A rheumatology journal has retracted a paper about treating knee pain after an institutional investigation found a mistake in the statistical process.
Over several months, the authors proposed a series of corrections to the 2014 study. However, the journal Annals of the Rheumatic Diseases (ARD) decided that there were “unresolved concerns” about the reliability of the data, and decided to retract the paper entirely, despite the authors’ objections.
Here’s the retraction notice:
Abou-Raya A, Abou-Raya S, Khadrawe T. Methotrexate in the treatment of symptomatic knee osteoarthritis: randomised placebo-controlled trial. Ann Rheum Dis 2014. Published Online First 27 Mar 2014. This paper has been retracted. Following publication ARD raised concerns with the authors and asked the Faculty of Medicine of the University of Alexandria to investigate. The report of the Investigation Committee included the following findings:
“Great concerns rose about the following points:
- Aggregation of age data at the age of 60 years constituting 26.4% of the placebo group and 22.2% of the Methotrexate group despite that data is ranging between 60 to 78 years and 60 to 84 years in both groups respectively.
- Correlation between age of the disease is evident among only the Methotrexate group (r=0.323, P=0.006) while no correlation between both variables in the placebo group (r=0.131, P=0.275).
- Coding system of the comorbidity variable is not clear – if it is either scale variable or categorical variable – which raises concerns about validity of the method of analysis used.
- Analysis of comorbidity (using data sent by the authors) revealed different results than that published (page 7 in output attachment).
- Baseline VAS was nearly perfectly correlated in the placebo group with its 2nd reading at 28 weeks (r=0.927, P<0.0001), while that correlation doesn’t not even exist among the Methotrexate group (r=0.023, P=0.859).
- “VAS mean difference” used in the paper was calculated using the mean percent difference which was not clear or mentioned in the research published.
- Errors were made in the analysis of “VAS mean difference” by including the missing values (given the code 99) in the analysis resulting in an evident errors (Page 12 output attachment).
- Analysis of the scores used yielding different results than that published.”
The Investigation Committee’s report concluded: “There is an unintentional mistake in the statistical process, with errors in collection of data in some groups”.
In the course of discussions with ARD about the paper, over a number of months, the authors provided three different data sets and proposed corrections to the text and two of the tables in the paper. ARD considered that the corrections to the paper were significant and was not able to agree to them for that reason and because of unresolved concerns about the reliability of the data in the paper and in the corrections. The authors do not agree with ARD’s concerns.
The 2014 paper — which found that the anti-cancer drug methotrexate reduced pain and improved function in people with knee arthritis — has not yet been indexed by Thomson Reuters Web of Science.
Tore Kristian Kvien, editor-in-chief of ARD from the University of Oslo in Norway, told us the journal has nothing more to add to the lengthy retraction notice.
We reached out to the first author of the paper, Anna Abou-Raya, and the University of Alexandria in Egypt for further clarifications. We’ll update the post with anything else we learn.
Like Retraction Watch? Consider making a tax-deductible contribution to support our growth. You can also follow us on Twitter, like us on Facebook, add us to your RSS reader, sign up on our homepage for an email every time there’s a new post, or subscribe to our new daily digest. Click here to review our Comments Policy. For a sneak peek at what we’re working on, click here.
Methotrexate for arthritis? I guess if you’re in enough pain, leukopenia and risk of liver damage might be worth it.
P.S. Just checked Wikipedia, and apparently it’s a lower dose than what you get for cancer treatment. That makes more sense.
There is so much of their retraction reasons that is wrong or irrelevant.
What is point 1 talking about?
Point 2 is meaningless. I assume they mean correlation between age and duration. Comparing p values is wrong and if they want to see if there is a difference they should fit a regression model to compare the slopes in both groups. It will probably find no difference.
Point 5 is not unexpected although the correlation does look high for the placebo group. Placebo patients would be expected to be more highly correlated within time points as nothing is done to them. Instead for the Methotrexate group there may well be different effects of the drug among different patients, with it possible that there may be two groups of patients, ones that responded and ones that didn’t which would remove baseline correlations.
Where the problems with point 5 fixed by fixing point 7 or did they still exist?
The problem with the 99 is that they should probably never have been 99 as that is within the data range. Maybe there is a typo in the retraction and it should be 999. Either way the first step in analysis should be to recode the missing and then print descriptive statistics including minimum and maximum for any variables to be used.