Gestational Age at Term and Educational Outcomes at Age Nine

We use data from the Fragile Families and Child Wellbeing (FFCWB) study, a national birth cohort study that randomly sampled births in 75 hospitals in 20 large US cities from 1998 to 2000. By design, ∼3/4 of the mothers were unmarried. Face-to-face interviews were conducted with 4898 mothers while still in the hospital after giving birth^21,22 ; additional information was collected from mothers’ and infants’ medical records from the birth hospitalization in 61 of the hospitals. The availability of medical records depended primarily on administrative processes of hospitals rather than decisions of respondents to make their records available. Mothers who completed postpartum (baseline) interviews were reinterviewed 1, 3, 5, and 9 years later. During the 9-year follow-up, interviewers obtained consent to contact the children’s teachers.

The eligible sample consisted of 3088 children born at 37 to 41 weeks’ gestation (singleton births only) with available medical record data (needed to ascertain GA) and no missing analysis variables from the postpartum survey (Fig 1). Of those, 1683 cases had no or incomplete teacher survey data, leaving an analysis sample of 1405 children.

We used a continuous measure of completed weeks’ gestation and a set of indicators for early term (37–38 weeks), term (39–40 weeks), and late term (41 weeks), as distinguished by the American College of Obstetricians and Gynecologists and the Society for Maternal-Fetal Medicine.

When the child was 9 years old, their teacher was asked: “Overall, how would you rate this child’s academic skills in each of the following areas, compared to other children of the same grade level?” The areas were mathematics, science and social studies, and language and literacy, and the response choices for each were far below average, below average, average, above average, or far above average. For each area, we created a binary outcome for average, above average, or far above average (versus far below average or below average) to characterize the child not being behind in terms of performance, but we assessed sensitivity to other cutoffs and to continuous measures. Additionally, we created a binary outcome for average, above average, or far above average ratings in all 3 subjects.

Teachers also completed the Cognitive Problems/Inattention Subscale of the Conners Teacher Rating Scale–Revised Short Form (CTRS-RSF),²³  in which they were asked to rate children on a number of common problems children have in school. Specifically, the teacher was asked to rate the child on a scale of 0 to 3 (0 = not true, 1 = just a little true, 2 = pretty much true, 3 = very much true) for each problem in the past month. We coded the child as having performed adequately in mathematics if the teacher reported “not true” or “just a little true” for “poor in arithmetic” and in reading if the teacher reported “not true” or “just a little true” for “not reading up to par.”

All adjusted models controlled for the child’s sex, the mother’s first birth, and measures of socioeconomic status, including the mother’s age, race and ethnicity, foreign-born status, education, marital status, and Medicaid coverage for the delivery (versus private or other insurance), all assessed at baseline (see Table 1 for age, race and ethnicity, and education categories). Medicaid is government-provided health insurance in the United States that is available to low-income pregnant women. Some models additionally controlled for obstetric risk factors and delivery complications assessed from the mothers’ medical records (detailed lists in notes to Table 1), any prenatal smoking, late prenatal care (initiated after first trimester), missing information on prenatal care timing, and cesarean delivery. In some models, we included measures of neonatal morbidity (respiratory conditions, hypoglycemia, and hyperbilirubinemia) to explore those conditions as potential pathways between GA and educational outcomes. Supplementary models included the infant having been in a NICU as an alternative measure of neonatal morbidity and alternatively excluded children born small for gestational age and low birth weight (<2500 g) to eliminate potential confounding from those characteristics.

First, we compared our analysis sample to the cases in the full cohort (excluding multiples) of 4803 that were not in our sample using t tests (for binary variables) or χ² tests (for categorical variables). Second, we documented children’s educational outcomes and sample characteristics, overall and by early term, term, or late term. Third, we estimated unadjusted and adjusted logistic regression models of associations between GA and children’s academic outcomes. The first adjusted model controlled for child sex and maternal characteristics, the second added obstetric risk factors and delivery complications, and the third further added neonatal morbidities. All estimates are presented as odds ratios (ORs) with 95% confidence intervals (CIs). We also replicated the full set of logistic regression analyses using inverse probability weights to account for sample loss from the eligible sample of 3188 children (Fig 1) and estimated models that limited the sample to children of United States-born mothers to further explore potential bias resulting from loss to follow-up.

