only. This is the planning step Cochrane's search guidance implies.
2. Deduplicate before screening. Remove duplicates so the fast stage is spent on distinct records.
3. Run abstract screening with exclusion codes and an advance default. Uncertain records get ABS-ADVANCE.
4. Run full-text screening with criterion-specific codes. Link multiple reports rather than counting them separately.
5. Reconcile counts against the flow diagram. PRISMA 2020's templates give you the structure; your reason codes give you the numbers.
6. Rerun the search before submission. Route late records through the same staged rules.

For the search step itself, WisPaper's scholarly search interface is where you find academic papers to feed into this pipeline. As records accumulate, My Library is where you save and upload papers and organize them, which keeps the screened set in one place rather than scattered across folders. When you need to interrogate the retained papers — for example, to check what a set of included studies reports on a specific criterion — Library QA lets you ask questions based on papers uploaded to My Library or added from search results. For longer literature-review tasks that sit around this screening work, Scholar Agent covers literature review, inspiration, and reproduction use cases.

One more check belongs at the manuscript stage: if your reference list was assembled from BibTeX, TrueCite checks BibTeX against real academic databases to flag hallucinated references — a useful final pass before the reference list goes to reviewers.

Reviewers often ask why a record was excluded. A staged log answers that directly: the reason code tells them which stage made the call and on what evidence. If a reviewer challenges an abstract-stage exclusion, you can show whether the abstract actually contained the disqualifying information or whether the record should have advanced. If they challenge a full-text exclusion, the criterion-specific code shows which criterion failed. NIH's rigor and reproducibility guidance treats transparent methods and careful assessment of prior knowledge as core to rigorous practice, which is exactly what a staged, coded screening log demonstrates.