P-values in baseline tables of randomised controlled trials are inappropriate but still common in high impact journals

Abstract

Testing for baseline differences between the treatment and control group in randomised controlled trials (RCTs) (see for example a recent article in this journal) is not appropriate, as pointed out by numerous papers and the CONSORT statement. Any baseline differences between the groups under study are by definition due to chance (as long as the randomisation was performed correctly). In addition, whether baseline differences are significant does not have any implications for the validity of the results of the study. For example, even a covariate that is balanced between treatment groups (according to a p-value) can affect the association between treatment and outcome. Previous studies reviewed a number of RCTs that were published in 1987 and 1997, and showed that testing for baseline differences was common (58% and 48%). We performed a review of all RCTs published in 2008, 2009 or 2010 in major general medical and cardiovascular journals – Annals of Internal Medicine, British Medical Journal (BMJ), Circulation, European Heart Journal, Journal of the American Medical Association (JAMA), Lancet, New England Journal of Medicine (NEJM) – to assess how often testing for baseline differences is currently reported. In 388 out of 1115 papers (34.8%) that reported baseline characteristics of the study population, a p-value in the baseline table itself or in the footnote accompanying the table was reported (Table 1). There were considerable differences between journals: more than half of the RCTs published in the JAMA or NEJM reported a p-value in their baseline table, whereas this was the case in less than 10% of the RCTs that were published in the BMJ or Lancet. In the two major cardiovascular journals it was also very common for RCTs to report p-values in the baseline table. Even though the proportion of papers on RCTs that report p-values in the baseline table seemed to have decreased over the last two decades, when comparing previous reviews with our data, it is still very