They adjusted the hazard ratios to remove the effect of diabetes. Would this not skew the data when they are not included in the sugar cohort? The confounding variables I'm talking about are behavioral, not clinical.