A multi-centre study agreement is a short written document, signed before the first participant is recruited, that fixes six things: who owns the data, how data moves between sites, which approvals each site holds, the authorship model and order, who sits in the writing group, and how disputes are settled.
A multi-centre collaboration can run smoothly through recruitment and still break down at the point of writing, when the pooled dataset sits on one server and three sites each believe they were promised second authorship.
What a multi-centre study agreement covers, and what it does not replace
A multi-centre study agreement is a scientific document between collaborators. It does not replace the contract your research office may require: where personal data crosses a national boundary, the sponsor usually needs a formal data use agreement on its own timetable.
Six items belong in it, each harder to settle later:
- Data ownership: who holds the master dataset, and what a site may do with its own.
- Data transfer: what is sent, in what form, under which identifiers.
- Approvals: the ethics approval each site holds.
- Authorship model: individual names, a group name, or both.
- Writing group: who drafts and by when.
- Dispute route: who decides, and where it goes then.
Settle data ownership and transfer before the first patient
Ownership is not the same as access
Data ownership in a multi-centre study usually rests with each institution for the records it generated, while the coordinating centre holds a licence to pool and analyse them for the agreed purpose. Write that split down, then answer: may a site publish a single-centre analysis of its own patients, and when? May the pooled data be reused for a second question? That answer settles your data availability statement.
What the transfer terms have to specify
A data transfer agreement states which variables leave a site, in what de-identified form, to whom, and for how long. Five points make it usable:
- The variable list.
- The de-identification standard.
- The transfer method.
- The deletion date.
- The custodian at each end.
Vagueness in any of them returns as a delay at information governance.
Every site needs its own approval, and the manuscript must name each one
Approval granted to the lead site does not extend to the others. Each centre needs its own ethics committee approval, documented local acceptance of a lead or single review, or whatever institutional authorisation its jurisdiction issues in place of one. Retrospective record reviews are not outside this; they still need a decision from a committee, usually a waiver of consent and in some jurisdictions a formal exemption determination.
Record each committee name, reference number and date. For interventional studies, register the trial at or before the first participant consents to enrol, in a primary register of the WHO ICTRP network or in ClinicalTrials.gov, and name every site there. Reporting requirements for the setting sit in the relevant EQUATOR checklist.
Choose the authorship model before anyone starts writing
Three authorship models work for a multi-centre study; name one in the agreement.
| Model | How the byline reads | What it suits |
|---|---|---|
| Individual authorship | Named authors in a fixed order | Few sites, a small team, everyone meeting the criteria in person |
| Group authorship | The study group is the author; members are listed within the article | Many sites, a long contributor list, no defensible way to rank the middle |
| Hybrid | Named authors, then on behalf of the named study group | A small writing group, with a wide collaborator network behind it |
Before choosing group or hybrid, ask how a group name will appear in the article record. A collaborator expecting their own name to be searchable deserves to hear otherwise now.
Recruiting patients is not, by itself, authorship
Under the ICMJE recommendations an author must meet all four criteria: substantial contribution, drafting or critical revision, approval of the final version, and accountability. Supplying cases satisfies the first only. This is where multi-centre agreements go wrong irreparably: a site promised authorship for recruiting writes nothing, and the corresponding author must choose between a broken promise and a false byline.
Write the promise correctly instead. A recruitment quota can set who is eligible for a named position, confirmed once that person has commented on a draft and approved the final text. Everyone else is a collaborator or acknowledged. Our post on who counts as an author sets out the criteria and the CRediT roles.
Fix the order by rule, not by name
People move institutions, leave research, or stop replying, so an order fixed to named individuals in year one has to be renegotiated in year three, with no rule left to settle it. An order fixed by rule survives: first author writes the first full draft; last author is the chief investigator; second and third go to the two highest-recruiting sites whose nominee met the ICMJE criteria; the rest alphabetical.
Treat the writing group as a job description, not an honour
The writing group is the three to five named people who draft the manuscript and see it to submission. Name what each produces and date each output; a group of nine with no allocation can go months without producing a draft. Record who takes each role:
- Who drafts the methods.
- Who runs the analysis.
- Who builds the tables and figures.
- Who checks the reporting checklist.
- Who is corresponding author.
The last role includes confirming that every author approved the submission, so the corresponding author's responsibilities are worth reading.
Copy this one-page multi-centre collaboration agreement
Complete the table below, then circulate it for signature by one named person per site before recruitment opens.
| Clause | Wording to complete and adapt |
|---|---|
| 1. Study and centre | Study title: ____. Coordinating centre: ____. Chief investigator: ____. Participating sites: ____. |
| 2. Data ownership | Each site retains ownership of the records it contributes. The coordinating centre is licensed to pool and analyse them for the study question above, and for nothing else without all sites agreeing in writing. |
| 3. Data transfer | Sites transfer only the variables in the agreed data dictionary (attached), de-identified to the standard at ____, by ____ (method), by ____ (date). Custodian at each site: ____. Pooled data retained until ____. |
| 4. Local analyses | A site may analyse and publish its own records alone, provided it tells the chief investigator in writing beforehand and does not submit before the primary multi-centre paper, or ____. |
| 5. Approvals | Each site confirms it holds its own ethics approval, documented local acceptance of the lead or single review, or the institutional authorisation its jurisdiction issues instead, before contributing data, and supplies the committee name, reference number and date. Trial registration: ____. |
| 6. Authorship model | The byline will use: individual authorship / group authorship / named authors on behalf of the ____ Study Group. Delete two. |
| 7. Authorship criteria | Named authorship requires all four ICMJE criteria. Contributing cases confers eligibility, not entitlement; a position is confirmed once that person has commented on a draft and approved the final version. Others are collaborators or acknowledged. |
| 8. Order | First author: ____. Last author: ____. Positions ____ to ____ allocated by ____ (state the rule, not names), applied when the first full draft is circulated. |
| 9. Writing group | Members: ____. Corresponding author: ____. Outputs and dates: ____. Comments on a draft are due within ____ days; silence counts as no comment, but every named author must approve the final text in writing. |
| 10. Timeline | Recruitment closes ____. Database lock ____. First full draft ____. Target submission ____. If no draft exists by ____, the chief investigator may reassign the first-author role. |
| 11. Disputes | Disagreements go in writing to the chief investigator, who decides within ____ days. If unresolved, the matter goes to the research integrity office of the coordinating institution, not to the journal. |
| 12. Signatures | One named individual per site, with the date. A site joining later signs the same document before contributing data. |
Why the dispute route must end at an institution, not at a journal
A journal cannot investigate an authorship dispute. It cannot see your recruitment logs or employment records, or compel answers. The COPE Core Practices require a journal to have a process for resolving disputes over authorship or author order; COPE's separate guidance points that process at the authors' institutions rather than at the journal. Clause 11 of the agreement above is the mechanism you control.
What we ask for at submission, and what we do not publish
Directive Publications does not publish a separate multi-centre or collaboration policy, and it would be wrong to imply otherwise. Our editorial policies align with COPE Core Practices, and the requirements bearing on a multi-centre submission are the ordinary ones: the corresponding author confirms that every listed author approved the submitted version, ethics approval is stated, conflicts are declared, and a data availability statement appears. Manuscripts reach us by email, so the corresponding author holds that record.
Before circulating the agreement, read our Author Guidelines and Publication Ethics pages so your clauses match what is asked at submission. For the clause that follows directly from your data ownership decision, see writing a data availability statement; if a byline is already contested, read how authorship disputes are resolved.