Some years ago our journal published a paper reporting concentrations of a substance in an organ in a small number of people of a particular occupational group who had died of a rare disease. The results have been reanalysed in two subsequent papers and discussed in five pieces of correspondence in two journals. The original paper contributes to a body of evidence used by the defence in some compensation claims in the USA. One of the authors of the original paper is prominent as an expert witness in such cases.
In the course of one of these compensation cases, some original laboratory results behind the original paper were disclosed to a court and were published in another journal. A lawyer complained to us that they undermined the original paper, which the complainant said should be retracted.
We have examined the original paper and the newly published data, and have concluded that the paper is consistent with the new data and the complaint cannot be upheld. However, in the course of this we noticed a completely independent problem: important statements made in the discussion in the paper do not agree with the results presented in the tables in the paper. Although this seems clear once it is pointed out, it has apparently not been noticed by authors of the seven subsequent publications on the data. Our conclusion is that we should not have accepted the original paper in its present form.
The heart of the inconsistency is that the text makes statements which it says are true of all the cases observed, but inspection of the results makes it clear that there is at least one exception, which weakens the impact when there are only a few cases anyway.
Close examination of the original paper also discloses that many of the measurements must have been at low levels, close to the limit of detection, and subject to large uncertainties which make the conclusions insecure statistically. The low level of the results is confirmed by the newly disclosed laboratory data. The paper does not discuss these uncertainties, and they have been ignored in references to the findings in later papers. Although this reinforces our view that we were wrong to accept the original paper as published, there is always room for argument about statistical analysis, so we regard this as a less serious problem than the inconsistency between the discussion and the tables.
The paper was processed before we started using online submission, and the reviewers’ and editor’s reports no longer exist.
We believe that this inconsistency would justify a notice of correction to the original paper, by the criteria in the COPE guidelines. However, the case does not fit the usual pattern because we are not responding to new information but to a realisation that we made a mistake and that we published a paper which was seriously flawed in parts—we would like to correct the paper because we have changed our mind about it.
Has the COPE forum any comments please?
The Forum suggested issuing a notice of correction but the editor should perhaps consult with the publisher’s legal department before publication. As the problem occurred nearly 10 years ago, another suggestion was to write an accompanying editorial explaining the whole case. The journal can issue a notice of correction without the approval or consent of the authors, but the advice was to contact the authors in the first instance and try to agree on the wording of a correction that is acceptable to all. The journal could draft the notice and send it to the authors for their comments. If agreement on the wording cannot be reached, the editor could suggest an arbitrator or panel of arbitrators. The editor could also allow the authors a chance to reply or comment further in the journal.
As advised by the Forum, the journal discussed the issue with their publisher's legal advisers and wrote to the authors proposing a notice of correction. They have just received a reply. The editor is trying to avoid involving arbitrators. Meanwhile, the editor has had a new submission from a third party reanalysing the original data.
Follow up (September 2012):
The author has proposed simply updating the table, and has given an explanation of all the inconsistencies except one. The faults in the paper have moved into the area of what the editors consider to be poor scientific judgement rather than deceit or factual error. The editors regret that these questions of judgement were not dealt with before publication, but in view of the age of the paper they have decided to accept the author's proposal just to correct the table, and to leave discussion of the paper's conclusions to other authors.