Analyses were conducted by using Stata Version 15.0 (Stata Corp, College Station, TX). This study was approved by the Rutgers Biomedical and Health Sciences Institutional Review Board.

Mothers of children in the analysis sample did not differ significantly from those in the original cohort that were not in our sample in terms of education, race and ethnicity, marital status, first birth, or Medicaid status, but they were slightly younger (24.9 vs 25.4 years) and less likely to be foreign born (13.9% vs 18.6%) (not shown in tables).

Approximately 28% of the children were early term, 60% were term, and 13% were late-term (Table 1). Roughly one-half of the mothers (48%) were non-Hispanic Black, 64% had a high school education or less, and 76% were unmarried, reflecting the oversampling of nonmarital births in the FFCWB study and strong associations between nonmarital childbearing, poverty, and entrenched social disadvantages associated with racial minority status in the United States.²⁴  Almost two-thirds of deliveries (63%) were covered by Medicaid. Of mothers with information on prenatal care timing, 51% initiated care in the first trimester (0.43/[0.43 + 0.41] = 0.51). Approximately one-fifth (18%) of the mothers smoked during pregnancy.

In both unadjusted and adjusted models, the continuous measure of GA in weeks was significantly associated with the odds of average or above-average rankings in mathematics, science and social studies, language and literacy, and all subjects combined at age 9 (Table 2). The associations were similar across outcomes (eg, mathematics [OR: 1.13; CI: 1.02–1.25], science and social studies [OR: 1.13; CI: 1.01–1.26], language and literacy [OR: 1.16; CI: 1.05–1.28], all subjects [OR: 1.16; CI: 1.05–1.27] in the model that adjusted for child sex, maternal characteristics, and obstetric risk factors and complications; model 3). Adjusting for socioeconomic factors attenuated the ORs somewhat, but adjusted estimates were insensitive across model specifications.

The corresponding models using indicators for early term and late-term birth (versus term) instead of the continuous measure of GA indicate positive associations between late-term birth and mathematics (eg, OR: 1.50; CI: 1.03–2.18; Model 7) and negative associations between early-term birth and language/literacy (eg, OR: 0.75; CI: 0.58–0.97; Model 7). Again, adjusted estimates were insensitive across model specifications.

Children of married and more educated mothers had higher odds of positive teacher ratings than those with unmarried and less educated mothers. Children with non-Hispanic Black mothers had worse teacher ratings than those with non-Hispanic white mothers (Table 3).

The patterns of estimates for this outcome were consistent with those above for performance in mathematics and language/literacy, but most estimates were not statistically significant, perhaps because the questions pertained to a short time period. A clear exception is that the associations between continuous GA and adequate performance in reading were statistically significant across all models. Again, the ORs were largely insensitive to model specification. Covariate estimates were similar to those reported on above (Supplemental Table 5).

Supplementary models that controlled for NICU stay, excluded children born small for GA, or excluded children who were low birth rate produced estimates similar to those in Tables 2 and 4, as did models that used different cutoffs for the educational outcomes, used continuous outcome measures of academic performance, or included indicators of each type of neonatal morbidity instead of the combined measure (not shown). Models that limited the sample to US-born mothers also produced similar estimates, suggesting that sample attrition by maternal nativity did not bias the estimates (not shown). Finally, using inverse probability weights to account for sample loss did not substantively or substantially change the estimates (not shown).

In a US birth cohort of children born at term (37–41 weeks), GA was positively associated with teacher-reported ratings of academic performance in mathematics, science and social studies, and language and literacy at age 9. Findings also suggested a general pattern of worse outcomes for children born at early term (37–38 weeks) and better outcomes for those born at late term (41 weeks) compared with those born at term (39–40 weeks). Compared with term, early term was negatively and significantly associated with performance in language and literacy and late-term was positively and significantly associated with performance in mathematics; the former is a common finding in the literature, whereas the latter is consistent with findings from some studies suggesting that a 41st week of GA is advantageous for academic performance.^7,12,19  Corresponding analyses using teachers’ reports of adequate performance in mathematics and reading in the past month indicated the same patterns but fewer statistically significant associations.

