Educational gain in cause-specific mortality: accounting for confounders