We observed a continuous increase in the risk for all-cause mortality throughout the whole range of glycated haemoglobin values in a non-diabetic UK population, with suggestion of a ‘basement’ to the graded relation for the low range. There was a similar pattern of association for cardiovascular and cancer-related mortality, with the strongest association being observed for cardiovascular mortality.
The shape of the association between a risk factor such as glycaemia and health outcome has major implications for the recommendation of targets for interventions. There is broad evidence for hyperglycaemia as a risk factor for cardiovascular disease [15, 16], extended very recently by results of a large collaborative meta-analysis of 102 studies on fasting glucose concentrations , but data on the low range of glycaemia are inconsistent. In addition to the ARIC study and NHANES, four earlier studies reported mortality risk for categories of glycated haemoglobin in non-diabetic populations, with one study suggesting a potentially increased risk  and three studies suggesting no increased risk with low levels of glycated haemoglobin [6, 7, 9]. However, low numbers of events, particularly in the low range of glycated haemoglobin (below ~5.0%), limit an accurate risk estimation in these studies. In contrast to the ARIC and NHANES results, our findings suggest that there is no lower threshold of glycated haemoglobin associated with increased risk for mortality.
We cannot distinguish in our analysis between a shallow linear increase in risk across the entirety of the distribution of glycated haemoglobin and a very flat relationship below a threshold above which risk increases linearly. In general it is difficult to distinguish between threshold relationships and linear ones and to identify the exact location of the threshold level when event rates are very low. Although there was no statistically significant departure from linearity, our data suggest that there may be a ‘basement’ level to the graded relation between glycaemia and health risk with potentially no relevant mortality benefit of further lowering glycaemic haemoglobin. Such a notion has been suggested recently for fasting glucose and cardiovascular disease . The basement range found in our study corresponds to the ‘normal’ range of glycated haemoglobin found in people in the general population who have normal glucose tolerance (mean glycated haemoglobin 5.3 ± 0.3%) . This is compatible with recent suggestions  that normal levels of glucose ought not to be associated with increased health risk given the essential requirement for glucose to sustain metabolic activity and life and consequently the tight regulation of blood glucose levels.
The shape of the risk relationship analysis of glycated haemoglobin with cause-specific mortality was similar to that for all-cause mortality, showing that the relationship with all-cause mortality was strongly mediated by cardiovascular mortality. The number of cardiovascular deaths in the lowest category of glycated haemoglobin was low, but previous findings on combined fatal and non-fatal cardiovascular events in the ARIC study and our population, resulting in larger numbers of events, also confirmed an increased risk only for the high range of glycated haemoglobin. Furthermore, we observed a similar shape of association for cancer-related mortality, with an increased risk for increasing glycated haemoglobin levels. Previous studies examining the association of glycated haemoglobin with incident cancer or cancer-related mortality were limited by small numbers of cases, which did not allow reliable risk estimation for the broad range of glycated haemoglobin levels [7, 20–23]. The largest analysis of a case–control set for colorectal cancer within the total EPIC cohort described a significantly increased risk of hyperglycaemia in agreement with our results on mortality from any cancer type . The underlying mechanism remains unclear and exploration is beyond the scope of this study. However, the important finding here is that there was no evidence for an increased risk of low levels of glycated haemoglobin for cancer-related mortality which, next to cardiovascular disease, is the leading cause of mortality in the elderly population.
The mechanistic pathways underlying the relationship between low glycaemia and health risk might differ overall between patients with treated diabetes and non-diabetic individuals. However, recent post hoc analyses of the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial also suggested that a subgroup of patients with refractory hyperglycaemia accounts for the increased mortality risk observed with the intensive treatment strategy in diabetic patients and not ordinary lower glycaemia [24–26].
The explanations for the different findings of our study compared with the ARIC study and NHANES remain to be determined. In contrast to our homogeneous study population of white Europeans, the design of the ARIC study and NHANES is multi-ethnic, with about 11–22% blacks included. Blacks have higher levels of glycated haemoglobin than whites, even after correcting for glucose levels . It is unclear whether these racial disparities have relevance for the susceptibility to glycaemia-associated complications. Analyses of NHANES suggested that the increased mortality risk associated with low levels of glycated haemoglobin is partly mediated by liver disease, specifically hepatitis C . We observed much lower levels of markers of liver damage compared with NHANES overall and particularly no increase in these markers in individuals with low glycated haemoglobin. Furthermore, the prevalence of seropositive hepatitis C assessed in a subset of the UK EPIC cohorts was 0.6% , which is also much lower than in NHANES participants (2.3%); this might therefore partly explain the divergent findings for the association between glycated haemoglobin and mortality risk.
The strengths of our study are the long and complete follow-up as well as the large sample size. We had nearly four times more individuals with glycated haemoglobin below 5% compared with the ARIC study and 10% more individuals compared with NHANES, which allowed us to estimate the association between glycated haemoglobin and mortality precisely for the range of maximum interest. A limitation of our study is the virtually exclusive inclusion of white Europeans. Thus, our results cannot be generalised to other ethnic groups and our study also cannot contribute to the exploration of ethnic differences in the relation of glycated haemoglobin and health outcome. Furthermore, our analyses are based on a single measurement of glycated haemoglobin at baseline. This could lead to underestimation of the true association due to error resulting from random variability; however, the within-participant variation of glycated haemoglobin in non-diabetic individuals is minimal (coefficient of variation 1.7%) . Finally, we did not have sufficient events to exclude an increased risk of mortality from specific cancer subtypes in individuals with low glycated haemoglobin. However, it is unlikely that an association with very rare diseases would underlie the findings of the ARIC study and NHANES or be relevant for the clinical use of glycated haemoglobin.
In conclusion, our findings in a large non-diabetic population of white Europeans do not support the concern about increased mortality risk associated with low levels of glycated haemoglobin. Differences in population characteristics including the prevalence of co-morbid conditions might explain the discrepant results across study populations. Further research is needed to explore determinants of low glycated haemoglobin which are concurrently associated with increased mortality.