@isabelleboutron.bsky.social
Prof of Epidemiology at University Paris Cité Director of the Centre for Research in Epidemiology and Statistics (CRESS) Head of the METHODS team within the CRESS Director of Cochrane France
@theaacr.bsky.social found that 23% of abstracts in manuscripts and 5% of peer-review reports submitted to its journals in 2024 contained text that was probably generated by large language models (#LLMs)
www.nature.com/articles/d41...
#AI #metaresearch
Automated verification of citations could help journals and peer reviewers identify questionable citation practices and reduce propagation of misinformation, improving the trustworthiness of scientific evidence.
In the #PubMedCentral-OA dataset, 34% (981 of 2898) of citation quotations were erroneous. #LLMs (@GeminiApp @GPT4oGPT) can be trained to detect citation errors with moderate accuracy, according to @uofilsystem.bsky.social researchers
#AI
The annotation process underscored the subjective nature of assessing promotional language, particularly for context-dependent adjectives like emerging and latest, and the need for refined constructs to capture gradations of promotional language. Despite these limitations, the pilot study indicates the potential for machine learning models trained on well-annotated datasets to contribute to the automated detection of hype. Future steps include modifying the annotation systems and experimenting with large language models under zero/few-shot regimes.
A pilot study testing the ability of #MachineLearning models trained on well-annotated datasets to automate detection of hype in biomedical research writing (eg ‘groundbreaking', ‘revolutionary’) showed 70-80% accuracy vs human detection
peerreviewcongress.o...
Methods to detect #LLM-generated #PeerReview: in ms, add command to LLM to:
1. Watermark (insert a random start string, fake citation, or technical term)
2. Hidden prompt (insert white-colored text, different language, font manipulation)
@cmu.edu
LLM-generated reviews matched or exceeded human reviewers on a few key dimensions of review quality. A fuller analysis will shed more light on the potential value of LLM peer reviews and how they could complement human peer reviewers’ work.
Comparing #LLM vs human #PeerReview of 5 @bmj.com mss, LLM reviews matched or exceeded humans on key elements of identifying strengths/weaknesses, commenting on writing, organization, and Tables and Figures, and constructiveness
peerreviewcongress.o...
#ZakKohane on #AI in #PeerReview: the surge in scientific publication is straining the conventional review system beyond capacity. @ai.nejm.org piloting AI-augmented fast track review: manuscripts w high prior of being good gets assigned to #GPT5 + a human, disposition reviewed by editorial team!
05.09.2025 13:27 — 👍 8 🔁 4 💬 0 📌 1
Detection of #AI-generated text at 10 @theaacr.bsky.social journals increased Q1 '23 after public release of ChatGPT; a policy prohibiting its use slowed detections in reviewer comments. Presence in mss was assoc'd w a higher rejection rate before peer review @evanko.bsky.social #cancer
03.09.2025 14:35 — 👍 6 🔁 2 💬 1 📌 1
The effect of our intervention was not statistically significant (mean citation rate, −0.007; 95% CI, −0.055 to 0.041) and held for all prespecified sensitivity analyses.
#RetractoBot auto-emails authors when papers they've cited are retracted. An RCT comparing it vs no notice showed no diffs in 1y citations
@bengoldacre.bsky.social @ndevito1.bsky.social @bennettoxford.bsky.social
@ox.ac.uk @oxprimarycare.bsky.social
www.retracted.net/
cc: @retractionwatch.com
The PNN involved 56 journals across 12 publishers, with Springer Nature (n = 29), Elsevier (n = 26), and Bentham Science Publishers (n = 25) having the most articles involved out of the total (n = 123). The involved authors were affiliated with 40 countries and 232 organizations.
The #Pharmakon Neuroscience Network, involving 361 authors from 40 countries in 123 articles/56 journals from 12 publishers 2019-2025, appears to have grown too rapidly to have developed naturally
@fauxsci.bsky.social @helenedraux.bsky.social @digital-science.com
(see also: doi.org/10.48550/arX...)
Unexplained textual similarities (eg, the same unlikely typo appearing in dozens of articles) common to many of these articles suggest the involvement of paper mills, organizations that mass produce, sell authorship on, and publish fraudulent scientific manuscripts at scale.
Scanning electron microscope image (SEM) metadata did not match the SEM manufacturer or model listed in the text in 21.2% of 11.3K materials science and engineering articles, a potential indicator of irreproducibility.
@reeserichardson.bsky.social
Read the full report at journals.plos.org/pl...
Gel blots (eg, Western blots) were the most frequently cited image type in problematic retractions, appearing in 1074 articles (51.68%). Between-article image reuse, where an image and its associated data are duplicated across different publications, was the most common context of misuse, identified in 1241 cases (59.72%). Notably, 982 retractions (47.28%) were attributed to paper mills. Image duplication was the predominant cause of retraction, accounting for 1827 cases (87.92%). Only 1 retraction was attributed to computer-or artificial intelligence–generated manipulation.
