Abstract and Introduction
Abstract
Objectives: To assess the effect of intensive insulin therapy on blood glucose amplitude variation and pattern irregularity in critically ill patients. To assess the association of these blood glucose signal characteristics with hospital mortality, independent of blood glucose level.
Design: Retrospective analysis of the databases of two previously published randomized controlled trials.
Setting: University hospital, 56-bed adult surgical intensive care unit and 17-bed medical intensive care unit.
Patients: One thousand five-hundred forty-eight surgical intensive care unit patients, admitted between February 2000 and January 2001, and 1200 medical intensive care unit patients, admitted between March 2002 and May 2005.
Interventions: In the two randomized controlled trials, patients were randomized to receive either intensive insulin therapy (targeting normoglycemia, between 4.4 and 6.1mmol/L) or conventional insulin therapy (infusing insulin when blood glucose levels were >12 mmol/L and stopping at 10 mmol/L).
Measurements and Main Results: Intensive insulin therapy significantly lowered mean blood glucose (5.8 vs. 8.4 mmol/L), hyperglycemic index (0.8 vs. 3.2 mmol/L), and glycemic penalty index (26 vs. 53), but it increased the mean daily difference between minimum and maximum blood glucose (mean daily δ blood glucose; 4.0 vs. 3.3 mmol/L). There was no significant effect on the standard deviation of the blood glucose measurements or on jack-knifed approximate entropy. In multivariable logistic regression analysis, corrected for baseline risk factors, blood glucose levels outside the normoglycemic range, higher mean daily δ blood glucose, higher standard deviation blood glucose, and higher jack-knifed approximate entropy were independently associated with hospital mortality.
Conclusions: The Leuven intensive insulin therapy strategy increased mean daily δ blood glucose while not affecting standard deviation blood glucose and jack-knifed approximate entropy. Increased blood glucose amplitude variation and pattern irregularity were associated with mortality, irrespective of blood glucose level. The reduced mortality observed with intensive insulin therapy in the Leuven trials cannot be attributed to an effect on blood glucose amplitude variation or entropy. Reducing amplitude variation and entropy of the blood glucose signal, irrespective of blood glucose concentration, may produce clinical benefits.
Introduction
Critical illness-induced hyperglycemia is associated with adverse outcome. Observational studies have revealed a J-curved relationship between blood glucose level and mortality, with a nadir approximately between 5 and 8 mmol/L (90–140 mg/dL). In patients with an acute coronary syndrome a similar association has been observed, with the lowest risk of mortality at blood glucose levels between 4.4 and 5.5 mmol/L (80–100 mg/dL). Intervention studies on the impact of prevention and treatment of hyperglycemia during critical illness have revealed conflicting results. Single-center studies performed in Leuven, Belgium, two in adults and one in pediatric patients, showed that targeting age-adjusted normoglycemia using insulin infusion (intensive insulin therapy [IIT]) reduced mortality by an absolute 3%. Multicenter studies using different control and/or intervention target ranges, different methodology for glucose measurement, and different strategies for insulin administration were not able to reproduce these results. The reasons for these discrepancies remain unclear.
There is growing evidence from experimental models and patient studies in diabetes mellitus that not only the blood glucose level but also the fluctuations in blood glucose may increase the risk of hyperglycemia-induced oxidative stress. Constant exposure to hyperglycemia may induce cellular protective mechanisms against this toxicity, whereas these mechanisms may be absent or insufficiently activated by intermittent exposure. Several recent studies in critically ill patients revealed an association between intermittent exposure to elevated glucose, further referred to as blood glucose "amplitude variation," and mortality. Uniform standards to quantify amplitude variation are currently lacking, in line with the lack of solid knowledge on what exactly in the dynamic characteristics of glycemia is causing this association with adverse outcome. In several retrospective studies, different indices have been proposed: glucose variability index,SD of all blood glucose measurements per patient recorded during intensive care unit (ICU) stay,SD together with other glucose-related indices, glycemic lability index, and mean amplitude of glucose excursions. Based on these retrospective observational associations, it remains unclear whether one of the two, reducing blood glucose amplitude variation or normalizing the average blood glucose level, would be more important or even solely responsible for the potentially beneficial effects of glycemic control. Using data from the two Leuven prospective, randomized, controlled intervention studies, together comprising 2748 patients, we examined markers of glucose level and amplitude variation in relation to tight blood glucose control and mortality. In the subset of medical ICU patients, from whom time series of all blood glucose measurements also are available, we assessed the hyperglycemic index (HGI), the hypoglycemic index (HoGI), and the glycemic penalty index (GPI) as markers for being in the normoglycemic range, and the SD of all glucose values per patient (sd blood glucose) as marker of amplitude variation. In addition, we calculated jack-knifed approximate entropy (JkApEn) of the blood glucose signal, a marker of "pattern irregularity", referring to characteristic patterns of variation over time. Entropy is a regularity statistic: large values correspond to a greater randomness, and smaller values correspond to a more regular signal with more instances of recognizable patterns in the data. In general, aging and disease is associated with decreased or increased entropy, depending on whether the system at interest is homeostatic or cyclic. JkApEn has less bias in smaller samples than approximate entropy. We analyzed the effect of IIT on amplitude variation and pattern irregularity, and we assessed their relative association with mortality in a multivariable logistic regression model that also included measures of being in the normoglycemic range and baseline risks.