A comparison of semiparametric approaches to evaluate composite endpoints in heart failure trials