Genome Biology has partially retracted a high-profile paper about an epigenetic biomarker of aging – a year and a half after the author alerted the journal to a software coding error that invalidated one of its findings.
The paper, “DNA methylation age of human tissues and cell types,” garnered some media coverage and forms the basis of its author Steve Horvath‘s work on measuring human aging. It has been cited 73 times, according to Thomson Scientific’s Web of Knowledge. The article is also recommended on the post-publication peer review site Faculty of 1000.
The lengthy, peer-reviewed erratum notice, written by Horvath, refers to several figures and files, as well as a conclusion:
All of my results from [1] that involve non-cancerous tissue or cancer cell lines remain valid but I have to report some corrections for the cancer tissue data. In particular, I have to retract the statement that cancer is associated with an increased DNA methylation age (i.e. positive age acceleration) in most cancer types. In fact, while some cancer types show positive age acceleration, others exhibit negative age acceleration.
The editor of Genome Biology, Louisa Flintoft, explained on the publisher’s On Biology blog why the article was corrected, rather than retracted:
So should we have retracted the paper? The effects of the error that Horvath reports are limited to the sections of his paper that look at cancer tissues. The preceding sections that deal with non-cancer tissues are unaffected, and indeed this is the bulk of the paper.
Although there have been some requests to retract the paper, we decided not to do this. We felt it of greater benefit to the research community to let the unaffected results and conclusions remain as part of the scientific literature so that this work can be built on. We have now published a detailed Erratum, which has been peer reviewed, in which Horvath explains the error and how it affected the results he originally reported.
The blog post explains that after the original paper was published on October 21, 2013, Horvath, a geneticist the University of California, Los Angeles, immediately informed the journal of his mistake. The post links to a brief online comment to the article that Horvath published two weeks later, identifying the problem and pointing readers to a more detailed explanation on his UCLA web site.
The post goes on to note that the guidelines of the Committee on Publication Ethics state that
…a formal correction notice must be published, which is permanently associated with the paper, both on the journal’s website and in databases such as PubMed.
The sound thinking behind this is that it’s important to see the full history of a published paper – readers must be able to view the flawed version that was originally published, along with a clear description of where errors lie. This is essential so that those who have already read the paper in its original form know exactly which of the original results and conclusions have changed. That would not be possible if the original version was simply replaced by a new, correct one.
We find it odd, therefore, that as of five weeks after the final erratum notice’s publication, neither the original article nor the PubMed entry link to it.
In terms of why there was such a long delay between when Horvath noted the mistake (November, 2013) and when the journal formally acknowledged it (May, 2015), Flintoft offered an apology:
Publishing a complicated Erratum such as this one is never a fast process as it involves back-and-forth between authors, editors and reviewers. However, the time taken to publish this Erratum has been unduly long and we sincerely apologize for this delay. Procedures are now in place so that the process will be much faster in future for any papers published in Genome Biology that require correction.
In response to our request for comment, the journal replied with a further explanation:
The Erratum was initially handled by an editor who subsequently left the journal, and we have apologised to the author for the resulting delay in completing this. We have since improved our internal procedures to ensure that such situations are better tracked and addressed more efficiently.
Horvath also expanded a bit on the details provided in the blog post and the retraction notice:
The article appeared in print Oct 21 2013. I caught the error after one week. I posted the correction on the same day that I caught the error. Specifically, I notified the journal Oct 27 2013 about the error and also included an updated version of the proofs. On the same day, I posted a correction on my webpage: http://labs.genetics.ucla.edu/horvath/dnamage/
Further, Genome Biology posted my corrections under “Reader comments” Nov 4, 2013.
Message: I tried to limit the potential damage by acting very quickly.
I also send the journal a corrected version of the proofs of my article on Oct 27 2013 hoping that the production team could still post the corrected version.
Unfortunately, the journal was not able to implement these changes right away since the horse was out of the barn. I am not able to comment why it took so long to post the correction.
I think the editor who handled this article moved on to another journal.
The good news is that the vast majority of my results were not affected. Only a couple of results pertaining to cancer tissue were affected. If anything the results got better (statistically more significant). The one noteworthy error was the statement that all cancer tissues exhibit positive age acceleration. I have to retract this particular statement.
Horvath added that the paper’s most important finding – “the development of a biomarker of aging that applies to most human tissues” – has since been validated in several subsequent articles.
Update 6/24/15 9:51 a.m.: The journal tweeted a response to the article, below:
@RetractionWatch Thx for pointing out lack of link in original art. Fixed last week, we can also confirm that pubmed linking sorted out too.
— Genome Biology (@GenomeBiology) June 24, 2015
Like Retraction Watch? Consider supporting our growth. You can also follow us on Twitter, like us on Facebook, add us to your RSS reader, and sign up on our homepage for an email every time there’s a new post.
While “mega-corrections” are unpopular here at RW, I do believe that a correction is the correct way to handle this case.
Correcting the literature by deleting large chunks of good data is not the way to go IMHO. You could correct the literature by republishing the good data, but that would mean you have a paper published two years after the original paper and way after papers of other groups who have reproduced the data. This really confuses everything and everybody.
In the digital age, I would actually like to see a completely corrected .pdf of the paper alongside the old .pdf that carries a big “corrected version available” across every page.
Interesting topic. Does anyone here know of similar cases of ‘mega-corrections’ ?