Is glucose variability associated with worse brain function and seizures in neonatal encephalopathy?

Is glucose variability associated with worse brain function and seizures in neonatal encephalopathy?

MANUSCRIPT CITATION:

Pinchefsky EF, Hahn CD, Kamino D, et al. Hyperglycemia and Glucose Variability Are Associated with Worse Brain Function and Seizures in Neonatal Encephalopathy: A Prospective Cohort Study. J Pediatr. 2019;209:23-32.

REVIEWED BY:

Jayasree Nair, MD and Vasanth Kumar, MD

Department of Pediatrics, Division of Neonatology,

University at Buffalo, Buffalo, NY

CORRESPONDENCE:

Jayasree Nair, MD

Program Director, Neonatal-Perinatal Medicine Fellowship

Assistant Professor of Pediatrics, University at Buffalo

Attending Neonatologist, John R. Oishei Children’s Hospital

1001 Main Street, 5th Floor

Buffalo, NY 14203

T: 716.323.0260 | F: 716.323.0294

[email protected]

KEYWORDS: hyperglycemia, glucose variability, neonatal encephalopathy

TYPE OF INVESTIGATION: Prospective cohort study

QUESTION: In full term infants with neonatal encephalopathy within 6 hours after birth, does hypoglycemia or hyperglycemia, compared to normoglycemia, correlate with brain function measures by aEEG or seizures?

METHODS

Design: Prospective cohort study, part of an ongoing “Neurological Outcome of Glucose in Neonatal Encephalopathy (NOGIN)” study cohort

Exposure: Infants were grouped based on the occurrence of normoglycemia (51-144 mg/dl), hypoglycemia defined as blood or interstitial glucose ≤50 mg/dL (2.8 mmol/L) and hyperglycemia as glucose >144 mg/dL (8.0 mmol/L)

Blinding: A blinded device the iPro2 monitor was used to record interstitial glucose concentrations for 72 hours and this data was not made available to the clinical care team.

Follow-up period: The study provides data collected in the first 72 hours only.

Setting: A single referral tertiary care hospital in Toronto, Canada, between August 2014 and March 2017.

Patients: Full term infants with neonatal encephalopathy, defined as abnormal consciousness with either neonatal seizures or abnormalities in tone and reflexes.

Exclusion criteria: suspected or confirmed congenital malformations, inborn errors of metabolism, congenital infections, gestational age <36 weeks, weight <1500 g, or if it was expected that a continuous glucose monitor (CGM) could not be attached within 6 hours of life.

Intervention: Continuous glucose monitors were used to record average interstitial glucose concentrations every 5 minutes. Episodes of interstitial glucose derangements were defined as 2 or more consecutive data points (≥10 minutes) outside the normal range (51-144 mg/dl). Episodes of glucose derangement were treated as contiguous if they were separated by brief periods of normoglycemia lasting ≤10 minutes. Clinicians treated newborns for hypoglycemia or hyperglycemia based on current standard of care using intermittent glucose testing. As per institutional protocol, blood glucose concentrations <49 mg/dL (2.7 mmol/L) were treated with intravenous dextrose, increasing glucose infusion rates or glucagon. Hyperglycemia was treated with insulin infusion as clinically indicated.

Outcomes: The primary analyses compared aEEG scores (background, sleep–wake cycling, and seizure scores) during 6-hour epochs containing episodes of hypoglycemia or hyperglycemia to normoglycemic epochs.

Statistical Analysis and Sample Size: No power or sample size analysis are mentioned. For each subject, mean, minimum, and maximum glucose levels were calculated based on interstitial measurements obtained over a 6-hour epoch. The area under the curve was calculated as the area of interstitial glucose concentration over/under the normal glucose range (hours*mg/dL), reflecting the extent and duration of the episode of interstitial glucose derangement. Glucose variability was quantified using the SD, coefficient of variation (SD/mean), and mean glucose rate of change per hour (mg/dL/h).  The primary analyses compared aEEG scores as mentioned above during 6-hour epochs containing episodes of hypoglycemia or hyperglycemia to normoglycemic epochs using generalized estimating equations for repeated measures with a linear scale response model. An independent correlation structure and robust variance estimators were used. Hypo- or hyperglycemic epochs were compared with normoglycemic epochs within the same patient as well as the overall group. Findings were adjusted for clinical markers of hypoxia–ischemia severity: Apgar scores, umbilical artery pH, and base deficit. Secondary analyses investigated the association of the aEEG scores with other glucose measures (mean, maximum or minimum glucose values, duration of glucose disturbance, and area under the curve) and measures of glucose variability (SD, coefficient of variation, and mean glucose rate of change per hour). Two-tailed tests with P values of < .05 were considered statistically significant.

