OBJECTIVE: We examined the relationships among gestational age at birth, maternal characteristics, and standardized test performance in Georgia first-grade students.
METHODS: Live births to Georgia-resident mothers aged 11 to 53 years from 1998 through 2003 were deterministically linked with standardized test results for first-grade attendees of Georgia public schools from 2005 through 2009. Logistic models were used to estimate the odds of failure of the 3 components of the first-grade Criterion-Referenced Competency Test (CRCT).
RESULTS: The strongest risk factor for failure of each of the 3 components of the first-grade CRCT was level of maternal education. Child race/ethnicity and maternal age at birth were also associated with first-grade CRCT failure irrespective of the severity of preterm birth, but these factors were more important among children born moderately preterm than for those born on the margins of the prematurity distribution. Adjusting for maternal and child characteristics, there was an increased odds of failure of each component of the CRCT for children born late preterm versus term, including for math (adjusted odds ratio [aOR]: 1.17, 95% confidence interval [CI]: 1.13–1.22), reading (aOR: 1.13, 95% CI: 1.08–1.18), and English/language arts, for which there was an important interaction with being born small for gestational age (aOR: 1.17, 95% CI: 1.07–1.29).
CONCLUSIONS: Preterm birth and low maternal education increase children’s risk of failure of first-grade standardized tests. Promoting women’s academic achievement and reduce rates of preterm birth may be important to achieving gains in elementary school performance.
- AIC —
- Akaike information criterion
- aOR —
- adjusted odds ratio
- CI —
- confidence interval
- CRCT —
- Criterion-Referenced Competency Test
- ELA —
- English/language arts
- EPT —
- extremely preterm (20–27 weeks)
- LPT —
- late preterm (34–36 weeks)
- MPT —
- moderately preterm (28–33 weeks)
- SGA —
- small for gestational age
What’s Known on This Subject:
Extreme prematurity is a well-established cause of cognitive and motor impairment. There is some evidence that late prematurity and modifiable maternal attributes may negatively influence scholastic achievement, including standardized test performance.
What This Study Adds:
We found that preterm birth significantly increases risk of first-grade failure rate even when the birth is just a few weeks before term gestation. Low maternal education status compounds the effect of prematurity.
A decade after No Child Left Behind education law was enacted in an attempt to improve schools, particularly for the nation’s poor and minority children, questions remain about the inability of many schools to make adequate yearly progress. In 2010, ∼38% of US public schools did not make adequate yearly progress; in Washington, DC, the number not making adequate yearly progress was 91%.1 Consequently, the stakes of standardized testing are high for students, schools, and parents alike. Parents often seek guidance from their pediatrician concerning school readiness and developmental needs of their child.2 However, there is a lack of research to support pediatric practice in this area.
A child’s ability to succeed academically is influenced by innumerable social, economic, and biological factors. Abundant evidence shows that preterm birth, particularly <33 weeks’ gestation, can adversely affect a child’s neurocognitive development.3–6 Impaired neurologic development is associated with later poor academic outcomes including grade retention, poor test performance, and special education referral.7,8 The impact of preterm birth on cognitive development may be further exaggerated by parental characteristics where there is evidence of synergism among maternal socioeconomic status, preterm birth, and academic performance.9 To date, limited research has focused on the academic difficulties of children across the prematurity spectrum. Instead, most existing studies focus on clinical measures of neurologic deficits among relatively small samples of children born at extremely (EPT) to moderately (MPT) preterm gestational ages. Although a great deal is known about the acute neurologic outcomes associated with preterm birth before 33 weeks’ gestation, much less is known about the subclinical developmental and academic outcomes associated with late preterm birth (LPT; birth between 34 and 36-6/7 weeks).
The developmental risks associated with late preterm birth were once thought to be minimal. However, studies have demonstrated that even infants who are at the margins of prematurity suffer disproportionate rates of clinical neurocognitive problems.3,10–12 These LPT infants may be susceptible to early- and long-term academic failure.11,13–15 Yet few, if any, studies have examined the relative impact of LPT birth on standardized test performance, the most common measure of academic performance and a principal determinant of grade retention in public schools. In 2012, Lipkind et al examined the relationship between prematurity and standardized test performance in New York City.16 They found a linear association between gestational age and test scores from 32 to 39 weeks’ gestation.16,17 In our study, we hypothesized that prematurity would significantly increase the risk of first-grade test failure among our sample when controlling for maternal (eg, age and educational attainment) and child characteristics (eg, gender and birth weight for gestational age). We expected that maternal and child socioeconomic characteristics would moderate the influence of prematurity on test failure (Fig 1). Although socioeconomic conditions are frequently blamed for the “achievement gap” in educational testing,17,18 the role of prematurity in educational achievement has largely been ignored. Our study sheds more light on this area.
