Outcome reporting bias in clinical trials

Abstract

Randomised controlled trials (RCTs) are designed and powered to measure one single outcome, calledprimary outcome(Sibbald & Roland, 1998; Barbuiet al., 2007). The primary outcome is the pre-specified outcome of greatest clinical importance and is usually the one used in the sample size calculation (Accordini, 2007). In addition to the primary outcome, RCTs may have several other outcomes, calledsecondary outcomes. In contrast with the analysis of the primary outcome, the analysis of secondary outcomes and its interpretation may be complicated by at least two factors:1)the trial may not have enough statistical power to detect differences (so it is possible to incur in a type II error, that is failing to see a difference that is present);2)increasing the number of secondary outcomes generates the problem of multiplicity of analyses, that is the proliferation of possible comparisons in a trial (and increasing the number of comparisons increases the possibility to incur in a type I error, that is detecting significant differences by chance). For all these reasons, the results of the analysis of primary outcomes is considered less susceptible to bias than the analysis of secondary outcomes.