Count the research behind the papers.
Five reports may describe only three studies
An initial article, a conference abstract and a follow-up paper can describe the same participants. Keep a study identity map alongside your reference library, so that extra reporting does not become an extra independent sample. [1] [3]
AcademicFix Editorial Team · Sources reviewed: Bu rehberi Türkçe okuyun
This guide is for thesis and article authors preparing systematic reviews or meta-analyses, particularly intervention reviews. It separates three decisions: which documents belong together, whether the study is eligible, and which results enter each synthesis. It does not prescribe a single method for every literature review.
On this page
First decide what you are counting
| Level | Meaning | Example |
|---|---|---|
| Record | A title, abstract or both indexed by a database or website | Two database entries pointing to one article |
| Report | A document providing information about a study | An article, thesis, preprint, abstract or registry entry |
| Study | An investigation with a defined participant group, interventions and outcomes | Research whose initial and follow-up results appear separately |
These definitions follow PRISMA 2020. Removing duplicate records of one report is different from linking separate reports of one study. Keep different publications as different bibliographic entries even when they share a study ID. A report can also describe more than one study. [3] [1] [2] [7]
As a practical check, do not let a DOI comparison make the study decision for you. Different DOIs do not establish independent participants. Similar titles are a reason to investigate, not a final match. [1] [7]
Investigate the link before assigning an ID
Look for a registration or protocol identifier and an explicit reference to the earlier report. Compare recruitment dates, setting, intervention, baseline characteristics and sample sizes. Authors and participant counts can differ across reports; unresolved cases may require investigator clarification. Cochrane describes these checks; the earlier PRISMA explanatory paper also discusses comparing authors, treatments, sample sizes and outcomes. [1] [2] [7]
Use this comparison checklist as an editorial aid:
- Do the reports cite the same registration or protocol?
- Were participants recruited at the same sites and dates?
- Do intervention duration, delivery and comparator agree?
- Does an explained subgroup or follow-up loss account for a smaller analysis sample?
- Does the explicit link agree with the numerical information?
Record both supporting and conflicting evidence. The same team can recruit a new sample; a changed author order does not settle the question either. Keep an unresolved link visible instead of forcing it into a group. [2] [7]
Build a map another reviewer can retrace
The following file structure is an AcademicFix editorial recommendation, not a mandatory software standard or a validated matching algorithm.
- Give every report a stable ID. Link R001, R002 and subsequent codes to DOI, title, document type, URL, filename and discovery sources. Preserve the fact that two databases found the same report after record cleanup.
- Create a comparison row. For a possible R001–R004 link, enter the evidence on each side and the PDF page or registry field. “Looks similar” gives the next reviewer little to verify.
- Assign study IDs from the evidence. Group confirmed links under S001. Mark unresolved cases, the missing information, the person responsible and a follow-up date. Do not force a decision to complete the spreadsheet.
- Review the decision with a colleague. Ask a second reviewer to check overlap and conflicts against the sources. Preserve both reasons when reviewers disagree, and keep a history if study IDs change.
- Connect each result to its source. Store study ID, report ID, outcome, measurement time, comparison and analysis sample size with each effect estimate. A forest plot row should lead back to a particular document and page.
Report how data were collected, how disagreements were resolved and any rules used to choose between conflicting reports. Retain secondary reports rather than discarding potentially useful information. [4] [1] [7]
For the bibliographic part, use the DOI lookup guide. Confirming a publication’s identity does not establish who its participants were.