We constructed a retrospective cohort by deterministically linking the birth records of all singleton live births to Georgia-resident mothers aged 11 to 53 years from 1998 to 2003 to the Georgia State Department of Education standardized testing data for first-grade attendees of Georgia public schools from 2004 through 2009. Figure 2 delineates the construction of the cohort.
From the birth record, we ascertained the gestational age, race, and birth weight of the infants, as well as specific maternal characteristic including maternal race/ethnicity, age at delivery, and years of education. From the educational record, we ascertained the child’s performance on all 3 first-grade test components (mathematics, reading, English/language arts [ELA]) of the Criterion-Referenced Competency Test (CRCT). The CRCT is a set of standardized tests administered to students in kindergarten through grade 8 at public schools in Georgia. It is designed to assess students’ mastery of the content outlined in the Georgia Performance Standards. CRCT performance is a key indicator of student success. Students in grades 3, 5, and 8 are required to pass the CRCT to be promoted to the next grade. A detailed description of the CRCT properties (eg, reliability) can be found on the web site of the Georgia State Department of Education19 and in the extant literature.19–24
Once cleaned, the databases were combined into one dataset using deterministic record linkages based on unique ID’s created from the child’s name and date of birth and mother’s last name. Records that could not be matched were compared with matched records to assess systematic differences in maternal and child characteristics and were excluded from subsequent analysis.
We included births based on the following criteria ascertained from the birth certificate: (1) live-born singleton infants with recorded birth weight between 400 to 5000 g and gestational age between 24 to 43 weeks; (2) absence of congenital anomalies and/or chromosomal defects; (3) child race categorized as non-Hispanic white, non-Hispanic African American, or Hispanic of any race; and (4) births that could be linked to student education records.
Three binary outcomes were considered: children were categorized as failing versus passing on the math, reading, and ELA components of the first-grade CRCT. The primary exposure of interest was the child’s gestational age at birth categorized as EPT (20–27 weeks), MPT (28–33 weeks), LPT (34–36 weeks), term (37–41 weeks), and postterm (42–43 weeks). A secondary exposure of interest was a measure of fetal growth restriction as indicated by an infant being less than the 10th percentile of weight for gestational age using national growth curves25 (small for gestational age [SGA]) versus greater than or equal to the 10th percentile of weight for gestational age (appropriate or large for gestational age).
Several covariates were considered in modeling as potential confounders of the relationship between gestational age and school performance. Maternal age at birth was categorized as 10 to 14, 15 to 17, 18 to 19, 20 to 24, 25 to 29 (reference category), 30 to 34, 35 to 39, and ≥40 years. Maternal highest attained education at time of birth was categorized as no high school diploma, high school diploma (or general equivalency diploma), some college, and ≥4-year college degree (referent category). Child gender was included with female children as the referent group. Child race was included as non-Hispanic white (referent category), non-Hispanic African American, and Hispanic of any race. The bivariate distributions of covariates by gestational age categories were assessed by using frequencies and χ2 statistics.
Logistic regression was chosen as the primary analysis because of the binary outcome data (ie, “pass” vs “fail” on each first-grade CRCT subject test). Initial crude models regressed each outcome variable (failure of math, reading, and ELA components of the CRCT) separately on each exposure and covariate. Subsequent stepwise regression was conducted assessing change in exposure-outcome association as indication of potential confounding. All possible 2-way interactions between exposures and covariates were assessed, including possible 2-way interactions between gestational age and growth restriction. Model fit was assessed by using the change in −2-log-likelihood and Akaike information criterion [AIC] statistic. Because all educational levels were not theoretically attainable for all maternal ages (eg, teen moms are unlikely to have completed college), models were repeated with restriction to mothers ≥18 years and found to provide consistent results. All analyses were conducted by using SAS 9.2 (Carey, NC).