Among 2078 retracted articles flagged for problematic image manipulation by @retractionwatch.com, gel blots were the most frequently cited image type (52%); image duplication was most common retraction cause (88%)
@phillcardenuto.bsky.social @loyolachicago.bsky.social @unicampoficial.bsky.social
Unraveling Spin & Selective Reporting in Medical #AI Research: A Cross-Sectional #MetaResearch Study
from #UNSWSydney researchers
is a poster presentation all 3 PRC days
Register to stop by, ask questions, say hello!
Learn more and register at peerreviewcongress.org
@jama.com @bmj.com
Authorship and Contributorship Criteria and Practices at Annals of African Surgery
from EIC James Kigera and AAS colleagues
is a poster presentation all 3 days of the PRC
Register to stop by, ask questions, say hello!
Learn more at peerreviewcongress.org
#AfricanScience
@jama.com @bmj.com
Development of a Tool for Addressing #COIs in Trials (TACIT) for Use in Systematic Reviews
from @alundh.bsky.social @isabelleboutron.bsky.social
@cochranedkcebmo.bsky.social @upcite.bsky.social
is a PRC poster presentation
Register to stop by, ask the authors questions!
peerreviewcongress.org
The latest Bias Methods Group newsletter is live! Dive into fresh insights and essential updates on tackling bias in research, as well as a call for abstracts for the 2025 annual meeting! Read it on the BMG website: methods.cochrane.org/bias/about-u...
19.06.2025 09:01 — 👍 4 🔁 4 💬 0 📌 0
Are you interested in presenting your research at the annual Bias Methods Group meeting? If so, please do not forget to submit your abstract to cochranemethodsbias@gmail.com by August 11. For more information, please click here: methods.cochrane.org/bias/news/20...
19.06.2025 09:04 — 👍 2 🔁 2 💬 0 📌 0
Two modules of the International Summer School in Public Health came to their conclusion yesterday. @cressumr1153.bsky.social and Institut Paris Public Health. After two days of intensive exchanges, workshops and teaching, these modules came to a close in a constructive and collaborative spirit.
12.06.2025 08:45 — 👍 1 🔁 1 💬 0 📌 0A special shout out to:
@rohitarorayyc.bsky.social
@paullusla.bsky.social
@jzsang.bsky.social
for working tirelessly to make this dream happen <3
And our advisors:
@isabelleboutron.bsky.social
David Moher
@geochurch.bsky.social
@nikbobrovitz.bsky.social
Excellent opportunity to learn innovative design, to interact with outstanding researchers in the field of epidemiology and statistics!!!
The international summer school of the @cressumr1153.bsky.social in the centre of Paris
Retour sur la 2ᵉ journée de l’École d’Été en Santé Publique du @cressumr1153.bsky.social & de l'Institut Paris Public Health !
Une journée animée autour d'ateliers pratiques et échanges pluridisciplinaires sur les grands enjeux de santé publique. @inserm.fr @upcite.bsky.social @univ-spn.bsky.social
On March 20, Astrid Chevance of the METHODS team at @cressumr1153.bsky.social and Luz Ascarate (philosopher) organized an interdisciplinary study day entitled “Suffering and pain”.
cress-umr1153.fr/en/rethinkin...
@inserm.fr
@upcite.bsky.social
@univ-spn.bsky.social
@inrae-france.bsky.social
L’École d'été internationale en Santé Publique organisée par @cressumr1153.bsky.social et l'Institut Paris Public Health débute aujourd'hui ! Du 02 au 13 juin, deux semaines de formation de haut niveau en Santé Publique. @upcite.bsky.social @univ-spn.bsky.social @inserm.fr @inrae-france.bsky.social
02.06.2025 13:09 — 👍 2 🔁 1 💬 0 📌 0
"French trial sponsors are urged to bolster clinical trial transparency": STAT News covers the recent French report urging that study results be published in a registry. I was very happy to be interviewed and provide some context.
15.05.2025 14:20 — 👍 2 🔁 1 💬 1 📌 1
#SemainedelaRechercheClinique 🔬 | @isabelleboutron.bsky.social mène des travaux visant à améliorer la conception, la transparence et l'interprétation des essais cliniques. Découvrez sa vision⬇️
@ap-hp.bsky.social @upcite.bsky.social @cochranefrance.bsky.social @cressumr1153.bsky.social @inserm.fr
Pierre-Yves Ancel, co-founding member of @cressumr1153.bsky.social, looks back at the history of the center created in 2014. An ambitious project to foster interdisciplinarity in response to public health challenges.
youtu.be/a4tv4VY5uE8
👉 Subscribe for more videos!
@upcite.bsky.social
New post on ResToRes's website "Massive collaboration sharpens tools for identifying problematic trials". It is about the recent publication of INSPECT-SR Stage 2 findings in The Journal of Clinical Epidemiology.
restores.univ-rennes.fr/highlights/m...
'Early bird' rates extended!
To allow as many people as possible to experience the unique and enriching International Summer School in Public Health we've decided to extend our early bird rates until May 19, 2025, offering a learning opportunity open to everyone!
cress-umr1153.fr/en/summer-sc...
