Association Between Transient Newborn Hypoglycemia and Fourth-Grade Achievement Test Proficiency: A Population-Based Study

Importance  Prolonged neonatal hypoglycemia is associated with poor long-term neurocognitive function. However, little is known about an association between early transient newborn hypoglycemia and academic achievement.

Objective  To determine if early (within the first 3 hours of life) transient hypoglycemia (a single initial low glucose concentration, followed by a second value above a cutoff) is associated with subsequent poor academic performance.

Design, Setting, and Participants  A retrospective population-based cohort study of all infants born between January 1, 1998, and December 31, 1998, at the University of Arkansas for Medical Sciences who had at least 1 recorded glucose concentration (a universal newborn glucose screening policy was in effect) was conducted. Medical record data from newborns with normoglycemia or transient hypoglycemia were matched with their student achievement test scores in 2008 from the Arkansas Department of Education and anonymized. Logistic regression models were developed to evaluate the association between transient hypoglycemia and school-age achievement test proficiency based on perinatal factors. Common hypoglycemia cutoffs of a glucose level less than 35 mg/dL (primary) and less than 40 and 45 mg/dL (secondary) were investigated. All 1943 normoglycemic and transiently hypoglycemic infants (23-42 weeks’ gestation) were eligible for inclusion in the study. Infants with prolonged hypoglycemia, congenital anomalies, or chromosomal abnormalities were excluded from the study.

Exposure  Hypoglycemia as a newborn.

Main Outcomes and Measures  The primary outcome was proficiency on fourth-grade literacy and mathematics achievement tests at age 10 years. We hypothesized a priori that newborns with early transient hypoglycemia would be less proficient on fourth-grade achievement tests compared with normoglycemic newborns.

Results  Perinatal data were matched with fourth-grade achievement test scores in 1395 newborn-student pairs (71.8%). Transient hypoglycemia (glucose level <35, <40, and <45 mg/dL) was observed in 6.4% (89 of 1395), 10.3% (143 of 1395), and 19.3% (269 of 1395) of newborns, respectively. After controlling for gestational age group, race, sex, multifetal gestation, insurance status, maternal educational level and socioeconomic status, and gravidity, transient hypoglycemia was associated with decreased probability of proficiency on literacy and mathematics fourth-grade achievement tests. For the 3 hypoglycemia cutoffs, the adjusted odds ratios (95% CIs) for literacy were 0.49 (0.28-0.83), 0.43 (0.28-0.67), and 0.62 (0.45-0.85), respectively, and the adjusted odds ratios (95% CIs) for mathematics were 0.49 (0.29-0.82), 0.51 (0.34-0.78), and 0.78 (0.57-1.08), respectively.

Conclusions and Relevance  Early transient newborn hypoglycemia was associated with lower achievement test scores at age 10 years. Given that our findings are serious and contrary to expert opinion, the results need to be validated in other populations before universal newborn glucose screening should be adopted.

At birth, the continuous utero-placental-umbilical infusion of glucose ends and reaches the lowest values during the first 1 to 2 hours,¹ stimulating counterregulatory mechanisms and promoting successful glucose homeostasis in healthy newborns. This occurrence is critical because the newborn brain principally uses glucose for energy and prolonged hypoglycemia has been associated with poor long-term neurodevelopment and neurocognition.^2-10 Nevertheless, little is known about whether early transient hypoglycemia, frequently considered to be a normal physiological phenomenon with no serious sequelae,^1,11 is associated with cognitive impairment.^3,4,8,12-14

Most newborn hypoglycemia outcome studies have limited generalizability because of the evaluation of at-risk infants only, a lack of nonhypoglycemic control subjects, and small sample sizes (including only symptomatic newborns, although hypoglycemic signs are nonspecific).¹⁵ As identified in the review by Boluyt et al,¹⁵ hypoglycemia cohort studies^{2,4-10,14,16,17} during the first week of life have reported follow-up findings at a mean age of 2.5 years. More recently, Tin and colleagues¹⁸ reported on the 15-year follow-up of preterm newborns with recurrent hypoglycemia and found no difference in their IQs compared with controls. However, that study and most other hypoglycemic outcome studies (except for the study by Lucas et al⁸) failed to control for maternal educational level and socioeconomic status, factors that are highly associated with childhood neurocognitive outcome and educational performance.^18-23

