Composite outcomes in randomized trials: greater precision but with greater uncertainty?