Results:

Of 80 subjects whose families were approached, 51 (64%) were enrolled of which 45 infants were eventually eligible for study analysis. Compared with normoglycemia, neonates with hypoglycemia had a greater mean gestational age and a greater mean umbilical artery pH. 37 infants (82%) demonstrated abnormal glucose values in the first 3 days - 29% hypoglycemia, 36% hyperglycemia and 18% with both. 34 episodes of glucose derangements were captured on CGM concurrent with aEEG monitoring; 16 episodes of hypoglycemia in 9 infants (20%) and 18 episodes of hyperglycemia in 13 (29%).

Both before and after adjustment for insulin treatment, aEEG epochs containing hyperglycemia displayed worse aEEG background scores (β 1.2, 95% CI 0.6-1.7 P< .001), less sleep–wake cycling (β 0.3, 95% CI 0.09-0.6, P= 0.009), and more frequent seizures (β 0.3, 95%CI 0.07-0.6, P= 0.01). No seizures were recorded during epochs with hypoglycemia. Greater fluctuation in interstitial glucose concentration (measured by a higher SD) was associated with worse sleep–wake cycling and more frequent seizures, and greater coefficient of variation was associated with more frequent seizures.

Study Conclusions

In neonates with encephalopathy, epochs of hyperglycemia were associated with worsening of brain function as assessed by aEEG, seizure burden and sleep-wake cycling scores.

Commentary

Despite advances in its management such as therapeutic hypothermia (TH), hypoxic-ischemic encephalopathy (HIE) is an important cause of mortality and adverse long-term neuro-developmental outcomes. Neonates with HIE are at increased risk of depleting their energy stores from transient hyperinsulinism  and developing concurrent hypoglycemia; the severity of encephalopathy and cellular damage varies with the severity of hypoglycemia (1). Early hypoglycemia in the first 6 hours of age is associated with the severity of HIE and adverse outcomes (death, neurodevelopmental impairment or significant motor disability) (2). Similarly, a higher incidence of death and/or moderate to severe disability has been reported with hyperglycemia and high glucose variability in infants undergoing TH for HIE (3, 4). However, in a more recent study, only hyperglycemic infants randomized to hypothermia had reduced risk of unfavorable outcome (defined as death and/or severe neurodevelopmental disability at 18 months) after adjusting for Sarnat stage and 5-min Apgar score; whereas, hypoglycemic and normoglycemic infants did not (5).

In view of the above contradictory findings, the study by Pinchefsky et al. is both interesting and timely. In neonates with encephalopathy, epochs of hyperglycemia, but not hypoglycemia were associated with worse global brain function and greater seizure frequency on aEEG monitoring. Worse background aEEG scores, sleep cycle scores and seizure episodes were noted during the hyperglycemia epochs. On longitudinal assessment of hypo- and hyperglycemic epochs, the time spent in hyperglycemia as well as maximum and mean glucose concentrations were significantly associated with aEEG scores.

Although the correlation between aEEG and hyperglycemia epochs was significant, the association between the two variables does not imply causation, which the authors clarify in the discussion. We often hope to show that changes in the explanatory variable (CGM) cause the changes in the response variable (aEEG). However, the observational study limits inferences regarding causality. Despite adjusting for clinical indicators of HIE severity (Apgar scores, umbilical artery pH and base deficit), an important confounder is the severity of HIE. Glucose derangement (hypo- and hyper) and variability could be in response to HIE or underlying predisposing conditions (eg maternal diabetes mellitus). The outcome measured relating to brain function, i.e. aEEG, seizure burden and sleep cycling scores may reflect the severity of HIE in these infants with hyperglycemic epochs, a proxy to abnormalities of brain function.

It is surprising that hypoglycemic epochs were not significantly associated with aEEG changes and seizure burden. The small number of infants studied and less time in hypoglycemia may account for the difference. Additionally, prompt and aggressive treatment of hypoglycemia, which is relatively standard in neonates, may have mitigated the effect. In contrast, treatment of hyperglycemia with insulin, as was seen in 3 out of the 16 infants with hyperglycemia, remains controversial with varying thresholds for therapy and adverse outcomes such as reduced linear growth and higher incidence of hypoglycemia (6).  A significantly higher cord pH and higher Apgar scores also suggest less severity of HIE in the hypoglycemic group, although the authors have adjusted for this in their final analyses.

In conclusion, this study adds to the body of literature on glucose variability in infants with HIE and provides new information linking glucose variability to adverse neurological outcomes.

