A retraction is a journal saying a paper should not be relied on. We followed retracted studies past that point, into the reviews doctors read, a national guideline and the trials that enrol patients. Each story says how far the chain goes, and no further.
How to read these
Five links in a chain.
Every story is rated by the furthest link we can prove. None claims the last one: no record shows which study shaped which patient’s care.
1. In the evidenceA review or guideline counts the retracted study as data.
2. Changes the answerWithout it, the review’s result is different.
3. Reaches a guidelineA guideline or decision tool is based on that review.
4. Cited by studies that enrol peopleA trial enrolled people and cites it as background.
5. Harms someoneA patient was hurt. We never claim this.
Story 1 · reaches a national guideline
Germany’s sepsis guideline repeats a finding that depends on a retracted trial.
In 2025 Germany’s national sepsis guideline based its advice on vitamin C given alone on one review. That review counts a trial its journal had already retracted. The guideline repeats one of the review’s findings: that vitamin C shortened the time patients with sepsis needed drugs to hold up their blood pressure. That finding depends on the retracted trial. Take the trial out and it is no longer statistically significant. The guideline still recommended against vitamin C, so the trial did not push the guideline toward vitamin C. But no one in the chain caught the retraction.
1
2
3
4
5
Furthest link proven: 3 of 5
Oct 2020A trial of 117 patients at one hospital in Hangzhou reports that high-dose vitamin C cut deaths from sepsis and shortened the use of blood-pressure drugs.
5 Oct 2022The journal retracts it. The authors gave no ethics approval number, the trial was never registered, and data in one of its tables (Table 2) were unclear. “None of the authors have responded to any correspondence from the editor or publisher about this retraction.”
5 Mar 2023A meta-analysis that counts the retracted trial is submitted, and published that October. It does not mention the retraction. In its analysis of blood-pressure drugs in sepsis, the retracted trial carries more than half the weight (57%).
2025Germany’s S3 sepsis guideline names that review as the basis for its advice on vitamin C alone, repeats its findings, rates the evidence “very low” and recommends against high-dose vitamin C.
Our re-run
We rebuilt the review’s analysis from its published figure, matched its result exactly, then took out the retracted trial.
Pooled effect on vasopressor duration, with and without the retracted trial
Three trials, as published
−0.59 (−0.99 to −0.19), p = 0.004
Without the retracted trial
−0.36 (−1.16 to 0.45), p = 0.38
Standardised difference in how long patients with sepsis needed blood-pressure drugs, 95% confidence interval. Below zero favours vitamin C. Without the retracted trial the interval crosses zero: no longer clear evidence of an effect. The review’s finding on deaths in sepsis holds without it (risk ratio 0.63, p = 0.009). Two other reviews that count the same trial keep their findings on deaths without it.
Why it mattersSepsis was linked to about 11 million deaths in 2017, one in five deaths worldwide. High-dose vitamin C is one of its most argued-over add-on treatments.
What it is notNot evidence that any patient was harmed, and not a claim that vitamin C helps or harms. The guideline panel was cautious and reached a sound answer. What failed is the check: nothing told the review’s authors, its peer reviewers or the panel that a trial inside it had been withdrawn.
Sources and our working
Lv SJ et al. Early use of high-dose vitamin C is beneficial in treatment of sepsis. Ir J Med Sci 2021;190:1183–1188. doi:10.1007/s11845-020-02394-1
Retraction note, 5 Oct 2022. doi:10.1007/s11845-022-03180-x
Zeng Y, Liu Z, Xu F, Tang Z. Intravenous high-dose vitamin C monotherapy for sepsis and septic shock: a meta-analysis of randomized controlled trials. Medicine 2023;102(42):e35648. Received 5 Mar 2023. Figure 4. doi:10.1097/MD.0000000000035648
Brunkhorst FM et al. S3-Leitlinie Sepsis, Update 2025 (AWMF 079-001), recommendation 8.16 and reference 487. Med Klin Intensivmed Notfmed 2025. doi:10.1007/s00063-025-01317-1
Rudd KE et al. Global, regional, and national sepsis incidence and mortality, 1990–2017. Lancet 2020;395:200–211. doi:10.1016/S0140-6736(19)32989-7
Our re-analysis: Hedges’ g with DerSimonian–Laird random effects, as in the review; it reproduces the published estimate before the trial is removed. Numbers and code are in the census repository.
