http://www.bmj.com/content/347/bmj.f6782?etoc= David Prieto-Merino, lecturer1, Liam Smeeth, professor2, Tjeerd P van Staa, professor23, Ian Roberts, professor1 Author Affiliations Correspondence to: D Prieto-Merino firstname.lastname@example.org Accepted 25 October 2013 Composite outcomes seem an attractive method to increase statistical power, but they can mask the effect of treatment According to international guidelines,1 2 outcome measures in a clinical trial should address the risks and benefits of a treatment, be relevant to patients, and be sufficiently common to make the trial feasible. In an attempt to meet these objectives many investigators select outcomes such as all cause mortality, all hospital admissions, or any adverse event. These outcomes can be thought of as composite outcomes as they combine multiple outcomes that are cause specific. All cause mortality is a popular outcome measure because it is believed to provide the net effect of the treatment, it seems more patient relevant than cause specific mortality, and it provides more outcomes so should increase statistical power. Another common approach to increase power is to use wide case definitions and sensitive tests. In recent years several papers have reviewed and debated the use of composite outcomes in clinical trials.3 4 5 6 7 8 9 Authors agree on their advantages and disadvantages, and a good summary can be found in a recent report from the European Network for Health Technology Assessment.10 Briefly, the main objection to the use of composite outcomes is that, if the treatment has different effects on the different components of the outcome, the net effect on the composite outcome is difficult to interpret. It also complicates patient management decisions, raising the question of which particular component outcome is more relevant for each patient. However, in the ongoing debate little emphasis has been given to the fact that, by including events that are not causally related to the treatment (either by using a composite outcome or by misclassifying events with a wide case definition), the overall effect of the trial will be diluted towards the null. In this paper we explain why dilution occurs, provide examples of trials where this has happened, and discuss how dilution can offset many of the supposed advantages. Read rest of the article here.