Background/Purpose: The quality of statistical analysis reporting is wanting even in our most prestigious journals. It stands to reason that active participation of biostatisticians/epidemiologists (b/e) in reporting would improve the situation. We tested the hypothesis that more close cooperation with a b/e would improve the quality of reporting randomized clinical trials (RCTs) rheumatology. We defined this close cooperation as the inclusion of a formal b/e among the co-authors and/or a declaration of formal statistical help in the study reports.

Methods: Two independent observers screened both by reading and electronic scanning, when applicable, the texts of all RCTs in Annals of the Rheumatic Diseases, Arthritis Care and Research, Arthritis and Rheumatology, Rheumatology Oxford published in 2015 and 2016. Using a pre-prepared worksheet, the observers specifically tabulated the presence of a b/e among the co-authors and/or formal acknowledgement of statistical help in the methods, the inclusion of effect sizes (the kind, whether they were specifically voiced as effect sizes, whether they could be calculated by the reader from the data presented or not given at all) and the presence of associated confidence intervals for the given effect sizes. Also tabulated were the improper aspects of p value reporting, including giving relative (>p>) instead of exact (p=) p values and the erroneous inclusion of p values in tables depicting trial entry data in randomized trials, since these tables, by definition, display randomized features (1). An arbiter (HY) decided the final tabulation when there were discrepancies between the 2 observers and there were up to 15 discrepancies in total where more than 1 discrepancy was possible per article.  

Results: The total number of RCTs was 134. There were 29 (22%) articles (Group I) in which a formal statistical help was acknowledged. In 26/29 this was a co-authorship and in 3, only a mention in the text. The remaining 105 (78%) of the articles made up Group II. The Table gives the findings. It is seen that the reporting of effect sizes and the desired way of reporting exact p values were significantly more common in Group I. On the other hand, reporting of confidence intervals and including p values for baseline data after randomization did not show significant differences between the groups.

Conclusion: The inclusion of a b/e, interestingly formally present in only 29/134 (22%) of the analyzed manuscripts, improved the reporting of effect sizes -at least rendering them calculable- and in reporting exact p values. The same cannot be said for reporting confidence intervals for effect sizes and giving p values for baseline data. A limitation of our work was the relatively small number of manuscripts in Group I. 

Table: The differences of the parameters between two groups

Reference:

1.        http://www.consort-statement.org/checklists/view/32–consort-2010/510-baseline-data

accessed June 4 2018