Right of reply. We wrote to the review’s corresponding author and the guideline’s coordinating author on 14 Oct 2026. Neither had replied by 28 Oct.
Story 2 · caught, two years late
A review counted a trial that its own reference list labels “Retracted”, and was retracted itself.
A 2024 review of adding sildenafil (Viagra) to standard treatment for women in threatened preterm labour said it would leave out retracted articles. Its own reference list gives one trial’s title as “Retracted: …”. It counted that trial anyway, and rated the evidence for three of its four outcomes as high certainty. In February 2026 the journal retracted the review, because of that trial.
1
2
3
4
5
Furthest link proven: 1 of 5
Oct 2018A trial of 239 women at Menoufia University, Egypt, published in BJOG, reports that adding sildenafil to nifedipine delays preterm birth.
27 Nov 2020BJOG retracts it, eight months after an expression of concern about its data. Two other sildenafil-in-pregnancy trials involving the same author were also retracted, in 2020 and 2024.
Jan 2024The review is submitted. Its methods exclude “retracted articles”. Its reference 26 begins “Retracted:”. The trial carries up to half the weight in its pooled results.
7 Feb 2026BMC Pediatrics retracts the review: the editor “no longer has confidence in the results” because a study in it had been retracted. The authors are invited to resubmit without it.
Our re-run
We rebuilt the review’s four analyses from its forest plots and took out the retracted trial. Every conclusion still holds: the smaller trials point the same way. The review’s own sensitivity analyses had already left the trial out for two of its four outcomes, pregnancy prolongation and birth weight, with the same result.
So the journal was right to act, and a corrected review may well reach the same answer. The story is the delay: a warning printed on the page, and two years before the review was retracted.
Why it mattersSildenafil in pregnancy is not harmless. In 2018 a Dutch trial of it for babies growing too slowly in the womb was stopped early, with more lung problems in babies whose mothers took it, and regulators warned against that use.
What it is notNot a claim that sildenafil for preterm labour harms babies. That was a different use. And here the system did correct itself, in the end.
Sources and our working
Maher MA, Sayyed TM, El-Khadry SW. Nifedipine alone or combined with sildenafil citrate for management of threatened preterm labour: a randomised trial. BJOG 2019;126(6):729–735. Retracted 27 Nov 2020 (Retraction Watch database).
Manouchehri E et al. Efficient administration of a combination of nifedipine and sildenafil citrate versus only nifedipine… BMC Pediatrics 2024;24:106. doi:10.1186/s12887-024-04588-3
Retraction Note. BMC Pediatrics 2026;26:102, 7 Feb 2026. doi:10.1186/s12887-026-06556-5
Pels A et al. Maternal sildenafil vs placebo in pregnant women with severe early-onset fetal growth restriction (STRIDER NL). JAMA Netw Open 2020. UK MHRA and Pfizer safety letter, Oct 2018.
Our re-analysis: Mantel–Haenszel and inverse-variance methods as stated in the review; numbers and code in the census repository.
Right of reply. We wrote to the review’s corresponding author on 14 Oct 2026. Dr Mona Larki replied on 24 Oct 2026. Dr Mona Larki sent this statement for publication on 28 October 2026:
The Maher trial was also summarized in Obstetric Anesthesia Digest in 2020. The version available to us did not display a retraction watermark, and its Google Scholar citation record did not indicate that the article had been retracted. These factors contributed to our failure to recognize its retracted status at the time.
Importantly, our review explicitly specified retracted publications among the exclusion criteria, and the Maher trial was excluded in the relevant analyses. However, the availability of a summary of the same trial in another journal, without an apparent retraction notice in the version and citation information available to us, led to this unintentional error.
After discovering the issue, we repeated all four analyses without the Maher trial. These analyses are also reported in our published article and demonstrate that the overall findings and conclusions remained unchanged. We requested a correction, but our request was not accepted. Our ability to respond to the journal’s final decision was affected by the unrest and internet disruptions in Iran at the time. We have since been invited to revise and resubmit the review.
Story 3 · sixteen years after the retraction
The Tai chi trial that would not stay retracted.
In 2010 a rheumatology journal retracted a trial of Tai chi for older women with knee arthritis, seven months after publishing it. The journal withdrew it after the lead author admitted copying text from another team’s paper and the university provided no formal documentation that the work had been done. Reviews of exercise for knee arthritis published in 2024, 2025 and 2026 still count it as one of their trials.
