Exactly when should an article be retracted? That is not so easy a question to answer, fraud and significant data or image manipulation are one fairly obvious group of reasons. Making errors and/or poor research design are more tricky. When the errors are such that the major results are unreliable, then retraction should be considered.
Retractions have been very few in neonatology, but one recent example, is an article that I blogged about previously, a systematic review of NIDCAP from a French group. The retraction notice states that the authors had included some data twice, which I hadn’t realized. Some of (or at least 2 of ) the articles that they included in fact were reporting data from the same infants.
The retraction notice also notes that there were errors in the description of the outcome variable. As far as I can see this is because some of the scores that were included were actually from the 1st version of the Bayley scales, but in the systematic review it was stated that they were from the 2nd edition. I already noted that it really isn’t appropriate to mix Bayley scores at different ages, especially very early scores at 9 months of age, mixing version 1 and version 2 scores makes that error worse. In fact the second edition of the Bayley became available in 1993, so all the babies in the first NIDCAP trial were examined with the first edition. These babies were also re-reported in the paper in 2009 by McAnulty, at which time some of them were 25 years old. So I think it is the duplicated publication of data in McAnulty 2009 which is the culprit.
I think the right decision has been made here, once those data points are subtracted, and once the Bayley 1 are evaluated separately to Bayley 2, then it seems unlikely that there will be any difference between NIDCAP and control. So the results were unreliable and the literature is better with a retraction.