This study adds to the previous literature, which considered a variety of educational outcomes, ages of children, and populations, by using a national US-based sample, assessing children aged >5 years, and exploring potential mediating roles of neonatal morbidities. The most comparable US-based study was focused on standardized test results at age 8 in New York City,¹⁷  whereas our study focused on teacher reports of academic performance at age 9 in a national urban sample. Test scores and teacher ratings of academic performance are distinctly different outcomes because a student can score well on a test but perform poorly in school or vice versa. As far as we know, in no previous studies have researchers explored neonatal morbidities as potential pathways linking GA to educational outcomes. We found that, although neonatal respiratory conditions, hyperbilirubinemia, and other morbidities are associated with both GA^25,26  and cognitive and educational outcomes,^27,28  even at term,^29,30  neonatal morbidities did not appear to underlie the associations between GA and teacher-reported educational outcomes among children born at term.

Adverse effects on cognition related to earlier birth may reflect alterations in brain development postnatally compared with in utero. Brain development begins soon after conception and continues into early adulthood, but there is increasing evidence that transferring brain growth and development from an intrauterine environment to an extrauterine environment is less than optimal, even in the absence of other neurologic risk factors, and that the human brain continues to develop uniquely in utero until the end of gestation.³¹  Although much of neurogenesis is complete by the third trimester, the ability of neurons to communicate is just starting. Synaptogenesis increases considerably in the last 3 months of gestation and continues postnatally in a period of exuberant synaptogenesis.³²  The greatest increase over the last month of gestation occurs in the visual and auditory cortex, followed by areas responsible for receptive language and speech production.³³  These new synapses are the most significant contributor to brain volume growth. Quantitative MRI data demonstrate that brain tissue volume between 29 and 41 weeks increases linearly, with the most substantial portion in the cerebral cortex.³⁴  Additionally, there is a fivefold increase in myelinated white matter between 35 and 41 weeks³⁴ ; consequently, lower GA limits the brain’s ability to effectively transmit information between neurons. These changes in white matter microstructure are particularly associated with improved mathematics performance.³⁵ 

The findings from this study also highlight the importance of disadvantaged minority status, nonmarital birth, and low maternal education as risk factors for poor school performance. These findings, in conjunction with our findings for GA, underscore the importance of comprehensive educational supports, especially in low-income neighborhoods. Children with inherent developmental risk, such as those born prematurely or with a known disability, are regularly referred for evaluation and support services, but children who are not thought of as high risk may be overlooked. Initiatives such as universal prekindergarten may help compensate for these unseen risks and provide for postpartum catch-up brain development.

A strength of this study was the use of data from a national US birth cohort study with data from medical records and maternal surveys linked to teacher surveys. These data allowed us to control for a broad array of potentially confounding socioeconomic, obstetric, and delivery-related risk factors; explore neonatal morbidities as potential pathways; and contribute to the extremely small literature on GA at term and educational outcomes in the United States.

Limitations include the urban sample and the oversampling of nonmarital births in the FFCWB study. Additionally, we relied on assessments of academic performance by teachers, which may be subjective but likely reflect students’ grades (because teachers assign the grades). Studies have found associations between teacher ratings and academic skills,^36,37  and studies from other countries using teacher ratings based on different questions have had similar findings to ours.^4,5,11  The questions we used to assess children’s academic performance have been used in US government surveys (including the Early Childhood Longitudinal Study-Kindergarten Cohort, which focuses on education) and a recent study of adverse childhood experiences and educational outcomes.³⁸  Moreover, patterns of estimates were similar when using relevant components of the validated CTRS-RSF. Loss to follow-up is another limitation, although supplemental analyses suggested minimal bias owing to attrition. Finally, causality cannot be firmly established in any observational study.

In light of limited US data currently available for studying educational outcomes associated with GA at term, new prospective studies involving larger, more diverse cohorts are needed. The ability to follow children and reevaluate them at various ages would provide valuable information on the extent of catch-up development and identify interventions that improve outcomes.

The findings from this study highlight the importance of longer gestation, even among term infants, and add to mounting evidence that early term elective deliveries should be avoided. Whereas many organizations suggest waiting until at least 39 weeks to deliver when possible, our findings add to a small group of studies suggesting that GA through 41 weeks is associated with improvements in some educational outcomes.

CI

confidence interval

CTRS-RSF

Conners Teacher Rating Scale–Revised Short Form

FFCWB

Fragile Families and Child Wellbeing

GA

gestational age

OR

odds ratio