Does a lower interventional glucose threshold affect neurodevelopment in at-risk neonates?

Does a lower interventional glucose threshold affect neurodevelopment in at-risk neonates? 

Manuscript Citation: van Kempen AAMW, Eskes PF, Nuytemans DHGM, van der Lee JH, Dijksman LM, van Veenendaal NR, et al. Lower versus Traditional Treatment Threshold for Neonatal Hypoglycemia. N Engl J Med. 2020 Feb 6;382(6):534–44. 

Carolyn McGann*, MD
Fellow, Division of Neonatology
Children’s Hospital of Philadelphia
mcganncm@email.chop.edu

Sarah Coggins*, MD
Fellow, Division of Neonatology
Children’s Hospital of Philadelphia
cogginss@email.chop.edu

Kevin Dysart, MD
Associate Medical Director, Division of Neonatology

Children’s Hospital of Philadelphia
Associate Professor of Clinical Pediatrics, Perelman School of Medicine at the University of Pennsylvania
dysartk@email.chop.edu

*These authors contributed equally to this work.

Corresponding Author: 

Carolyn McGann, MD

Contact Address: 3401 Civic Center Boulevard Philadelphia, PA 19104

Phone: (267) 847-9362

Email: mcganncm@email.chop.edu

Keywords: hypoglycemia, newborn, randomized controlled trial, intention-to-treat, per-protocol

TYPE OF INVESTIGATION: Prognosis

QUESTION: In neonates born at > 35 weeks gestation who require hypoglycemia screening, does a lower glucose treatment threshold (36 mg/dL) compared to a traditional glucose threshold (47 mg/dL) affect psychomotor development at 18 months? 

METHODS

Design: Randomized Controlled Trial (Non-Inferiority)

Allocation: Neonates that had moderate hypoglycemia, defined as 36 to 46 mg/dL (2.0 to 2.5 mmol/L), were allocated in a 1:1 ratio, stratified by study center and predefined hypoglycemia risk factor subgroup. To ensure allocation concealment, a block randomization sequence was generated by web-based software at the time of enrollment.

Blinding: Due to the nature of treatment strategies, clinicians could not be blinded to the allocated group, however, psychologists assessing the primary outcome at 18 months were blinded to the study group assigned to the patient.

Follow-up period: Birth to 18 months

Setting: 17 academic or teaching hospitals in the Netherlands from October 2007 to April 2011

Patients: Neonates > 35 weeks and > 2,000g with one of the following risk factors for hypoglycemia: small for gestational age (SGA, birth weight <10^th percentile), large for gestational age (LGA, birth weight >90^th percentile), neonates born at 35 0/7 weeks to 36 6/7 weeks (near term or late preterm), infants of diabetic mothers (IDM)

Exclusion criteria: Severe perinatal asphyxia (three or more of the following: signs of intrauterine asphyxia, arterial umbilical cord pH <7.10, initiation of spontaneous respirations > 5 min after birth, 5 minute APGAR < 5, multiorgan failure), severe perinatal infection requiring support beyond antibiotic therapy, respiratory support requirement, hypotension requiring vasopressors, major congenital malformations or strong suspicion of a syndrome, inborn error of metabolism or strong suspicion of one, hyperinsulinism, first glucose <36 mg/dL

Intervention: The lower-threshold group had a target glucose of >36 mg/dL. 

Control: The traditional-threshold group had a target glucose of >47 mg/dL.

Glucose screening occurred at 1, 3, 6, 8, 12, 18 and 24 hours of life in all eligible neonates. Local feeding protocols were not adjusted for the study; thus, randomization was stratified by site. Infants who had a glucose less than the threshold for their assigned group had the carbohydrate intake increased by 1-2 mg/kg/min (enteral or IV) at the discretion of the clinical team and then had a glucose repeated in one hour with the goal of achieving target glucose levels within 3 hours. Symptomatic hypoglycemia was treated with 2.5 mL/kg dextrose 10% fluid regardless of assigned group.           

Outcomes:

Primary outcome: Psychomotor development at 18 months measured by the Bayley-III Scale

Secondary outcomes: Costs for medical treatments, number of glucose measurements, number and severity of hypoglycemia episodes, type of treatment required, length of hospital stay, breastfeeding duration, hospital admissions up to 18 months 

Analysis and Sample Size: Investigators set a noninferiority limit of -0.5 SD which translates to 7.5 points on the Bayley Scale, or about a one-month developmental delay.  Anticipating 15% withdrawal of study participants, a sample size of 200 for each of the 4 subgroups was estimated using a study power of 90% and a one-sided alpha level of 0.025.  They performed intention-to-treat analysis for the entire study population as well as the four subgroups.  For the primary outcome, they conducted post-hoc analyses using multiple imputation of missing data, which are the data presented.  For the secondary outcomes, unadjusted results are presented. 

