Confounding and missing data in cost-effectiveness analysis: comparing different methods