Even though we adjusted models for multiple confounding factors we were unable to adjust models for other important factors for which measures were not available, including glycaemic control, plasma insulin level, type of diabetes, duration of diabetes, and level of renal function