Managing Large-Scale Reviews

To manage a large-scale systematic review, deduplicate into one corpus, give one coordinator a single screening queue, and release records in named batches to fixed pairs of reviewers. Both people decide on the same batch independently before either opens the next. Disagreements follow an escalation path the protocol names before screening starts, and only cleared full texts go to a second pair for duplicate extraction. Large-scale reviews fail on coordination, not on search syntax: a fully written search can still stall when several thousand records have no owner.

Who holds the queue

Role	Holds	Does not hold
Protocol owner	The protocol, and the decision when a pair still cannot agree	Day-to-day assignment of records
Coordinator	One list: batch name, who has it, and whether it is waiting, in dual review, disagreed, or cleared	A private pile nobody else can see
Screening pair	Independent title and abstract decisions, then independent full-text inclusion decisions, on the batch they were both given	The next batch, until this one is finished or escalated
Extraction pair	Independent outcome data on reports that have already cleared inclusion	Records still disagreed, or still in screening

MECIR C39 is mandatory: at least two people decide independently whether each study is eligible, and the disagreement process is defined in advance. Dual title-and-abstract screening is desirable, not mandatory. This plan does it anyway, because that is where an informal pile loses records. MECIR C46 makes duplicate extraction of outcome data mandatory. MECIR C45 makes duplicate extraction of study characteristics highly desirable. This plan uses the same batch rule for both, so extraction does not become a second informal pile.

A worked plan for several thousand records

Copy this into a protocol. It is not a completed review, and it does not prescribe a batch size. The scale is a search that returned several thousand records. The batch is the slice both reviewers can finish before the next one is released. The rule is the shared slice, not a particular count.

1. Deduplicate first, so the queue is one corpus. The deduplication guide is that step. Screening each database export on its own is how one paper is decided twice and another is skipped.
2. The coordinator gives one named batch to both screeners. The protocol states how many records are in it. Neither person opens the next batch until this one is cleared or escalated.
3. Each screener decides without seeing the other’s decision. Screening strategies is the two-stage pass: a liberal title-and-abstract screen, then full text against the eligibility criteria.
4. Disagreements are listed on the batch. The pair discusses them. If they still disagree, the record is escalated and the protocol owner decides. A Maybe from either screener goes into the same resolution queue as a disagreement, not a soft include. C39 requires that path to be defined before screening starts. A named third person is one way to write it. An unwritten path is the failure.
5. Only a cleared full text goes to the extraction pair, who use the same batch rule for outcome data. A disagreed record is not extracted. When the two extractors' values differ, a field-level adjudication ladder is one written way to resolve each cell.
6. An idle screener waits, or helps close the open batch. They do not start a private pile so the queue looks busy.

The best-practice sequence starts with the protocol that should name these roles. A browser draft of a search is not the queue. Saving your SRT work covers the search export, the deduplication audit, and the flowchart counts. The screening list is the shared file or screening tool the coordinator updates. This site does not store that list.

What an informal queue does

An informal queue is everyone screening ad hoc from a shared folder, a chat thread, or a library with no status on the record. The same record is opened twice, or by nobody, because each person assumes the other will take it. One screener starts records the other was never given, so the pair no longer holds two independent decisions on the same set. Disagreements sit in a message, and extraction starts on a report that has not cleared inclusion. A replacement screener cannot see what is finished. The number of records screened cannot be rebuilt, because nobody listed who decided what.

Editing the Boolean does not fix that. One list, named batches, and an escalation path with a person already on it does.

Keep the counts the write-up will ask for

The queue is also where the reporting numbers come from. PRISMA 2020 item 16a asks for the path from records identified to studies included, ideally as a flow diagram (Page et al., BMJ 2021). If the coordinator's list records, per batch, how many records were screened, excluded at title and abstract, sought as full text, and excluded with a reason, the flow diagram is a sum of rows rather than a reconstruction months later.

Column on the coordinator's list	Where it ends up
Records in the batch	Records screened
Excluded at title and abstract	Records excluded
Full texts sought, and not retrieved	Reports sought for retrieval, reports not retrieved
Full texts excluded, with the reason	Reports excluded, by reason
Included full-text reports, grouped by study	Reports of included studies, and unique studies included

The identification side (records per database and duplicates removed) comes from the export and the deduplication audit, as in the dedupe-to-flowchart worked example. Trial-register records go in Registers, alongside databases in the identification column. Agency reports, websites, organisations, and citation searching go in the "other methods" column; grey literature searching shows how to log them. When the review is about to be written up, PRISMA reporting guidelines explained maps those numbers to the checklist items. If the deadline is the real constraint, rapid review methods covers which steps can be shortened and the risk each cut adds.