The American Academy of Pediatrics 2011 report on glucose homeostasis acknowledged that screening and management of newborn hypoglycemia is “a controversial issue for which evidence is lacking but guidance is needed.”^11(p575) These guidelines are based on expert opinion and lack support from high-quality long-term follow-up studies.^11,13,24 Therefore, it remains to be determined whether transient newborn hypoglycemia (defined as a single initial low glucose concentration, followed by a second concentration above a cutoff) is associated with subsequent impaired scholastic performance. To address this knowledge gap, we compared initial newborn glucose concentrations in 1998 from the universal glucose screening database available at the University of Arkansas for Medical Sciences (UAMS) with their 2008 matched student achievement test scores.^8,13 We abstracted perinatal and maternal factors evaluated in earlier neonatal hypoglycemia studies and included maternal educational level and socioeconomic status. Our objective was to evaluate an association between transient newborn hypoglycemia and proficiency on school-age achievement tests, which are real-world assessments that predict high school graduation, college attendance, and long-term adult success.^25-28 We considered the hypoglycemia cutoffs of a glucose level less than 35 mg/dL (primary) and less than 40 and 45 mg/dL (secondary) to define hypoglycemia^8,13 and hypothesized a priori that newborns with early transient hypoglycemia would be less proficient on fourth-grade achievement tests compared with normoglycemic newborns with similar perinatal factors. (To convert glucose concentration from mg/dL to millimoles per liter, multiply by 0.0555.)

Box Section Ref ID

Included in the study were all infants (23-42 weeks’ gestation) born at the UAMS in 1998 who had at least 1 recorded glucose concentration and who survived to hospital discharge without major congenital anomalies (including microcephaly) or chromosomal abnormalities. Thirty-four infants with prolonged hypoglycemia (having at least the first 2 glucose concentrations under a cutoff) were excluded. The study received UAMS institutional review board approval. In addition, a Health Insurance Portability and Accountability Act of 1996 waiver and an Arkansas Department of Education Family Educational Rights and Privacy Act waiver were obtained.

The UAMS universal newborn glucose screening policy was to obtain an early glucose concentration from all newborns, and compliance exceeded 99%. Plasma glucose concentrations (including collection, laboratory receipt, and verification date and times) were available from the laboratory database and were determined in the pediatric laboratory within the neonatal intensive care unit using the glucose oxidase method (Glucose 2 Analyzer; Beckman Coulter). The pediatric laboratory met approval standards of the Clinical Laboratory Improvement Act, and the equipment was routinely checked to ensure quality control. While there were no written hypoglycemia treatment guidelines in 1998, generally newborns with initial (1-3 hours after birth) glucose concentrations of 35 mg/dL or less received intravenous dextrose or an early feeding. Follow-up values for hypoglycemic infants were obtained approximately 1 hour after the initial value.

Four gestational age groups were defined. These included (1) full term (≥37 to 42 weeks), (2) late preterm (≥34 to <37 weeks), (3) preterm (≥28 to <34 weeks), and (4) extremely low gestational age (<28 weeks) newborns. Gestational age from medical records was uniformly recorded (eg, 26 weeks’ gestation would apply to newborns identified as “26+ weeks” or “26 weeks and 0-6 days”).

Data were retrieved from the 1998 UAMS medical records and from the Arkansas Department of Health birth certificate and Arkansas Department of Education databases. Data abstracted from the newborn record included name, birth date and time, birth weight, estimated gestational age, size for gestational age, race, sex, Apgar scores, singleton or multiple status, and the presence of polycythemia (hematocrit ≥65%). Data abstracted from the mother’s record included name, age, insurance status, educational level, gravidity, prenatal care, medical conditions, delivery route, pregnancy-associated and obstetrical conditions, and history of smoking or substance abuse. Data were stored in Research Electronic Data Capture (REDCap)²⁹ (1UL1RR029884) hosted at the UAMS Translational Research Institute.

Newborn names and birth dates from medical records and Social Security numbers from the Arkansas Department of Health birth certificate database were compared with student names, birth dates, and Social Security numbers from the Arkansas Department of Education database by one of us (G.H.) who is an expert in identity resolution and matching for longitudinal education data.³⁰ Positive identification was defined as an exact match of the Social Security number or (if a Social Security number was not available) a newborn’s name and birth date with those of a student’s. Multiple techniques were used to increase the number of positive identifications for newborn-student pairs while limiting false positives by using frequency-based confidences.³⁰ If positive identification could not be made with the newborn’s name, birth date, or Social Security number, a combination of the newborn’s information and the mother’s name was used. If unsuccessful, no further matching attempts were made.

