How to select studies for a systematic review: deduplication and PRISMA flow

A concrete task with RigorMed

1. Organise selection decisions into a reviewable PRISMA count table

   These are suggested task inputs and outputs to request, not a new execution, a customer outcome or a guarantee of completion.

   Materials to provide: Prespecified eligibility criteria, sourced search records, actual deduplication, initial screening and full-text decisions, and exclusion reasons.

   Requested outputs / delivery goal: A selection-ledger structure, a PRISMA count-checking table and items awaiting adjudication.

2. An example request

   Distinguish records, reports and studies. Using the actual decisions I provide, organise a selection ledger and PRISMA count-checking table, listing conflicts and missing reasons. Do not perform independent screening that has not been done on my behalf.

3. What still needs review

   This is not a completed systematic review, independent dual screening or meta-analysis. Original-text decisions and disagreement resolution remain the responsibility of the designated reviewers.

Four steps from search results to included studies

1. Prespecify the question and eligibility rules

   Before screening, define population, intervention or exposure, comparator, outcomes, design and necessary date or language limits. Explain restrictions, reviewer roles and disagreement procedures. Pilot the rules on a small set; retain unclear records rather than excluding them automatically because an abstract omits information.

2. Deduplicate records without losing provenance

   Save databases, complete searches, dates, export batches and identifiers. Check duplicate candidates using DOI, PMID, title, authors and year; retain merge mappings and sources. A preprint, conference abstract and journal article may be different reports of one study, not disposable duplicates.

3. Separate initial screening, retrieval and full-text assessment

   Screen titles and abstracts, seek potentially eligible full texts, then apply the prespecified rules. Distinguish reports not retrieved from reports read and found ineligible. Log stage, decision, reason, reviewer, date and resolved disagreements; an AI summary is not a final eligibility assessment.

4. Link reports to studies and reconcile the flow

   Use registration numbers, locations, dates, populations and methods to link reports; retain their additional information. Count records, reports and studies separately. Neither one article per study nor a participant-flow counting scheme can be assumed.

A reconciled synthetic PRISMA example

Assume a new review used only databases and registers, removed only duplicates before screening, and mapped each retained candidate record to one report sought. All numbers are synthetic teaching data: no real search or review was conducted.

Reconcile stages: 1000 − 200 = 800; 800 − 680 = 120; 120 − 8 = 112; 112 − 92 = 20. Linking the 20 included reports identifies 18 studies; subtraction alone cannot establish that linkage. Other sources, automated removals or an updated review need the appropriate official template and separately documented branches.

Six checks before reporting the flow

Check real screening records, not merely the drawing:

• Were eligibility rules, reviewers and disagreement procedures prespecified?
• Are search sources, dates, strategies and duplicate mappings retained?
• Are initial exclusions, non-retrieval and full-text exclusions separate?
• Can exclusion decisions be traced to rules, original texts and records?
• Are multiple reports linked correctly and counted separately from studies?
• Do text, supplements and diagram agree, without confusing literature with participants?

Automation may assist organisation and flagging, but disclose its actual use, human decisions and verification. Software checks do not replace review authors’ responsibility.

Frequently asked questions