Missing values were imputed using 10-fold multiple imputation for each cohort, using the remaining variables as predictors.
Analysis of health status outcomes for surviving participants used multiple imputation for missing data: 50 imputed datasets were created and estimates combined across datasets with Rubin's rules.
To adjust for sample attrition, all of our analyses were conducted using ten imputed datasets created via chained equation multiple imputation for missing values, a method recommended over listwise deletion of cases with missing observations (e.g., Widaman 2006).
Furthermore, although we used multiple imputation for missing values, separate models were created using original and imputed data, and no significant differences were seen.
Approximate ICERs for the full imputed sample were -£15,500/QALY (last observation carried forward) and -£6,000/QALY (multiple imputation) for total healthcare costs and -£3,900/QALY (last observation carried forward) and -£3,100/QALY (multiple imputation) for combined social and healthcare costs.
The missing values were replaced using linear regression for multiple imputation for continuous variables and using logistic regression for binary variables (command mi impute).
Rubin DB (1987), Multiple Imputation for Nonresponse in Surveys, John Wiley.
Additional file 1 Multiple Imputation for Missing Data in RAD Data.
Multivariate analysis used a lasso regression (Tibshirani 1996; Musoro 2014), including multiple imputation for missing values and bootstrap validations, adapted for high numbers of variables (biomarkers).
N = 1,288 with multiple imputation for missing values.
Group allocation was maintained using multiple imputation for missing values.