To protect confidentiality, data were securely transmitted to the Arkansas Department of Health, where Social Security numbers were added. This new file was then securely transmitted to the Arkansas Department of Education, where achievement test results were added and anonymized. This file was transmitted back to us as an encrypted file. The final data set included UAMS-born participants who were successfully matched with their student achievement test scores.

The Benchmark Examination was developed to assess student competencies in literacy and mathematics according to Arkansas education standards for each grade, and all Arkansas public school students in grades 3 through 8 have been mandated to take the tests since 1997 (http://www.ArkansasEd.org). Scores were designated categorically as advanced, proficient, basic, or below basic based on scaled scores of 0 to 1000, assigned according to the percentage of correct answers. Proficient or advanced scores (proficient) represent performance at or above grade level, while basic and below basic scores (nonproficient) represent performance below grade level. Children with significant cognitive disabilities do not take the Benchmark Examination and were not included in the analyses.

The primary outcome variables were proficiency (yes or no) on fourth-grade literacy and mathematics achievement tests. We chose fourth-grade (age 10 years) achievement test proficiency as our primary outcome of interest because children were first exposed to the Benchmark Examination in third grade and would be expected to be more familiar with achievement test procedures. In addition, other studies^31-33 have linked perinatal characteristics with school-age cognitive, academic, and school performance at ages 8 to 11 years.

To compare characteristics of hypoglycemic and normoglycemic newborns, we used a 2-sample t test with unequal variances for continuous variables, rank sum test for count variables, and Fisher exact test for categorical variables for the 3 hypoglycemic cutoffs of a glucose level less than 35 mg/dL (primary) and less than 40 and 45 mg/dL (secondary). The odds ratios (ORs) and 95% CIs from the univariate logistic regression for potential independent predictors of achievement test proficiency were then determined.

Variables with P < .10 from the univariate analysis were considered as possible covariates for the multivariable logistic regression models. A stepwise backward elimination method was used in which the least significant variable was removed at each step until the remaining variables were significant (P < .05) (STATA 13; StataCorp LP). Collinearity was evaluated using a variance inflation factor. Adjusted ORs and 95% CIs for variables in the final logistic regression models for literacy and mathematics proficiency by the 3 cutoffs were then determined.

We matched 1395 of 1943 newborns (71.8%) having normoglycemia or transient hypoglycemia with their achievement test scores. Characteristics of the study sample by hypoglycemic and normoglycemic status using each of the hypoglycemia cutoffs are listed in Table 1 and Table 2. The mean (SD) birth weight and estimated gestational age of the matched cohort were 2881 (855) g and 36.8 (3.8) weeks, respectively. Most newborns were full term and late preterm. Overall, 94.7% (1321 of 1395) of newborns were of black or white race, and 50.3% (702 of 1395) were male. Five-minute Apgar scores less than 7 were present in 6.4% (88 of 1395) of infants. In addition, 5.9% (82 of 1395) of infants were from multifetal gestations. As is typical of a university hospital population, 81.5% (1137 of 1395) of families had Medicaid or no insurance, and 25.4% (335 of 1319) of mothers had an educational level beyond high school. The mothers of 4.9% (68 of 1395) of newborns had diabetes mellitus. The cesarean section rate was 31.0% (432 of 1395). Most characteristics of newborns unmatched to achievement test scores were equivalent to those of matched infants^34,35 except that there were fewer black infants and fewer newborns with Medicaid in the unmatched cohort (eTable 1 in the Supplement).

In general, hypoglycemic infants were smaller, less mature, less commonly full term, and more commonly from multifetal gestations. In addition, they were more frequently polycythemic, infants of mothers with diabetes mellitus, and born to mothers with pregnancy-related conditions and were less likely to have been delivered vaginally. These results are summarized in Table 1 and Table 2.

Among 1395 matched newborns, the mean (SD) initial glucose concentration was 59.5 (19.9) mg/dL (range, 13-231 mg/dL), and the median (interquartile range) was 57 mg/dL (48-68 mg/dL) (Figure 1). The median (interquartile range) time to specimen collection was 89 minutes (68-115 minutes) after birth, and specimen results were reported within 25 minutes after collection. This short turnaround time was because of staffing with dedicated phlebotomists and the presence of a pediatric laboratory within the neonatal intensive care unit. Transient hypoglycemia occurred in 6.4% (89 of 1395), 10.3% (143 of 1395), and 19.3% (269 of 1395) of newborns with cutoffs of a glucose level less than 35, 40, and 45 mg/dL, respectively. For transiently hypoglycemic newborns, follow-up glucose concentrations were obtained at a median (interquartile range) of 70 minutes (22-156 minutes).