EBM LESSON: Generalized estimating equations

A variety of statistical methods are available for analyzing longitudinal data in study participants. Ideally, analysis of longitudinal data must take into account the correlation between repeated observations on the same subject. That is, two observations from the same subject would be more similar than two individual measurements taken on distinct subjects. The classic method of analyzing studies involving more than one measurement on each subject is the repeated measures analysis of variance or repeated measures ANOVA. It compares means across one or more dependent variable(s) based on repeated observations. However, a limitation of repeated measures ANOVA is the inability to incorporate covariates, or characteristics of the participants other than the actual treatment (dependent variable), that may affect the outcome of the study. These limitations can be overcome by either mixed-effect modelling or use of generalized estimating equation (GEE) to fit a model for longitudinal/clustered data analysis. The mixed-effect model is an individual-level approach adopting random effects to capture the correlation between observations on the same subject. GEE is a population-level approach based on providing the population-averaged estimates of parameters (7) and is described as one of three most commonly used analytical approaches for group randomized trials (8).

The easiest way of modeling observations of repeated measurements is to use a linear model, wherein covariates have an additive effect on outcome. If the variables do not follow a linear relationship, a generalized linear model (GLM) would be appropriate. While using GLM, one must assume that errors are independent and identically distributed. However, this is not the case, as observations for each individual are correlated. Possible solutions to account for this individual correlation include use of generalized linear mixed modeling (GLMM), which requires assumptions of parametric data. GEE provides a nonparametric way to analyze such data.

GEE as proposed by Liang and Zeger, has several advantages (9). This is a computationally simple method compared to mixed-linear models with no distributional assumptions. Besides longitudinal data, which has measurements at different points in time, clustered data, in which measurements are taken on subjects who share a common characteristic can also be analyzed by GEE, making it a common modeling approach used in cluster randomized trials to account for within-cluster correlation (10). This method can be used with many different types of regression models. With GEE regression modelling, “the estimated regression coefficients are valid even if the correlation assumptions on which the analysis is based are not precisely correct” (11).

There are some limitations identified with GEE. This methodology has issues with small sample sizes, with the sandwich estimator tending to produce p-values that are too small (12). Additionally, the problems GEE experiences of a downward bias, thus underestimating the variances with finite sample sizes can become exacerbated when coupled with a rare outcome. Under these circumstances, alternate variance estimators such as Rogers, Pan or Morel could be better choices than the traditionally used Liang-Zeger method (13).

References

1. Basu P, Som S, Choudhuri N, Das H. Contribution of the blood glucose level in perinatal asphyxia. Eur J Pediatr. 2009;168(7):833-8.
2. Nadeem M, Murray DM, Boylan GB, Dempsey EM, Ryan CA. Early blood glucose profile and neurodevelopmental outcome at two years in neonatal hypoxic-ischaemic encephalopathy. BMC pediatrics. 2011;11:10.
3. Chouthai NS, Sobczak H, Khan R, Subramanian D, Raman S, Rao R. Hyperglycemia is associated with poor outcome in newborn infants undergoing therapeutic hypothermia for hypoxic ischemic encephalopathy. J Neonatal Perinatal Med. 2015;8(2):125-31.
4. Al Shafouri N, Narvey M, Srinivasan G, Vallance J, Hansen G. High glucose variability is associated with poor neurodevelopmental outcomes in neonatal hypoxic ischemic encephalopathy. J Neonatal Perinatal Med. 2015;8(2):119-24.
5. Basu SK, Salemi JL, Gunn AJ, Kaiser JR, CoolCap Study G. Hyperglycaemia in infants with hypoxic-ischaemic encephalopathy is associated with improved outcomes after therapeutic hypothermia: a post hoc analysis of the CoolCap Study. Archives of disease in childhood Fetal and neonatal edition. 2017;102(4):F299-F306.
6. Alsweiler JM, Harding JE, Bloomfield FH. Tight glycemic control with insulin in hyperglycemic preterm babies: a randomized controlled trial. Pediatrics. 2012;129(4):639-47.
7. Gibbons RD, Hedeker D, DuToit S. Advances in analysis of longitudinal data. Annu Rev Clin Psychol. 2010;6:79-107.
8. Feng Z, Diehr P, Peterson A, McLerran D. Selected statistical issues in group randomized trials. Annu Rev Public Health. 2001;22:167-87.
9. Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics. 1986;42(1):121-30.
10. Huang S, Fiero MH, Bell ML. Generalized estimating equations in cluster randomized trials with a small number of clusters: Review of practice and simulation study. Clin Trials. 2016;13(4):445-9.
11. Mathew DEF, V.T. Using and understanding Medical Statistics (textbook); Chapter 14. The analysis of longitudinal data. 2007.
12. Fitzmaurice GM, Laird NM, Ware JH. Applied longitudinal analysis: John Wiley & Sons (textbook); Chapters 13,14. 2011.
13. Rogers P, Stoner J. Modification of the Sandwich Estimator in Generalized Estimating Equations with Correlated Binary Outcomes in Rare Event and Small Sample Settings. Am J Appl Math Stat. 2015;3(6):243-51.