Hypoglycemia and Risk of Death in Critically Ill Patients

BACKGROUND Whether hypoglycemia leads to death in critically ill patients is unclear. METHODS We examined the associations between moderate and severe hypoglycemia (blood glucose, 41 to 70 mg per deciliter [2.3 to 3.9 mmol per liter] and <= 40 mg per deciliter [2.2 mmol per liter], respectively) and death among 6026 critically ill patients in intensive care units (ICUs). Patients were randomly assigned to intensive or conventional glucose control. We used Cox regression analysis with adjustment for treatment assignment and for baseline and postrandomization covariates. RESULTS Follow-up data were available for 6026 patients: 2714 (45.0%) had moderate hypoglycemia, 2237 of whom (82.4%) were in the intensive-control group (i.e., 74.2% of the 3013 patients in the group), and 223 patients (3.7%) had severe hypoglycemia, 208 of whom (93.3%) were in the intensive-control group (i.e., 6.9% of the patients in this group). Of the 3089 patients who did not have hypoglycemia, 726 (23.5%) died, as compared with 774 of the 2714 with moderate hypoglycemia (28.5%) and 79 of the 223 with severe hypoglycemia (35.4%). The adjusted hazard ratios for death among patients with moderate or severe hypoglycemia, as compared with those without hypoglycemia, were 1.41 (95% confidence interval [CI], 1.21 to 1.62; P<0.001) and 2.10 (95% CI, 1.59 to 2.77; P<0.001), respectively. The association with death was increased among patients who had moderate hypoglycemia on more than 1 day (>1 day vs. 1 day, P = 0.01), those who died from distributive (vasodilated) shock (P<0.001), and those who had severe hypoglycemia in the absence of insulin treatment (hazard ratio, 3.84; 95% CI, 2.37 to 6.23; P<0.001). CONCLUSIONS In critically ill patients, intensive glucose control leads to moderate and severe hypoglycemia, both of which are associated with an increased risk of death. The association exhibits a dose-response relationship and is strongest for death from distributive shock. However, these data cannot prove a causal relationship.