Impact of early glycemic variability on mortality and neurologic outcome of very low birth weight infants: Data from a continuous glucose monitoring system*

From January 2013 to January 2015, a single-center prospective study was conducted at the Neonatal Intensive Care Unit of the Department of Pediatrics, Jagiellonian University, a third level neonatal center in Krakow, Poland. Neonates were eligible for the study if they weighed 1500 g or less at birth and were admitted to the unit during the first day of life. Exclusion criteria were: major congenital malformations, a diabetic mother, a suspicion of an inborn error of metabolism, perinatal trauma, and asphyxia.

A continuous glucose monitoring system was used to measure glucose (Guardian Real-Time CGM®, Medtronic, Northridge, USA). The continuous glucose monitoring sensor (SofSensor®, Medtronic, Northridge, USA) was inserted into the subcutaneous tissue of the lateral side of the thigh, from which measurements were recorded every five minutes for six days. Both the accuracy and safety of the CGM system have been validated in VLBW neonates [5]. The CGM system was calibrated at least three times a day with a blood sample obtained from the radial or umbilical arterial lines (which were infused only with a normal saline solution). Point of care blood glucose analysis of samples routinely used for clinical management (Cobas b 221 POC, Roche Diagnostics, Germany) were used to calibrate the CGM system. The normal range of interstitial glucose concentration was predefined between 70 and 180 mg/dL. The incidence of hyperglycemia and hypoglycemia based on sensor interstitial glucose concentration was investigated. Glycemic variability (GV) was assessed by standard deviation (SD), coefficient of variation (CV) calculated as the percent value of SD divided by the mean glucose and mean amplitude of glucose excursion (MAGE) defined as

where ne is the number of excursions with amplitudes greater than one SD and Ag( i) is the ith glucose excursion greater than the SD of the whole monitoring. Both the data from the CGM system as well as the calibration values were not blinded and were available for clinicians. A uniform nutrition strategy and intervention protocol was used in the cases of predefined hyper- or hypoglycemia.

The objective was to evaluate the ability of early glycemic variability to predict death (primary outcome) and neurologic morbidity (secondary outcomes): retinopathy of prematurity requiring treatment (ROP), severe intraventricular hemorrhage (grade 3 or 4 IVH), and periventricular leukomalacia (PVL). We selected the following biomarkers of interest: glycemic variability assessed by the standard deviation of the mean glycemia (SD), the coefficient of variation (CV) and the mean amplitude of glycemic excursion (MAGE), episodes of hypoglycemia (<70 mg/dL), and episodes of hyperglycemia (>150 mg/dL).

The diagnosis of intraventricular hemorrhage was determined through cranial ultrasonography (Envisor HD 11, Philips, USA) performed, according to the standardized protocol, within the first day of life, then at 1 week of life, and subsequently at 1- to 2-week intervals. Hemorrhage grades 1 and 2 were determined as mild hemorrhage and grades 3 and 4 as severe. All ultrasonography was performed by neonatologists certified by the Polish Society of Ultrasonography.

Periventricular leukomalacia was recognized by magnetic resonance imaging of the brain performed for infants at near term-equivalent age (GA of 36-39 weeks) by using a 1.5 T (General Electric Healthcare, Milwaukee, USA) magnetic field induction MR system with 8-channel coil. Morphological brain changes were assessed using standard T1-weighted, T2-weighted and T2-FLAIR sequences. The scans were analyzed by a single neuroradiologist experienced in pediatric MRI examination.

Retinopathy of prematurity was diagnosed according to the guidelines of the International Classification of Retinopathy of Prematurity revisited [6]. Stages of ROP were determined according to the International Classification of ROP (Stages 1-5). In this study, examinations were performed by a pediatric ophthalmologist after mydriasis using a binocular indirect ophthalmoscope (Keeler Instruments Inc., USA) and was verified by digital imaging using a RetCam imaging system (Clarity Medical Systems, USA). Laser therapy was performed in infants with indications outlined in the ETROP criteria (ROP 1 Type): any stage of ROP with plus disease in Zone I; stage 3 ROP without plus disease in Zone I; stage 2 or 3 ROP with plus disease in Zone II.

The study protocol was approved by the Jagiellonian University Ethics Committee, and written informed consent was obtained from a parent of each infant.

Categorical variables are presented as numbers (n) and percentages (%); continuous variables are presented as median (interquartile range, IQR or 95% CI for median), arithmetic mean (± standard deviation, SD) for normally distributed data, and geometric mean (÷/× geometric standard deviation, GSD) for non-normal positively skewed data.

First, an initial univariate analysis was performed to determine the risk factors associated with outcomes, utilizing either Student's two-tailed t-test, Welch’s test, or the Mann-Whitney test to compare continuous variables and Fisher’s exact test to compare categorical values. Then, multiple logistic regression was used to evaluate the risk factors determined by univariate analyses. The odds ratio (OR) and 95% confidence interval (CI) were calculated for each risk factor. Assumptions of linearity and normally distributed residuals were visually checked using residual and P-P plots. The variance inflation factor (VIF) was used to exclude multicollinearity among predictors in the multiple logistic regression model. Probability values (p-values) below 0.05 were considered statistically significant. Analysis was carried out using MedCalc Statistical Software (version 16.8.4, Ostend, Belgium; https://www.medcalc.org; 2016).

Seventy-four patients were enrolled in the study. The baseline demographic and clinical characteristics of the patients are detailed in table I.

A total of 102,334 glucose measurements were delivered by the continuous glucose monitoring system. The average time below 70 mg/dL was 3.14%, and above 150 mg/dL, 10.35%. The mean glucose concentration and ranges of selected methods of glycemic variability assessment were calculated and are presented in table II. This shows the arithmetic means and standard deviation (SO) or the geometric means and geometric standard deviation (GSD) for positively skewed data.

Univariate analysis revealed that glycemic variability occurring in the first week of life, glucose median concentration, incidence of hypoglycemia <70 mg/dL, and incidence of hyperglycemia >150 mg/dL were not be associated with mortality before term-equivalent age (tab. III).

Univariate analysis suggested an increased risk for development of severe intraventricular hemorrhages for CV (p=0.025) and MAGE (p=0.032) (tab. IV). Simple analysis also revealed an increased risk for ROP requiring treatment for hyperglycemia (p=0.018) and glycemic variability (SD p=0.019; CV p=0.026; MAGE=0.029) (tab. V). There was no association between glucose disturbances occurring on the first week of life and risk for PVL (tab. VI).

Multivariate linear regression was performed to examine the associations between severe and explanatory variables selected based on results of univariate analyses and earlier research. To reduce multicollinearity SD and CV were removed from multiple regression model. The multivariate linear regression model revealed that only gestational age (OR 0.98; 95% CI 0.97–0.99) and duration of oxygen exposure (OR 1.02, 95% CI 1.01-1.36) are associated with severe ROP. The multivariate model did not show a statistical association between MAGE and risk for ROP requiring treatment (OR 1.74; 95% CI 0.57-5.32). The Hosmer-Lemeshow test result indicated the fitness of the overall model (tab. VII).

Multivariate analysis did not show a statistical association between glycemic variability and risk for severe intraventricular hemorrhages (OR 1.31; 95% CI 0.16-10.67). No interaction or multicollinearity problems were observed for the final model. The Hosmer-Lemeshow test result was 0.129, which indicated the fitness of the overall model (tab. VII).