In a report 2 years ago based on the UK General Practice Research Database, Currie et al.1 obtained information on 27 965 diabetic people receiving oral agents in combination and on 20 005 diabetic people who initially received oral agents and subsequently received insulin. Over a 5-year follow-up period, the lowest mortality was for those with a mean A1c of 7.5%. Either higher or, worrisomely, lower A1c levels were associated with lower survival. This was particularly true in the insulin-treated group, for whom mortality nearly doubled at either A1c 10.5% or A1c 6.0% compared with A1c 7.5%.1 Two recently published epidemiologic studies of diabetic patients with renal insufficiency show similar patterns. Shurraw et al.2 identified 23 296 diabetic patients with a glomerular filtration rate (GFR) lower than 60 mL/min per 1.73 m2, finding worse survival among those with A1c either >8% or <6.5%, although progression to end-stage renal disease (ESRD) only worsened at A1c >9%. In the second study performed on 54 757 diabetic patients receiving maintenance hemodialysis, Ricks et al.3 found that at A1c ≥10%, 3-year mortality was 19% greater than at A1c 7%–7.9%; however, mortality increased even more (by 35%) at A1c ≤5%. What is the biological implication of a “U-shaped curve”? Efforts to understand the relationship between obesity and illness have been stymied by such relationships, although the divergence between smokers and non-smokers in relationships between body mass index (BMI) and mortality (with smokers showing higher mortality at lower BMI, and non-smokers exhibiting higher mortality at high BMI)4 has led to the hypothesis that the effects of lower weight are an effect of ill health.5 Some studies excluding cigarette and alcohol users and those with diagnosed illnesses still suggest higher mortality at both lower and higher body weight levels.6 Similar reports of U-shaped curves exist for blood pressure,7 alcohol use,8 fasting insulin,9 and the duration of sleep.10 The question for A1c is then whether the U-shaped curves suggest that type 2 diabetic patients are optimal at a moderately elevated level of A1c, with a causal relationship between low (i.e. normal for non-diabetic people) A1c and adverse outcome, or whether this apparent effect is due to confounding factors that lead to both lower A1c and worse outcome. Conceptually, it is difficult to understand how normal glucose levels for non-diabetic people would (directly) lead to adverse outcome among those with diabetes. However, it is important to realize that the relationship between glycemia and outcome may be less consistent as more advanced disease develops. Thus, in the study of Shurraw et al.2 the relationship between A1c and likelihood of progression to ESRD was significant only for a GFR >40 mL/min per 1.73 m2 and was progressively stronger at higher baseline GFR levels. Similarly, in the Veteran’s Administration Diabetes Trial there appeared to be benefit of intensive glycemic control in those diabetic patients with baseline coronary calcium scores below, but not above, 100.11 It may be that certain advanced levels of diabetic complications do not benefit from improved glycemia, with intensive control only leading to a greater likelihood of harm from hypoglycemia. The notion that improved glycemia mediates adverse outcome then inherently conflates a biochemical measure of glycemic control with the measures taken to achieve lower glycemia. New approaches to improving glycemia may well allow benefit without adverse effects.12 Furthermore, although the relationship between A1c and glycemia is significant, in the study of Ricks et al.3 just over one-third of the variance in glucose was explained by the A1c level, and low glucose was much weaker in its association with mortality than was low A1c.3 It may be that A1c is not sufficiently precise to truly measure overall glycemic exposure.13 This may particularly be the case with renal insufficiency or with anemia, among the situations in which A1c can be particularly misleading,14 but it is important to realize that approximately 20% of diabetic people have A1c levels differing by >1% from that which would be predicted from their mean glucose.15 Do the “U-shaped curve” phenomena then imply that normalization of glycemia for type 2 diabetes is misguided? The jury is still out, but it seems that we should continue to optimize glycemic control for those diabetic patients with relatively early disease for whom the progression of complications can be halted while making sure that we avoid any adverse effects of treatment, particularly those associated with hypoglycemia.