1
2
3
4
5
Furthest link proven: 1 of 5
The recordReviews in 2017, 2024, 2025 and 2026 (two) count it as data. Some reviews did catch it. A 2022 study, for example, left the trial out because it had been retracted, and noted that an earlier review had counted it, where it produced a large effect on a walking test.
What it is notWe have not re-run these reviews, and do not claim their conclusions depend on one trial. The point is how long a retracted paper keeps travelling, and that catching it depends on whether each review team happens to look.
Sources
Ni GX et al. Tai chi improves physical function in older Chinese women with knee osteoarthritis. J Clin Rheumatol 2010;16:64–67. doi:10.1097/RHU.0b013e3181cf344f. Retraction notice doi:10.1097/RHU.0b013e3181e18d34.
Reviews counting it: PLoS One 2017 (PMC5266306); Medicine 2024 (doi:10.1097/MD.0000000000039520); Luo Y et al., J Orthop Surg Res 2025 (doi:10.1186/s13018-025-05973-z); Li Y, Zhai Z, Guo B et al., Front Public Health 2026 (doi:10.3389/fpubh.2025.1710610); Pan R, Mi A, Cheng H, Front Med 2026 (doi:10.3389/fmed.2026.1821838).
The 2022 study that excluded it: Clinical relevance of Tai Chi on pain and physical function in adults with knee osteoarthritis. Sci Prog 2022 (PMC10450487).
Right of reply. We wrote to the corresponding authors of the five reviews on 14 Oct 2026. Dr Luo replied on 15 Oct that the team had not known of the retraction, and on 16 Oct that they would ask the journal for a correction. In Dr Luo’s words: “To confirm: we will proceed with an erratum (correction) to our review regarding the retracted Ni et al. (2010) trial.” The other authors had not replied by 28 Oct.
Story 4 · reaches patients
Studies enrolled patients while citing papers retracted after a misconduct investigation.
In 2022 a journal retracted more than 100 papers by one anaesthetist, mostly short reports on nerve blocks, the injections that numb part of the body for surgery. A sample notice, as reported by Retraction Watch, says an investigation “concluded that no research was conducted and all the data including patient backgrounds were fabricated.” Retraction Watch reported the retractions on 10 Feb 2022; the formal notices for the papers these studies cite are dated April and May 2022. Since then, ten studies registered on ClinicalTrials.gov, nine of them trials, have listed those papers among their references. The registry now marks every one of those references “RETRACTED”.
1
2
3
4
5
Link reached: 4 of 5. Links 1 to 3 are about reviews; this story is about trials.
10
studies (nine trials) registered after the retractions list the retracted papers, in seven countries.
497
people enrolled in the five studies that have finished.
391
more planned in the other five.
The patternAll 11 citations now read “RETRACTED” in the studies’ own registry entries; the registry takes the label from PubMed. Most of these studies test the same block the retracted paper described. One cited paper reports complications in “498 consecutive cases”.
What it is notThe papers are cited mainly to describe a block, beside other evidence; we did not find a study whose design depends on them. Nerve blocks are widely used, and no patient is known to have been harmed. The point is that the citations still stand, years after the retractions.
The wider sweepAcross the 22,553 retracted papers in our census, 219 studies (182 of them trials) registered on ClinicalTrials.gov after a retraction still list the retracted paper. Of the 224 study-paper pairs, about a quarter cite the paper for a side point and a third test the same treatment. We read seven strong cases in full. In three it is one of several supporting references; in none could we show that the design depends on it.
Why the label mattersThe registry shows the warning, taken from PubMed, on the study’s own page. In all ten, the citation is still there. A published review did the same (Story 2).
The ten studies
The ten registered studies that cite the retracted papers
Retraction Watch, “More than 100 of an anesthesiologist’s papers retracted”, 10 Feb 2022, quoting the Journal of Clinical Anesthesia notices.
Retraction Watch database: 124 retractions for the author, 102 of them in the Journal of Clinical Anesthesia. Notice dates from PubMed (for example PMID 35393191, 5 Apr 2022).
ClinicalTrials.gov API, queried 14 Oct 2026 for every retracted PubMed ID in our census frame.
