How to Link Multiple Papers That Report the Same Study

A single trial can surface in your search results as a protocol, a primary outcomes paper, a secondary analysis, a conference abstract, and a long-term follow-up. If you count each as a separate study, your evidence table inflates, your risk-of-bias assessment fragments, and your synthesis double-counts participants. The fix is a deliberate linking step: extract a small set of identifying fields from every candidate report, group reports that share those fields, and then choose which report to trust for each outcome you need.

This article gives you a repeatable sequence for that work, with diagnostic checkpoints, a worked example, a reusable extraction table, and stop rules for when the evidence is too thin to link confidently. The method follows the logic Cochrane recommends for selecting studies: plan from eligibility criteria, search multiple sources, document the process, deduplicate, screen in stages, and explicitly link multiple reports of one study.

Cochrane's guidance treats "linking multiple reports of one study" as a named step in study selection, not an afterthought. That placement matters. Linking sits after deduplication and screening, and before you finalize which studies enter your review. If you skip it, two failures follow. First, the same participants appear twice in your narrative or meta-analysis. Second, you may attribute an outcome to the wrong report, because the paper that first reported the trial is often not the paper that reports the outcome you need.

A second reason is auditability. Reviewers and editors increasingly ask how you moved from a raw record set to a final study list. A linking log — one row per report, one group ID per study — answers that question directly and makes your selection reproducible.

Cochrane recommends planning searches from eligibility criteria. In practice, this means writing down, before searching, which populations, interventions, comparators, outcomes, and study designs you will accept. For a linking task, add one more line: which report types you will treat as eligible (for example, primary reports and secondary analyses) and which you will treat as companion documents only.

This decision determines what counts as a "study" versus a "report." A trial registration record and a protocol are usually companion documents; the primary results paper is usually the anchor report. Decide this in advance so you are not making it up mid-screening.

Cochrane recommends using multiple sources and combining free-text terms with subject headings. PubMed supports field tags, phrase searching, filters, Advanced Search, and saved searches, and its Search Details panel shows how your terms were translated and combined — a useful record of what you actually ran. Google Scholar supports searching by author and title, date sorting and limits, Cited by, Related articles, and locating alternative versions of a work, which helps when a trial has several reports under slightly different titles.

Record the date, database, exact query string, filters, and result count for each search. This is the backbone of your method trail. If you want a scholarly search interface to run and revisit these queries, you can start from WisPaper search or read the public entry point for finding academic papers.

Checkpoint: Before screening, confirm you can reproduce each query from your log. If you cannot, fix the log now — not after you have screened 400 records.

Cochrane recommends deduplicating records and screening in stages. A practical staging is: title/abstract screen, full-text screen, then linking. Keep the linking step separate from eligibility screening. A report can be eligible for the review and still need to be grouped with its siblings; a report can be ineligible on its own but still tell you which study it belongs to.

During screening, capture the reason for every exclusion. When you reach full text, begin extracting the identifying fields described next — even from reports you may later exclude, because they can help you recognize a study's other reports.

The core of linking is a small, consistent extraction. For every candidate report, record:

1. Trial identifier — registration number or other study ID, exactly as printed.
2. Sample size — the randomized or analyzed N, with the population described.
3. Recruitment dates — the enrollment window, not the publication year.
4. Authors — at minimum the first author and the senior author, plus any overlapping author group.
5. Sites — countries, centers, or setting.
6. Interventions — the arms, doses, and comparators as described.

No single field is decisive. Trial identifiers are the strongest signal when present and consistent, but many reports omit them or print them inconsistently. Sample size and recruitment dates together are a strong pair, because a genuine companion report usually describes the same enrollment window and the same or a subset population. Overlapping authorship and sites add corroboration. Interventions should match or be a clearly described extension.

Checkpoint: If two reports agree on identifier but disagree on sample size, do not assume error. A secondary analysis may legitimately use a subgroup. Record the discrepancy and resolve it at the outcome-selection step.

