Handle Preprints and Later Journal Versions as Linked Evidence

Decide the policy before screening

A preprint is a report version, not automatically a different study. Specify whether preprints are eligible, how their status will be displayed, and when a later journal article will replace or supplement extracted information. Do this before seeing whether a preprint’s findings favor the review’s hypothesis.

Cochrane’s guidance on collecting data treats studies rather than individual reports as the underlying unit and recommends planning how to resolve inconsistent sources. Apply that principle to versions without assuming that the newest report contains every useful detail.

Build a version chain

Assign a stable study identifier. Under it, list the preprint identifier, version date, journal citation, supplementary files and any registration number. Preserve the exact version used for each extraction. A replacement download at the same landing page should not overwrite the audit trail.

Compare the versions using concrete fields: participant count, recruitment dates, outcome definitions, analysis population, exclusions, assessment times, statistical models and estimates. Mark each field unchanged, changed with explanation, or unresolved. Changes in wording alone do not establish changes in the underlying data.

Work through a hypothetical discrepancy

Imagine a preprint reports 200 randomized participants and a 30-day result for 180 participants. The journal version reports the same 200 randomized participants but analyzes 190 at 30 days. The ten-person difference is not ten new trial recruits. Check whether additional outcomes were obtained, an exclusion rule changed, or one report contained an error.

Suppose the later supplement explicitly documents recovery of ten previously unavailable measurements. Record the journal result as the selected analysis under the review’s source hierarchy, retain the earlier extraction, and document why it changed. If no explanation exists, contact the investigators or retain the discrepancy as unresolved rather than averaging the estimates.

Update without double counting

One study represented by a preprint and journal article contributes one eligible study to the review, although both reports may supply information. Keep separate counts of records, reports and studies in the review flow. A protocol may use a different eligibility rule for non-peer-reviewed reports; explain that rule rather than describing peer review as proof of correctness.

Before final synthesis, search for later versions, withdrawals and corrections. Preserve what was available at the stated cutoff. If new information changes an extraction after the cutoff, document the amendment and rerun affected analyses. The deliverable is a traceable version decision, not a claim that publication status alone settles reliability.

Source verification: 6 November 2026. Examples are hypothetical.