Suspected paper-mill articles drew more citations but fewer readers in 20 cancer journals

The 4,085 figure is the output of a classifier. The team's previously developed detection model compares each paper's title and abstract with those of retracted paper-mill articles, and a paper that reads similarly gets flagged ^[3]. That puts the flagged share at about 12.3% of the 33,159 articles ^[13]. The study's own title calls the distortion "Suspected" ^[1]. The Phys.org report does not give the model's false-positive rate, the size of the citation gap or the source of the reader counts.

The design is what makes the result worth attention. Citations come from readers, so a paper's citation count and its readership should rise together, and the nonflagged papers are the comparison group ^[4]. Flagged papers broke that link. "Having more citations but fewer readers makes no sense, as citations come from readers," Scancar said ^[9].

A second pattern fits the first. Flagged papers mostly cite and are cited by other flagged papers ^[5]. We'd expect papers that readers valued to draw citations from outside the flagged set as well.

The thing this doesn't tell you is who placed the citations. Adrian Barnett, the corresponding author and a statistician at QUT, said the team had documented a pattern that suggested widespread, coordinated manipulation of citation counts by paper mills ^{[12] [7]}.

The impact-factor point is Barnett's, and he states it as a possibility. "Artificially high citation rates can inflate journal impact factors, amplify unreliable findings and slow scientific progress," he said ^[8]. The measured result is narrower. Flagged papers clustered in a subset of journals, including Journal of Experimental & Clinical Cancer Research and Molecular Cancer, while other journals were only marginally affected ^[6].

We think the supportable reading is journal by journal. In the clustered journals, a citation count says less about trust than it appears to. Across all 20 journals, the study does not support that claim.

Barnett's concern is downstream. "A danger is that artificially inflated citation counts may distort the evidence base by misinforming future research and clinical translation," he said ^[11]. The authors' proposed remedy is screening: "Our results support the screening of articles and journals for the patterns of citation manipulation we have described," Scancar said ^[10].