The mean (SD) fourth-grade literacy test score and the proficiency rate were 544 (165) and 32% for hypoglycemic (<35 mg/dL) newborns vs 583 (195) and 57% for normoglycemic newborns (≥35 mg/dL). The mean (SD) mathematics test score and the proficiency rate were 562 (94) and 46% for hypoglycemic newborns vs 589 (104) and 64% for normoglycemic newborns. Test scores for hypoglycemic and normoglycemic newborns were similar among the 3 cutoffs (Figure 2).

The final logistic regression models for literacy and mathematics contained the same covariates irrespective of the hypoglycemic cutoff used (Table 3). Because birth weight and gestational age were highly collinear, we used only one of these covariates (the categorical classification of gestational age groups) in the final models. Collinearity issues were not found between the remaining covariates in the final models.

Transient hypoglycemia was significantly associated with decreased probability of proficiency on literacy achievement tests (adjusted ORs, 0.49, 0.43, and 0.62) and on mathematics achievement tests (adjusted ORs, 0.49, 0.51, and 0.78) for the 3 hypoglycemia cutoffs, respectively. Factors positively associated with increased probability of proficiency on achievement tests in all 6 models were female sex, singleton birth, full-term status, white and other race, private insurance, maternal educational level exceeding high school, and primiparous birth. The covariates had similar effect sizes in the independent literacy and mathematics models irrespective of the cutoff used. All logistic regression models fit the data adequately and demonstrated acceptable discrimination. When excluding the most premature and high-risk infants (extremely low-gestational-age newborns), similar associations were observed between transient hypoglycemia and achievement test performance compared with the whole study sample (eTable 2 in the Supplement).

Despite more than 50 years of neonatal hypoglycemia research, uncertainty remains about which newborns to screen, whether transient hypoglycemia has untoward long-term effects, and what concentration or range of glucose concentrations should be used to define neonatal hypoglycemia, as well as about management strategies and the incidence of transient hypoglycemia. Moreover, a link between transient hypoglycemia and poor neurodevelopmental long-term outcome has been difficult to establish because recommendations for screening and managing hypoglycemia are largely empirical and not based on long-term follow-up studies or neurodevelopmental testing.^11,13,24 Most important, we found that early transient newborn hypoglycemia was not uncommon (6.4%, 10.3%, and 19.3% of newborns using cutoffs of <35, <40, and <45 mg/dL, respectively), and our observed incidences were consistent with but appropriately lower than the incidences from higher-risk populations.³⁶ By leveraging existing data and resources available in Arkansas, including the UAMS policy of universal newborn glucose screening, it was determined that most infants born at the UAMS remain in Arkansas and later attend public schools. Because of excellent working relationships with the Arkansas Department of Health and Department of Education, we were well positioned to link early transient newborn hypoglycemia with achievement test performance. After controlling for multiple covariates and using several cutoffs to define hypoglycemia, we observed that transient hypoglycemia in a heterogeneous cohort of newborns born at a university hospital was associated with lower fourth-grade achievement test scores, real-world assessments that predict educational and economic success.^25-28

Consistent with previous studies, we observed that newborns with private insurance (a proxy for socioeconomic status²²) were more proficient on fourth-grade tests than newborns who had Medicaid or who were self-pay.¹⁹ White race and female sex were associated with greater proficiency than black race and male sex.^19,21 Newborns whose mothers had more years of education had higher test scores. Newborns of primiparous women fared better than those who had siblings.²⁰ After controlling for multiple variables, we also found that maternal and obstetrical diagnoses were not associated with test proficiency.

Preschool-aged children with severe and prolonged hypoglycemia as newborns generally are neurodevelopmentally delayed.⁶ In a large study by Lucas et al⁸ of premature infants, recurrent moderate hypoglycemia was associated with developmental delay at age 18 months. In contrast, Tin and colleagues¹⁸ replicated the study by Lucas et al⁸ and reported no difference in IQs between adolescents who had recurrent hypoglycemia (glucose level ≤45 mg/dL) compared with matched controls. Despite an unparalleled 15-year follow-up, the study by Tin and colleagues¹⁸ did not control for socioeconomic status or maternal educational level, factors known to be associated with childhood psychometric and academic performance.^19-23 Both the study by Lucas et al⁸ and our study adjusted for socioeconomic status and maternal educational level and observed differences in neurodevelopmental and educational outcomes.

