The infectious self-plagiarism of radiologist Hedvig Hricak

The journal Radiology has the highest impact factor among diagnostic imaging journals. Why would such a major Radiology journal publish a study that also appeared in Croatian Medical Journal, which is a minor local journal. I do not understand – they are not in the same league.

LikeLike

Dear radiology md – There are two possibilities as to why this was published in Radiology. Either the editors gave the authors a “pass” or the editors did not know.

I have to think that the editors did not know. I published in Radiology and there is a sole submission policy. When you look at both articles it looks like Croat Med J. 2015 article came from Belgrade and the Radiology, 2016 came from New York and Belgrade, but the figures are the same. I doubt the patients flew transatlantic for this study!!

I don’t know how the authors got away with the sole submission policy and how the same figures appear in both.

LikeLike

Hello all:
Read the link below…The editor Kressel warns people about what might happen to them if they plagiarize!!!

he even warns people that blogs like this may catch you!!!

http://pubs.rsna.org/doi/full/10.1148/radiol.12122201

Plagiarism and Redundant Manuscript Screening

This year we will introduce one additional new initiative in manuscript processing. Our journal and a number of others have been concerned about the issues of plagiarism and redundant manuscript submissions (2–7). A number of studies have suggested that as many as 10% of submissions to biomedical journals may be redundant (8). The availability of software to identify plagiarism and redundant publication has served to further heighten our awareness of the prevalence of this problem (9). Web sites such as “Deja vu: MEDLINE Duplicate Publication Database” (10) and “Retraction Watch” (11) serve to further underscore both the importance of the problem and the disappointing frequency with which it occurs. This year, as part of our pre–peer review process, we will be screening submitted manuscripts for plagiarism and/or redundancy by using a dedicated software package. Problematic manuscripts will be returned to the authors without peer review. As noted in our publication information for authors and on the author contributions disclosure forms that we require for every submission, Radiology accepts manuscripts of original research only with the understanding that they have been contributed solely to this journal; authors must attest that a manuscript on the same or similar material has not already been published by them and has not been and will not be submitted to another journal by them or by colleagues on their study before their work appears in Radiology. We hope that this initiative will serve to deter authors from this inadvisable practice and further enhance the scientific integrity of this journal.

LikeLike

Roos

If Radiology uses special software to detect plagiarism and/or redundancy, why didnt they catch the Blažić , Radiology, 2016 publication?

It has most of same patients from Blažić Croat Med J. 2015 paper.

LikeLike

I understand the humor in some of the comments, but this is a very serious subject. I believe the accurate language for most of these paper is “salami sliced publishing”. Salami sliced publishing has been written about and in my judgement is far more serious than self-plagiarism. Self-plagiarism may be have a copyright problem and self-loving, but misleading salami slicing in my opinion is scientific misconduct and dangerous to the scientific health community.
Salami sliced publications has been written about in the scientific press

There is a big cost as writen in Nature Materials

http://www.nature.com/nmat/journal/v4/n1/full/nmat1305.html

It is well dezcribed in this publication (Salami publication: definitions and examples Vesna Šupak Smolčić )

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3900084/

And it is even dezcribed by the international publisher Elsevier

Click to access Quick_guide_SS02_ENG_2015.pdf

Elsevier writes “as a general rule, as long as the “slices” of a broken up study share the same hypotheses, population, and methods, this is not acceptable practice. The same “slice” should never be published more than once”

There are certain examples when publishing separate papers is tolerated, but with open access and internet publishing the argument that there are separate audiences who read different journals is not usable anymore. In salami sliced papers the text is different so that is the reason they are not identified by plagiarism software.

There are major dangers to salami sliced publications. This summarizes it best from the USA Office of Research Integrity (http://ori.hhs.gov/plagiarism-16) :

“This type of misrepresentation can distort the conclusions of literature reviews if the various segments of a salami publication that include data from a single subject sample are included in a meta analysis under the assumption that the data are derived from independent samples. For this reason, the breaking up of a complex study containing multiple dependent measures into separate smaller publications can have serious negative consequences for the integrity of the scientific database. In certain key areas of biomedical research the consequences can result in policy recommendations that could have adverse public health effects.”

What is clear is how authors should handle possible salami slicing. Nature Materials tells it “Nature Materials explicitly requires that all authors provide details and pre-prints of all papers that are under consideration, in press or recently published elsewhere that could have any relevance to their submitted work. Ultimately, though, this relies on the integrity of authors”

There are times when a complete data set can be republished, for instance if there are new methods to analyze the data that was not available at the time of first paper. However, the decision if two manuscripts are separate is decided by the editors, not the authors. But the authors must be honest.

In these Hricak publications it seems the close date of all the groups of submissions are very problematic. The detailed comments in this blog are helpful – they show that sometimes no information or a very very tiny bit of information was provided about the other paper. In pubmed Hricak has 556 papers – its hard to think that she and her colleagues were not aware of the proper rules of publishing.

LikeLike

Dear Leonid:

I am the author of the Radiology paper on renal tumor growth rate. I would like to clarify a few things:

The “striking” “improbable” mathematic situation you mentioned, i.e., the doubling time of patients’ tumors “ranged from −248 to 72 days (mean, 474 days; median, 811 days)”, was taken out of context. The original results were written as such: “The median RDT of all renal tumors was 0.45, corresponding to a DT of 811 days, with a mean RDT of 0.77, corresponding to a DT of 474 days. The RDT range was −1.47 to 5.06, corresponding to a DT range of −248 to 72 days.”

