Variables associated with government scheme membership in univariate analysis (p<0.1) were included in the initial multivariate model in addition to important potential confounders such as gender, and retained if their removal markedly altered the model fit.