A study by Brand and colleagues,⁵ considered to be of high quality by Boluyt et al,¹⁵ evaluated the effects of transient hypoglycemia during the first day of life in a cohort of 75 healthy term large-for-gestational-age infants and detected no significant difference in neurodevelopment between children with and without hypoglycemia at age 4 years, although normoglycemic children performed significantly better on a reasoning subscale test. Brand and colleagues⁵ did not adjust for socioeconomic status or maternal educational level and acknowledged in their discussion that the observed suggestion of poorer outcomes in transiently hypoglycemic infants could have been significant with a larger sample size.⁵ Compared with the study by Brand and colleagues,⁵ our study had a much larger sample size, investigated a heterogeneous population of newborns instead of at-risk infants only, followed up newborns much longer, and controlled for maternal educational level and socioeconomic status.

There has been considerable debate about whether a single glucose concentration or an operational threshold should be used to define neonatal hypoglycemia and to guide interventions,¹³ as well as about whether the same threshold is appropriate for all newborns. Plasma glucose concentrations of 18 to 47 mg/dL have been used in other studies^24,37 as cutoffs for defining hypoglycemia. In the study by Lucas et al,⁸ the investigators statistically tested cutoffs of 9 to 72 mg/dL and found that a cutoff of 47 mg/dL (during the first 9 weeks of life) was most highly correlated with 18-month development. Instead of using post hoc analyses to find the most significant cutoff, we examined 3 glucose concentrations as possible conservative cutoffs for defining hypoglycemia based on clinical considerations and practices used by neonatologists in 1998.

This study has notable strengths that have real-world implications. These include the large sample size and the consideration of perinatal covariates, maternal educational level, and socioeconomic status. In addition, our study used several clinically relevant cutoffs and glucose concentrations from a universal glucose screening database rather than from at-risk infants only, as well as objective, valid, and blinded outcomes. Our primary outcomes at age 10 years exceeded the mean 2.5-year follow-up from most previous hypoglycemia cohort studies.^{2,4-8,10,14,16,17}

However, there were some limitations to our study. We used retrospective observational data, and glucose concentrations were based on specimens collected at times that were at the discretion of bedside nurses. Glucose concentrations were evaluated only for the first 2 values, and we could not determine the duration of hypoglycemia or the minimal glucose concentration and their effects on test scores. Furthermore, hypoglycemia treatment strategies were not examined. In addition, we were able to match only 71.8% of newborns with their achievement tests. However, most characteristics were equivalent between matched and unmatched infants. The limitations of using existing data from medical records and state birth certificate and student achievement test databases in a retrospective cohort study are far outweighed by the ability to make valid assessments about newborn glucose homeostasis based on our careful conservative matching techniques.³⁰ Information regarding signs of hypoglycemia were inconsistently available from medical records, and anecdotally most newborns with transient hypoglycemia were asymptomatic. We could not account for personal childhood characteristics (primary household language, disabilities, school absenteeism, etc) and diagnoses or for 10 years of environmental influences and nutrition, similar to previously performed hypoglycemia cohort follow-up studies. Barker et al³⁸ also found associations between fetal or infant parameters and adult disease and did not account for intervening influences. Given that so many factors not accounted for in this study could have affected fourth-grade test performance, the fact that at age 10 years we still observed differences in school performance between transiently hypoglycemic and normoglycemic newborns is noteworthy. Although this single-center study has limitations and its results may be most generalizable to university hospital populations, establishing associations between transient hypoglycemia and childhood achievement is important in informing future newborn hypoglycemia recommendations and may serve as preliminary data for a definitive prospective trial.

Current guidelines recommend screening only in newborns with symptomatic hypoglycemia or those at risk of developing hypoglycemia.^11,24 Contrary to these expert opinion guidelines, our study suggests that early transient newborn hypoglycemia is associated with poorer academic performance at age 10 years. While our study did not prove that transient newborn hypoglycemia causes poor academic performance, we believe that the findings raise legitimate concerns that need to be further investigated in other newborn cohorts. Until our results are validated, however, universal newborn glucose screening should not be adopted. High-quality long-term follow-up studies are needed to direct future newborn hypoglycemia screening and treatment guidelines.

Accepted for Publication: May 20, 2015.

Corresponding Author: Jeffrey R. Kaiser, MD, MA, Section of Neonatology, Department of Pediatrics, Baylor College of Medicine, 6621 Fannin St, Mail Code WT 6-104, Houston, TX 77030 (jrkaiser@texaschildrens.org).

Published Online: August 24, 2015. doi:10.1001/jamapediatrics.2015.1631.

Author Contributions: Dr Kaiser had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Kaiser.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: All authors.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Bai, Swearingen.

Study supervision: Kaiser.

Conflict of Interest Disclosures: None reported.