MAIN RESULTS:

Of 5,363 eligible neonates, 722 had moderate hypoglycemia and 689 were randomized. As anticipated, 15.5% of patients did not have follow up at 18 months. Maternal and neonatal characteristics were similar for both groups.

Primary Outcome: The Bayley scores at 18 months were similar between the lower- and traditional-threshold groups; cognitive scores were 102.9 vs. 102.2 (mean difference 0.7, 97.5% CI -1.5 to 2.9) and motor scores were 104.6 vs 104.9 (mean difference -0.3, 97.5% CI -2.4 to 1.8), in the lower- vs traditional-treatment threshold groups respectively.  The noninferiority limit was not crossed. These results held true in post-hoc analyses adjusting for missing data both when it was assumed that missingness was completely at random and not random. 

Secondary Outcome: In the lower-threshold group, fewer neonates received IV dextrose boluses (1.8 vs. 3.4%), continuous IV glucose (6 vs. 21%), and tube feeds (5 vs 13%). Neonates in the lower-threshold group received fewer glucose measurements (6.4 vs. 7.0) and fewer supplemental oral feeds (5 vs 7), but had more episodes of severe hypoglycemia and two serious events that were unrelated to the trial. The hospital length of stay for both mothers and newborns, breastfeeding duration, and healthcare costs were not significantly different between groups. 

STUDY CONCLUSION: In healthy newborns with risk factors for hypoglycemia, a lower threshold for intervention for hypoglycemia did not lead to worse neurodevelopmental outcomes at 18 months when compared to a traditional glucose threshold. There was no difference in healthcare costs, breastfeeding duration or hospital length of stay. 

COMMENTARY: 

Uncertainty persists regarding the ideal threshold for therapeutic intervention in hypoglycemic infants. It is known that complex postnatal hormonal shifts, in addition to patient-level risk factors, predispose to hypoglycemia within the first days of life. Hypoglycemia is also associated with brain injury; energy failure in the setting of severe or prolonged hypoglycemia may lead to neuronal excitotoxicity and cell death. Multiple observational studies have identified an association between neonatal hypoglycemia and neurodevelopmental impairment in childhood^1,2,3, though adverse neurologic outcomes may be mitigated with appropriate hypoglycemia therapy⁴. In this multicenter non-inferiority trial, a lower interventional glucose threshold (<36mg/dL) was noninferior to a traditional higher threshold (<47mg/dL) with regard to neurodevelopmental scores at age 18 months. 

Hypoglycemia management in newborns is a controversial topic, as evidenced by the lack of alignment between the American Academy of Pediatrics Committee on Fetus and Newborn and the Pediatric Endocrine Society, who recommend targeting thresholds of >45 and >50mg/dL in the first 24 hours of life, respectively^5,6. Institutional approaches to hypoglycemia risk stratification, screening, and management also vary widely.  Stratified analyses demonstrated no impact of study center on the overall outcome in the HypoEXIT trial, and it is notable that infants in the lower-threshold group received fewer interventions for hypoglycemia (regardless of route of therapy), without acute severe adverse events. 

 While clinicians strive to identify hypoglycemia among at-risk infants, potential negative impacts of over-screening (such as late breastfeeding initiation) must be balanced,⁷ which they addressed in this study by choosing secondary outcomes related to burden, efficacy and healthcare costs. Both groups had low rates of exclusive breastfeeding which may have impacted the duration of breastfeeding.  Length of stay was similar between groups, at around 4 days. Differences in these two outcomes may have been apparent in settings with different feeding protocols, higher breastfeeding rates and shorter average length of stay after delivery. 

This study was analyzed by intention-to-treat (ITT), citing the need to allow a pragmatic approach to individualized patient care. However, given the non-inferiority design of this trial, an ITT analysis may have reduced strength to identify true treatment inferiority. Protocol adherence was not reported in this study, though a per-protocol analysis may have provided additional insight as to the efficacy of the traditional and lower glucose intervention thresholds.

Despite a well-defined prospective study, questions remain about interventional thresholds among IDMs with hypoglycemia. Because the IDM subgroup had unexpectedly low enrollment (only 82 of the estimated 200 patients required), this study was underpowered to fully assess the noninferiority of a lower glucose threshold in this population of infants. Hypoglycemia management in IDMs warrants further scrutiny, as the pathophysiology appears to be distinct from that of late preterm and/or growth-restricted infants (e.g., hyperinsulinemia-mediated vs. decreased glycogen stores and immature carbohydrate metabolism). 