Right of reply. We wrote on 14 Oct 2026 to the four study teams that list a contact (NCT05461092, NCT05377541, NCT05897307, NCT06155097). None had replied by 28 Oct.
The bigger picture · what retraction never catches
93 trials failed a trust check. Journals had flagged 3.
Our stories follow papers that journals have already retracted. Those are the easy cases: the warning exists. The harder question is how much untrustworthy research carries no warning at all. Every independent check we found gives the same answer: far more than retraction ever marks.
1 in 500
randomised controlled trials in PubMed is marked retracted: 1,328 of 672,661 (our count, 14 Oct 2026).
93 of 374
trials in 18 Cochrane reviews were set aside after a trust check. Only 3 had been retracted or flagged by a journal.
22 months
median wait for a journal’s first action after concerns were raised about one research group later found to have fabricated data.
Cochrane’s own checkReviewers re-checked every trial in 18 Cochrane reviews of nutrition in pregnancy and after birth. A quarter could not be confirmed as trustworthy, and six reviews needed updating because their findings changed. In the authors’ words, the result confirms “that the bar for retraction is considerably higher than for trustworthiness assessment.”
When an editor sees the raw dataAn editor at Anaesthesia checked trials submitted to the journal from 2017 to 2020. Where authors sent individual patient data, he judged 44% to contain false data; with only summary tables, he could spot 2%. These were submissions to the journal, not published papers. The point is how much stays invisible without the raw data.
A wider sampleApplying 72 draft trust checks to 95 trials drawn from 50 Cochrane reviews, assessors had some concerns about 25% and serious concerns about 6%; the authors note some checks were hard to apply and may have overstated concerns.
Even proven fabrication is slow to clearBetween 2013 and 2020, researchers sent concerns about 292 papers from one bone-health group to 77 journals. A university investigation later found the group’s lead author had fabricated data. By October 2020, 115 of the papers had any notice. The median wait for a journal’s first action was 22 months.
What researchers sayIn a meta-analysis of surveys, about 2% of scientists admitted fabricating or falsifying data at least once. In a national survey of 6,813 academic researchers in the Netherlands (published 2022), an estimated 8.3% had fabricated or falsified data in the previous three years.
What this is notNot a claim that a quarter of trials are fabricated. “Set aside” mostly means questions the authors never answered. No one can say exactly how much research is false. What every check shows is that retraction marks only a small part of it, and that, as our census shows, even that part keeps being used.
Sources
Weeks J et al. Trustworthiness assessment as an inclusion criterion for systematic reviews: what is the impact on results? Cochrane Evid Synth Methods 2023. doi:10.1002/cesm.12037
Carlisle JB. False individual patient data and zombie randomised controlled trials submitted to Anaesthesia. Anaesthesia 2021;76:472–479. doi:10.1111/anae.15263
Wilkinson J et al. (INSPECT-SR). Assessing the feasibility and impact of clinical trial trustworthiness checks via an application to Cochrane Reviews. J Clin Epidemiol 2025. doi:10.1016/j.jclinepi.2025.111824
Grey A, Avenell A, Bolland M. Timeliness and content of retraction notices for publications by a single research group. Account Res 2022;29:347–378. Hirosaki University investigation (2017), reported by Retraction Watch, 6 Dec 2017. doi:10.1080/08989621.2021.1920409
Fanelli D. How many scientists fabricate and falsify research? PLoS One 2009. doi:10.1371/journal.pone.0005738. Gopalakrishna G et al. Prevalence of questionable research practices, research misconduct and their potential explanatory factors. PLoS One 2022. doi:10.1371/journal.pone.0263023
PubMed counts: randomized controlled trial[pt] and the same AND retracted publication[pt], run 14 Oct 2026.
Honest accounting
What we tested, including what came to nothing.
We re-ran five reviews that counted a retracted trial in a pooled result. In one, a finding that a national guideline repeats is no longer statistically significant without the trial (Story 1). In the other four, the results we tested held, usually because the other trials pointed the same way. One bad trial is often absorbed by the rest. The finding is not that every review is wrong. It is that the checkpoints missed the retraction, again and again.
For reporters
Everything here can be checked.
Each story links its sources. The census data, protocol and code are public, and the data release is archived at doi:10.5281/zenodo.23358099. Census figures are pending human review. To correct anything on this page, or for interviews, data pulls or a story in your beat, write to SaneCite at letters@sanecite.com.