A worked map: five reports, three fictional studies
All studies, codes and participant counts below are invented for teaching. They are not research findings or a measured error rate. Imagine three separate investigations evaluating a training programme.
| Report | Information in the document | Study ID | Mapping decision |
|---|---|---|---|
| R01 | Study A: initial article; 120 participants at baseline; 3-month results | S01 | Initial results for A |
| R02 | A: 12-month follow-up; same baseline group; 106 analysed; explicit link to R01 | S01 | New follow-up information, not a new independent study |
| R03 | Study B: conference abstract; 84 participants; 3-month results | S02 | Early report of B |
| R04 | B: full article; same 84 participants, recruitment dates and protocol; explicit link to R03 | S02 | More detailed report of B |
| R05 | Same research team, new recruitment period and protocol; 72 new participants | S03 | Independent study C |
The map contains 5 reports and 3 studies. Adding S01’s baseline 120 to its follow-up analysis sample of 106 gives 226 numerically, but not 226 independent participants. The 106 belong to the same baseline group at a later assessment. Deleting R02 would instead lose the 12-month information.
If R03 and R04 give the same 3-month outcome for S02, do not enter them as two independent study rows in the same meta-analysis. Document which report supplies the result and why; retain the other document’s methods and explanations. Rules for selecting across reports and resolving inconsistencies should be explicit. [2] [4]
R05 belongs to the same team but contains 72 new participants from a different investigation. Merging it into S01 merely because the authors overlap would hide study C. The exercise therefore tests mistaken merging as well as mistaken separation.
One study ID does not mean one usable result
Repeated measurements from the same participants are dependent. A standard meta-analysis cannot treat them as independent effects. Separate, planned short- and long-term analyses are one possible approach; methods that account for dependency are another. A linkage table alone does not solve the statistical modelling problem. [5] [4]
In this example, S01 could contribute its 3-month result to a prespecified short-term analysis and its 12-month result to a separate long-term analysis. Keep S01 on both rows so their relationship stays visible. Assess eligibility for each outcome, time frame and comparison; do not assume every study belongs in every analysis.
Where several scales or times fit an outcome definition, specify how results will be selected rather than inventing a rule after seeing their direction. Tabulate what is available. Cochrane discusses prespecified selection; PRISMA 2020 asks reviewers to describe the process and explain changes. [6] [4]
Transfer counts to the correct reporting level
PRISMA 2020 separates duplicate records removed before screening from the studies and reports ultimately included. If this example were the review’s entire included set, the final counts would be 3 studies and 5 reports. That does not reveal the number of search records. R02 does not become a duplicate database record just because it shares S01 with R01. [1] [4]
PRISMA is a reporting guideline, not a certificate of methodological quality. Completing a diagram cannot establish that linkage or analysis decisions were correct. [3]
Before submission, trace one result backwards: effect estimate → measurement time → study ID → report → source page. Then check whether that study enters the same analysis twice as if the rows were independent. This is an editorial audit exercise, not a substitute for reviewing the analysis method.
Keep source records connected to your writing with the Word and Zotero workflow. Next, use the reference-list audit to check cited works against listed entries. Browse the English guide hub for related tasks.
Sources and limits
Sections below were read on 4 November 2026. Cochrane chapters share one organizational source; PRISMA papers belong to one guideline programme. They are not counted as seven independent organizations. The fictional map and proposed file workflow are AcademicFix teaching material. The 2009 paper supplies historical methodological corroboration; current reporting should use PRISMA 2020.
- Cochrane Handbook, chapter 4. 4.6.1–4.6.3: studies, linkage criteria and record deduplication.
- Cochrane Handbook, chapter 5. 5.2.1: study/report relationships and uncertain linkage.
- PRISMA 2020 statement. Scope and Box 1: record, report and study; reporting versus methodological assessment.
- PRISMA 2020 explanation and elaboration. Items 9, 10a and 16a; extensions accounting for dependency. Full text supplied by Europe PMC.
- Cochrane Handbook, chapter 6. 6.2.4: repeated observations and alternative follow-up strategies.
- Cochrane Handbook, chapter 9. 9.3.3: multiple measures/time points and prespecified selection.
- PRISMA 2009 explanatory paper, methodological discussion. Items 6 and 10: multiple reports, linkage and inconsistent information. Historical corroboration; use PRISMA 2020 for current reporting.