Your live review is citing dead papers
Two publisher clean-ups in the second half of August account for 133 retracted papers between them.
Elsevier's Ceramics International passed 56 retractions for 2026 on 20 August. The stated reasons are image overlap, data integrity problems and paper-mill output, with 15 cases involving suspicious authorship changes during peer review. Forty-three of the notices cite PubPeer directly. The whole sequence was set off by an email from two sleuths in April 2025.
Separately, on 17 August, Sage retracted 77 papers from two journals it had acquired from IOS Press, citing third-party involvement, peer-review manipulation, citation manipulation and tortured phrases. Sage acknowledged that it holds limited information about how these papers were originally published, which is a candid thing for a publisher to say and a genuinely difficult position to be in.
Neither batch is in a field most readers here work in. The number that matters is not the count. It is the interval.
Sixteen months
Concerns raised in April 2025 produced retraction notices in August 2026. That is the normal shape of the process. Investigations are slow because they have to be, editors are cautious because retraction is a serious accusation, and acquired journals come with archives nobody on the current staff assembled.
Now put a systematic review alongside that timeline. Searches run in month one. Screening finishes in month three. Data extraction and risk-of-bias assessment run through month six. Peer review adds another six. A review submitted a year after its search date is unremarkable; two years is common.
Somewhere in that window, papers you included get flagged, investigated and retracted, and nothing tells you. The version of the literature you screened is a snapshot that is already ageing when you take it, and the reference list you submit was accurate at a point in the past.
The problem is worse than the retraction itself. A retracted trial that stays in a meta-analysis is not just an embarrassing citation. It contributes to a pooled estimate, weights the analysis in proportion to a sample size that may be fictional, and shows up in a Summary of Findings table that clinicians read.
Four points where a re-check costs almost nothing
At protocol stage, decide in writing how retracted and flagged studies will be handled, including expressions of concern, which are not retractions and require a judgement rather than an automatic exclusion.
At full-text screening, check each included study once. Crossref carries retraction metadata, the Retraction Watch database is queryable, and Zotero will flag retracted items in a library automatically using that data.
Before submission, re-run every included study. This is the check that gets skipped, and it is the one that most often catches something, because it is the furthest from the search date.
At revision, run it again. Peer review commonly adds six months to a year, and reviewers do not check this for you.
For anything published in a journal with a known integrity problem, PubPeer is worth a look as well. It runs ahead of the formal record by exactly the interval described above, and a paper with substantive unanswered comments is at minimum worth a sensitivity analysis.
Who had access to the data?
A separate argument from 25 August is narrower and more interesting.
Rhodes, Schumacher and Shoag argue that researchers using the TriNetX dashboard never touch raw data and do not control the analytic decisions, yet routinely sign the standard statement that the authors had full access to the data and take responsibility for its integrity. More than 4,400 PubMed articles referenced the platform as of 19 August. They call for mandatory disclosure of dashboard use, contributor credit for the platform staff who do control those decisions, and correction of misleading access statements.
An honest caveat: their piece is about authorship and accountability. It does not discuss meta-analysis, and the relevance to evidence synthesis is inferential rather than stated.
The inference is worth making anyway. A fast-growing body of studies drawn from the same federated network, analysed through the same dashboard, on overlapping patient populations, is precisely the kind of literature a systematic review will pool. Standard practice treats each published study as an independent observation. In this case they may share source data, analytic defaults and even the same underlying patients, and there is currently no way for a review team to assess how much overlap exists.
If your review is likely to include several TriNetX studies, that is a question to raise at protocol stage rather than discover at peer review.
Paper mill or paper mine
The Journal of Clinical Epidemiology published a paper on 20 August offering a tentative explanation for the surge in Global Burden of Disease database papers. The distinction it draws is useful: not fabrication, but industrial-scale extraction from an open dataset, producing large volumes of technically valid, minimally informative analyses.
This is directly an eligibility problem. Any field now flooded with database re-analyses forces a decision about whether they belong in the evidence base at all, and that decision needs making in the protocol rather than improvised during screening. Options include excluding pure database re-analyses, including them but analysing separately, or setting a methodological threshold. All three are defensible. Deciding after seeing the results is not.
The common thread
None of this is about fraud, and the framing of research integrity as a fraud problem is part of why review teams treat it as somebody else's job.
The practical issue is that the published record changes after you read it. Papers are retracted, access statements turn out to describe something other than what happened, and whole categories of study appear faster than methodological norms for handling them.
A review that names its inclusion date, states how retracted studies were handled, and re-checks its references before submission is doing something modest. It is also doing more than most.
References
Degen, P. M., Riebler, A., & Held, L. (2026). Paper mill or paper mine? A tentative answer to the sharp increase in research papers based on the Global Burden of Disease database. Journal of Clinical Epidemiology, Article 112464. https://doi.org/10.1016/j.jclinepi.2026.112464
Orrall, A. (2026, August 17). Sage issues dozens of retractions from two journals it acquired. Retraction Watch. https://retractionwatch.com/2026/08/17/sage-issues-dozens-of-retractions-from-two-journals-it-acquired/
Orrall, A. (2026, August 20). Ceramics journal pulls dozens of papers flagged for integrity concerns. Retraction Watch. https://retractionwatch.com/2026/08/20/ceramics-journal-pulls-dozens-of-papers-flagged-for-integrity-concerns/
Rhodes, S., Schumacher, F. R., & Shoag, J. E. (2026, August 25). Guest post: The uncomfortable truth an explosion of papers reveals about authors' access to data. Retraction Watch. https://retractionwatch.com/2026/08/25/guest-post-trinetx-data-accountability-statement/