Are multiple primary outcomes analysed appropriately in randomised controlled trials? A review

ObjectivesTo review how multiple primary outcomes are currently considered in the analysis of randomised controlled trials. We briefly describe the methods available to safeguard the inferences and to raise awareness of the potential problems caused by multiple outcomes.Methods/designWe reviewed randomised controlled trials (RCTs) in neurology and psychiatry disease areas, as these frequently analyse multiple outcomes. We reviewed all published RCTs from July 2011 to June 2014 inclusive in the following high impact journals: The New England Journal of Medicine, The Lancet, The American Journal of Psychiatry, JAMA Psychiatry, The Lancet Neurology and Neurology. We examined the information presented in the abstract and the methods used for sample size calculation and statistical analysis. We recorded the number of primary outcomes, the methods used to account for multiple primary outcomes, the number of outcomes discussed in the abstract and the number of outcomes used in the sample size calculation.ResultsOf the 209 RCTs that we identified, 60 (29%) analysed multiple primary outcomes. Of these, 45 (75%) did not adjust for multiplicity in their analyses. Had multiplicity been addressed, some of the trial conclusions would have changed. Of the 15 (25%) trials which accounted for multiplicity, Bonferroni's correction was the most commonly used method.ConclusionsOur review shows that trials with multiple primary outcomes are common. However, appropriate steps are not usually taken in most of the analyses to safeguard the inferences against multiplicity. Authors should state their chosen primary outcomes clearly and justify their methods of analysis.