|
Abstract:
|
Clinical trials frequently consider a vector of binary events as primary outcome. Primary outcome may be a single binary indicator, or composite, of whether any vector event occurs in a patient. A composite outcome is deemed desirable because it 1) increases incidence of primary outcome, and therefore thought to increase power, 2) removes need to focus exclusively on a particular outcome, say, particular complication, when experimental intervention expected to reduce more than one complication type. Other options are to analyze the count of events per individual, analyze each outcome individually, and multivariate setting where the treatment-by-outcome interaction can be assessed. Trial designers should consider strengths and weaknesses of various approaches, including interpretation, error rates, correlation among outcomes, and the inherent unequal weighting across outcomes.
|