Please note the repeated word “corresponding”. We ranked tumors by RDT values first, established the range and median of RDT second, and then finally found the DT values of tumors with such RDT range and median values. The RDT and DT values do not have a rank-preserving relationship, that is, when you rank the tumors based on one of these values, the other value will not have the same ranking. Therefore, the DT values quoted above were of the tumors with the smallest, greatest and median RDT values, not the actual smallest, greatest and median DT values. This can be confusing. Maybe the following simplified table will help:

Tumor 1 RDT -1.47 (lower end of range) DT -248 days decrease over time
Tumor 2 RDT 0.45 (median) DT 811 days median growth
Tumor 3 RDT 5.06 (higher end of range) DT 72 days fastest growth

On one end of the spectrum, tumor 1 shrank over time, with negative RDT and DT. Tumor 2 showed median growth, with RDT of 0.45, well within the range of -1.47 to 5.06. This tumor would on average double every 2+ years. On the other end, tumor 3 was the fastest growing one, with the greatest RDT value of 5.06, and its doubling time was only 2+ months, much shorter than that of tumor 2, and it should be.

As can be seen from the simplified table, the tumors were sorted by RDT, not DT. The ranking of DT was different from that of RDT. If you had ranked these three DT values from the smallest to the largest: -248 days, 72 days, and 811 days, yeah, you could say the range was -248 to 811 days, with a median of 72 days, but that would be meaningless. Because the median DT of 72 days did not correspond to the tumor with the median growth rate, but rather, the fastest growing one.

Simply ranking the tumors based on their DTs would work only if there were no regressing tumors with negative DTs (imagine there were only tumor 2 and tumor 3, but no tumor 1). Because in that case, the RDTs and DTs would actually be rank-preserving to each other, although in an inverse order, and our paper probably wouldn’t have raised so many questions. Unfortunately we were dealing with a more complicated scenario. We had to rank tumors based on their growth first (and RDT is a good indicator for this purpose), and then identified the DTs corresponding to tumors that shrank the most, grew the fastest (thus the range), and the one that was in the middle (median).

In my 2009 Response to Dr. Park, I explained the above, although not in such great details. I also quoted a paper by Jennings et al., who had used the same method to calculate lung cancer growth rate. In their results, their DT corresponding to median RDT was 207 days, also outside the DT “range” of -50 to 26 days. I felt that my Response had adequately answered this question, yet it seems that my Response had been either largely ignored or not well understood. It is also quite strange that in the presence of other studies with similar findings and methodology, our study was singled out and kept being criticized. Whoever is interested can look up the references given in our original paper regarding every step of the calculations.
A probable explanation for the “mathematical irregularities” you mentioned regarding Figure 2 is that, so many data points are clustered in the left lower corner, if your “median change line” just deviated from 26.5% to 27.5%, the dots would be divided very differently. I cannot address your other comments regarding this figure, since you did not specify the problem.

In terms of Figure 3, the apparent discrepancy is due to the fact that, the 21 tumors with DTs of more than 730 days (RDT < 0.5) did not include the regressing tumors with negative DTs. But when RDT values were calculated, the 21 tumors with very long DTs corresponded to small RDTs < 0.5, and the regressing tumors with negative DTs corresponded to negative RDTs. These two groups now would have their RDTs combined together in the first two bars of Figure 3. However, since there were 7 tumors with negative RDTs, the total number of tumors in bars -1 and 0 should have been 28, whereas Figure 3 appears to show two to three tumors more than that. Without access to the database, my best speculation is that this was caused by a rounding error. Many data points were clustered around RDT of 0.5 (Figure 4). A few tumors with RDT slightly greater than 0.5 probably got rounded down to 0.5 and included in bar 0. But these few tumors with RDTs slightly greater than 0.5 did not account for the 21 tumors of RDT < 0.5.

Regarding the 2013 Correction, the reason for the “disappearance” of Dr, Wang and myself was that we had both relocated away, and could not be reached. I can only imagine how painstaking it must have been for my former colleagues to follow the digital trail I had left behind, and to retrace every step I had taken for the study. The Correction was written mainly by statisticians, to whom I did not have a chance to explain the science behind my study, therefore what is confusing to you must have caused them an even greater confusion.

I agree with their statement that “the relationship between DT and RDT is not rank-preserving; that is, ranking patients based on DT values does not correspond to ranking them based on RDT values”, — this is exactly why the DT corresponding to median RDT was not the actual median DT and appeared to be out of range. This statement basically is a reiteration of what I had tried to explain in my previous Response to Dr Park. But unfortunately, just like my Response, this statement from the statisticians also went largely ignored or not being understood. I could see that in the Correction, the DTs were ranked to make the median DT appear within range. As I discussed before, you can rank whatever you want to rank, but ranking DT like this in this case is meaningless. Nor do I understand some of the other comments in the Correction such as, “DT values were incorrectly calculated based on RDT values”, because RDT values were calculated from DT, as stated in our Methodology, not the other way round. Not having participated in the re-analysis, I do not know exactly what calculation error was found.

I hope the above comments help to clarify a few things. Thank you and good luck.

LikeLike