Create one row per report and one group ID per study. Sort by trial identifier first, then by recruitment dates and sample size, then by author overlap. Where no identifier exists, group on the combination of dates, N, sites, and interventions, and mark the group as "linked without identifier" so the basis is visible.

Keep a short note on why each group was formed. "Same registration number" and "same enrollment window, same N, overlapping authors, no identifier printed" are different levels of confidence, and your note should say which applies.

Grouping is only half the job. For each outcome in your review, choose the report that actually reports it, at the timepoint and in the population you specified. Work outcome by outcome:

• Prefer the report that presents the outcome with the fullest data for your analysis.
• Prefer the prespecified primary report for the primary outcome, and secondary analyses for outcomes they uniquely report.
• When two reports give the same outcome, note both and state which you extracted and why.
• When a later report supersedes an earlier one for the same outcome, record the supersession rather than silently dropping the earlier paper.

This is where a linking table earns its keep: it lets you point to exactly which report supplied each number in your evidence table.

Suppose your search returns four records that look related. Report A is a protocol with a registration number and a planned enrollment of 600. Report B is the primary results paper, prints the same registration number, reports 598 randomized across 12 sites in a 2019–2021 enrollment window, and gives the primary outcome. Report C is a secondary analysis of a biomarker subgroup, same registration number, same sites, N = 240. Report D is a conference abstract with no identifier, first author overlapping with B, 598 participants, and the same intervention arms.

Your grouping: A, B, C, and D form one study group. A is a companion document. B is the anchor report for the primary outcome. C is the source for the biomarker subgroup outcome, and its smaller N is expected, not a conflict. D is a companion abstract; you would not extract its numbers if B reports them more fully, but you keep it in the log because it corroborates the group.

Your outcome map then reads: primary outcome → B; biomarker subgroup → C; protocol-defined outcomes → A for definitions, B for results. Every number in your manuscript traces to a named report.

Keep this table alongside your screening log so the whole path from record to extracted number is visible in one place.

Cochrane recommends documenting the search and selection process and rerunning searches before publication. Build your trail so a colleague could repeat it:

• Save every query string with its date, database, and filters.
• Keep the raw record export before deduplication.
• Keep the screening log with exclusion reasons.
• Keep the linking table with group IDs and linking basis.
• Keep the outcome map that ties each extracted number to a report.
• Note the date of any search rerun and what changed.

If your team works from a shared set of PDFs, you can save and organize the reports in My Library so the linking table and the source files stay together. For questions that depend on the specific papers your team has collected, Library QA lets you ask questions based on papers uploaded to My Library or added from search results.

Linking is a judgment task, so define in advance when to stop and when to escalate.

Stop and record as a single report when no other record shares an identifier, enrollment window, sample size, author group, sites, or interventions. Mark it as unlinked and move on.

Stop and flag as uncertain when fields partially overlap — for example, same authors and intervention but different dates and N. Record both possibilities and state the uncertainty in your notes rather than forcing a merge.

Escalate to a second reviewer when a merge decision would change which studies enter the review, when two reports conflict on sample size or dates in a way you cannot explain, or when a trial identifier appears on reports that otherwise look unrelated.

Escalate to the study authors or registry when the identifier is ambiguous or the report set is incomplete and the outcome you need is missing. Cochrane's emphasis on multiple sources supports checking registries and companion documents before concluding a report does not exist.

Rerun before submission. Cochrane recommends rerunning searches before publication; a new report of an already-linked study can change your outcome map even though your study count stays the same.

Linking is mechanical enough to benefit from tooling and judgment-heavy enough to keep under your control. A scholarly search interface helps you run and revisit documented queries; a library keeps your report set together; question-answering over your own collected papers helps you check what a specific report says without re-reading it end to end. For longer literature-review tasks, Scholar Agent is the area for extended research work. If you are coordinating the linking log across a team, Research Projects is where you can create and manage a project. And before submission, when your reference list is final, TrueCite checks BibTeX against real academic databases to flag hallucinated references — a useful last pass after your linking table is settled.