This study contributed data suggesting that it may be safe to use lower glucose thresholds in at-risk newborns. However, this needs to be studied further in different hospital contexts, targeting particular subgroups and with longer developmental follow up to ensure safety.   

EBM LESSON:

An important consideration in statistical analysis of randomized controlled trial results is the approach to accounting for protocol adherence. Intention-to-treat analyses group participants based on their randomized treatment allocations, and results are analyzed without regard for patient cross-over, dropout, or protocol deviation. This is considered a conservative approach, as it maintains original randomization and sample size. However, because patients are analyzed according to treatment allocation and not necessarily to ultimate treatment received, the ability to detect differences in treatment effect is biased to the null hypothesis (that the treatments are not different from one another). In contrast, per-protocol analysis is restricted to only those participants who adhered to the treatment protocol they were initially randomized to, and excludes participants who cross over or drop out. Benefits of this approach include a better assessment of treatment efficacy, but it is also at risk for bias due to loss of randomization and potentially non-random protocol adherence as patients may receive different therapies than those they were randomized to and rates of protocol violations may differ between study groups.

However, non-inferiority trials are designed to test the null hypothesis that a treatment is inferior to a control (based on a predefined noninferiority margin), with the alternative hypothesis being that the treatment is not unacceptably worse than the control. In this case, a hybrid approach of intention-to-treat and per-protocol analysis schemes is recommended, taking into account data missingness/participant dropout and protocol deviation rate⁸. As intention-to-treat analyses are intrinsically biased to equivalence, use in a non-inferiority trial may erroneously conclude that a treatment is not worse than the control (when in reality, the treatment may truly be inferior). Per protocol analyses may be more conservative in non-inferiority trials, as they examine differences based on treatment received and are more likely to detect a treatment effect. However, protocol violations and missing data can bias this analysis in either direction.^8,9. If both analytic approaches demonstrate no difference between treatments, a conclusion of non-inferiority is strengthened; however, attention should be paid to missing data and protocol violations, and a data analysis plan should be determined a priori.

Acknowledgment: 

The Journal club is a collaboration between the American Academy of Pediatrics- Section of Neonatal Perinatal medicine and the International Society of Evidence- based neonatology (EBNEO.org)

REFERENCES:

1. Harris DL, Alsweiler JM, Ansell JM, Gamble GD, Thompson B, Wouldes TA, Yu T-Y, Harding JE. Outcome at 2 years after dextrose gel treatment for neonatal hypoglycemia: follow-up of a randomized trial. J Pediatr. 2016; 170: 54-59.
2. Kaiser JR, Bai S, Gibson N, Holland G, Lin TM, Swearingen CJ, Mehl JK, ElHassan NO. Association between transient newborn hypoglycemia and fourth-grade achievement test proficiency: a population-based study. JAMA Pediatr. 2015; 169(10): 913-21. 
3. Shah R, Harding J, Brown J, McKinlay C. Neonatal glycaemia and neurodevelopmental outcomes: a systematic review and meta-analysis. Neonatology. 2019; 115(2): 116-126.
4. McKinlay CJD, Alsweiler JM, Ansell JM, Anstice NS, Chase JG, Gamble GD, Harris DL, Jacobs RJ, Jiang Y, Paudel N, Signal M, Thompson B. Neonatal glycemia and neurodevelopmental outcomes at 2 years. NEJM. 2015; 373: 1507-1518.
5. AAP Committee on Fetus and Newborn. Postnatal glucose homeostasis in late-preterm and term infants. Pediatrics. 2011; 127 (3): 575-579.
6. Thornton PS, Stanley CA, De Leon DD, Harris D, Haymond MW, Hussain K, Levitsky LL, Murad MH, Rozance PJ, Simmons RA, Sperling MA, Weinstein DA, White NH, Wolfsdorf JI. Recommendations from the Pediatric Endocrine Society for evaluation and management of persistent hypoglycemia in neonates, infants, and children. J Pediatr. 2015; 167(2): 238-245.
7. Mukhopadhyay S, Wade KC, Dhudasia MB, Skerritt L, Chou JH, Dukhovny D, Puopolo KM. Clinical impact of neonatal hypoglycemia screening in the well-baby care. J Perinatol. 2020. https://doi.org/10.1038/s41372-020-0641-1
8. Sanchez MM, Chen X. Choosing the analysis population in non-inferiority studies: per protocol or intent-to-treat. Stat Med. 2006, 25(7): 1169-1181.
9. Schumi J, Wittes J. Through the looking glass: understanding non-inferiority. Trials. 2011; 12(1): 106.