Reference-based multiple imputation for missing data sensitivity analyses in trial-based cost-effectiveness analysis.