The population included 301,948 participants without a personal history of cancer (193,221 men, 44.8 ± 12.0 years and 108,727 women, 45.1 ± 14.2 years). Table 1 summarises the mean ± SD blood glucose levels for men and women for each quintile, along with minimum and maximum values. The mean ± SD blood glucose levels in men were: Q1, 4.81 ± 0.24 mmol/l; Q2, 5.29 ± 0.08 mmol/l; Q3, 5.49 ± 0.08 mmol/l; Q4, 5.84 ± 0.13 mmol/l; and Q5, 6.88 ± 1.61 mmol/l. Overall, blood glucose levels were lower in women than in men. The blood glucose level that typically defines normoglycaemia was observed in Q3 (5.49 ± 0.08 mmol/l [5.39–5.61] in men and 5.22 ± 0.08 mmol/l [5.11–5.33] in women), and the mean blood glucose level in Q5 was lower than the threshold defining diabetes (7 mmol/l) in both men and women.
The general characteristics of the population are shown in Table 2. Mean age increased from Q1 to Q5 (39 ± 12 years to 51 ± 11 years, respectively). Glucose-lowering agents (considered as a diagnosis of diabetes in this analysis) were used by 6.06% of participants, with the lowest proportions in the first three quintiles (3.38–5.44%) and reaching 8.90% in Q5. Declared aspirin treatment was similarly distributed at approximately 20%.
Table 3 presents the percentage of deaths attributed to cancer and non-cancer, according to sex, over the median 9 years of follow-up. Mortality rates increased from Q1 to Q5: all-cause, from 1.47% to 5.54% in men and from 0.69% to 3.99% in women; cancer-related, from 0.59% to 2.57% in men and from 0.30% to 1.69% in women; non-cancer-related, from 0.88% to 2.97% in men and from 0.39% to 2.30% in women (p < 0.0001 for all comparisons for linear trends in men and women). However, it is important to note at this stage that risk of death was significantly increased among participants in Q5.
Table 4 shows the adjusted mortality risk in each blood glucose quintile compared with Q1 (reference group) according to the different models. The relationship between blood glucose and cancer-related death does not appear to be linear, particularly in comparison with all-cause mortality. The HR for all-cause mortality went from 1.13 (Q2) to 0.93 (Q4) to 1.16 (Q5), while the HR for cancer-related death went from 1.10 (Q2) to 1.11 (Q3) to 1.04 (Q4) and then rose to 1.36 (Q5). An important result is that the blood glucose level that typical defines normoglycaemia is not associated with an excess risk of cancer-related death mortality.
In the Q5, an increased risk of all-cause and cancer-related death was observed after adjustment for age and sex. This excess risk remained significant after adjustment for age, sex and glucose-lowering treatment (model 2). After full adjustment (model 3), the excess risk of all-cause death was no longer present. In contrast, a significantly lower mortality rate was noted in Q4 compared with Q1. For cancer-related death, after full adjustment, the excess risk associated with Q5 reached 17% (HR [95% CI] 1.17 [1.03, 1.34]), corresponding to population attributable proportion of cancer of 3.22%. Additional analyses including lipid-lowering treatment did not modify this result (data not shown). For non-cancer-related death, analysis by quintile did not reveal any excess risk of death associated with blood glucose level.
Considering blood glucose as a continuous variable, the risk of death (HR [95% CI]) associated with an increase of 0.56 mmol/l of blood glucose was 1.04 (1.02, 1.06) for cancer, 1.04 (1.03, 1.06) for non-cancer and 1.04 (1.03, 1.05) for all-cause. The relationship with blood glucose remained significant after the introduction of a quadratic term (blood glucose squared) in the model for cancer-related deaths but disappeared for non-cancer-related deaths.
Table 5 shows the risk of cancer-related death according to normoglycaemia or hyperglycaemia (blood glucose ≥7 mmol/l) status and the use of glucose-lowering medication. Only the group presenting with high blood glucose levels without treatment was exposed to a significantly increased risk of cancer-related death.
After full adjustment, participants with a blood glucose level ≥7 mmol/l but who were not receiving glucose-lowering treatment, considered as incident diabetes, had an excess risk of cancer-related death (HR [95% CI] 1.26 [1.10, 1.45]). This risk was similar for non-cancer-related death (HR [95% CI] 1.66 [1.49, 1.86]). This excess risk for cancer-related death became non-significant when participants were treated for diabetes irrespective of their glucose control. However, participants receiving glucose-lowering medication, considered as having ‘known diabetes’, presented an excess of risk of non-cancer-related death (HR [95% CI] 1.45 [1.10, 1.92] for normoglycaemia and 1.60 [1.29, 1.99] for hyperglycaemia). These results suggest that incidence of diabetes is associated with cancer- and non-cancer-related death, and that glucose-lowering medication reduces the risk of cancer-related but not non-cancer-related death.
These results have significant implications for additional benefits of glucose-lowering medications on reducing cancer-related deaths. Although detailed information on prescription of glucose-lowering medications was not available, such a long-term effect of treatment is worth emphasising.
The relationship between blood glucose and cancer-related death differed according to the type of cancer (Table 6). For gastrointestinal cancer and leukaemia, an excess risk of death was observed in Q5, which remained significant after full adjustment (50% and 78%, respectively). For lung and prostate cancers, the significantly higher risk of death associated with Q5 disappeared after adjustment for confounders. For kidney and breast cancers, there were no associations between blood glucose level and mortality rates. For other cancers, the risk of death was significantly lower, regardless of the model used, suggesting a low incidence of cancer when blood glucose is within the normal range. For other cancers, the risk of death was significantly lower, regardless of the model used, when blood glucose is within the normal range, compared with higher blood glucose, and a low incidence of cancer was observed.
Tobacco-related cancers were grouped together (lung, mouth and throat, stomach, and bladder) and their relationship was analysed according to tobacco use. The relationship for cancer-related death and hyperglycaemia was not statistically significantly different between smokers and non-smokers with tobacco-related cancers (HR [95% CI] 1.19 [0.88, 1.60] and 1.45 [1.08, 1.95], respectively,). A non-linear trend between cancer-related death and blood glucose level was also observed for specific cancers, confirming a non-coincidental result for the study.