There were 628 115 eligible births from 1998 through 2003; this number represents the total number of singleton live births to Georgia-resident mothers who otherwise met the inclusion criteria. Of this cohort, 314 328 or 53% were successfully linked to their first-grade standardized test scores. There are 4 possible sources for loss to follow-up in this cohort. First, out-migration from Georgia between a child’s birth and entrance into first grade is expected. According to interstate migration data from the US Census and Internal Revenue Service, there is an average out-migration rate from Georgia of 2.5% to 3% per year.26,27 Therefore, an estimated 12.5% to 15% of the original cohort may have moved out of state before entering the first grade, −3% per year.26,27 A second source of loss is our reliance on standardized tests that are only mandated for students enrolled in Georgia public schools (traditional and charter). Based on educational and US Census data sources, 8% to 9% of Georgia children attend private schools, and an additional 2% to 5% are homeschooled or otherwise not enrolled in public schools.28–30 Therefore, 10% to 14% of the original cohort may not be represented in the public school CRCT database. On the basis of educational and US Census data sources, 8% to 9% of Georgia children attend private schools, and an additional 2% to 5% are homeschooled or otherwise not enrolled in public schools.28–30 Third, a small proportion of children born in Georgia may have died before entering the first grade. Finally, the deterministic linkage process used in relating individual birth records to first-grade educational records could fail to create true links because of changes or errors in the recording of identifying variables used in linkage.
Loss to follow-up in this setting is particularly concerning if it is differential with respect to key risk factors or outcomes. Although we cannot assess the educational outcomes for children who were not linked, we can compare the distribution of birth characteristics for children who were linked to the total birth cohort population. The linkage rate is relatively stable within strata of key variables, with some variation. Specifically, linkage was highest among children whose mothers had a high school degree or some college education (51%–60%) but was lower among children whose mothers had ≥4 years of college (44%). The linkage rate was 53% for children born term and 51% for those born preterm. The linkage rate was highest for children whose mothers were non-Hispanic African American (57%) but lower for non-Hispanic white (52%) and Hispanic (48%) mothers. Importantly, linkage rates were consistent across the birth years of interest.
The characteristics and distribution of subjects by maternal and child characteristics are described in Table 1. As shown, <1% of the 314 328 matched births were EPT, almost 2% were MPT, ∼8% were late preterm, the vast majority (∼86%) was term, and ∼4% were postterm.
The results of our estimation of the probability of failure of each component of the first-grade CRCT are given in 3 separate models. For all models, maternal age at birth, maternal highest attained education at time of birth, child gender, and child race/ethnicity were included as covariates.
Tables 2, 3, and 4 report results from crude and adjusted logistic regression models for the probability of a child failing math (Table 2), reading (Table 3), and ELA (Table 4). The results are similar for each outcome, and only results for math are detailed here. The first column presents the independent crude associations between gestational age and each covariate with probability of failing. The second column presents results from a multivariable adjusted model that includes all maternal and child covariates except SGA. Finally, the third column presents a fully adjusted model with all covariates and SGA. For math and reading (but not ELA), there were no significant interactions between SGA and preterm category. Overall there is a clear and significant dose-response association between gestational age category and the probability of failing the CRCT in first grade. Specifically risk for failure is significantly increased in each category including LPT and postterm compared with term (adjusted odds ratio [aOR] for LPT compared with term 1.17, 95% confidence interval [CI] 1.13–1.22). There is some attenuation of the association between gestational age category and outcome comparing the adjusted to the crude model, suggesting that maternal and child characteristics confound this association to some degree, but a robust independent association persists. Importantly, model fit as indicated by either the AIC or the pseudo-R2 improves in the fully adjusted (AIC = 217 280, pseudo-R2 = 11.8%) compared with the crude model (AIC = 236 812, pseudo-R2 = 0.44%).
In both the crude and adjusted logistic models, maternal age and education, child race/ethnicity, gender, and SGA status were significantly associated with failure of each component of the CRCT.
The relative influence of the model covariates on the probability of test failure varied considerably across the gestational age categories (Fig 3). The inclusion of the covariates in the logistic model reduced the probability of failure of each component of the CRCT among the moderately preterm group substantially more so than it did for the EPT and LPT groups. For example, the probability of failure of the ELA component was reduced by 9.8% and 8.5% among the EPT and LPT groups but by 14% among the moderately preterm group. The average reduction in the aOR was 13%, 16%, and 10% among the EPT, MPT, and LPT groups, respectively.
We found that being born “preterm” versus “term” increases a first-grade child’s risk of failure of each of the 3 components of the CRCT when controlling for maternal age at birth, maternal education, maternal race/ethnicity, child race/ethnicity, child’s gender, and year of birth. Increasing severity of preterm birth was associated with increasing likelihood of failure of all components of the CRCT in a significant dose-response pattern. Of note, even children born LPT (34–36 weeks’ gestation) experienced significantly increased risk of failure for each of the 3 components of the CRCT.
Of the factors considered in our model, the strongest risk factor for failure of each of the 3 components of the CRCT in first grade was maternal education with a consistent inverse exposure-response relationship between level of maternal education and risk of failure of all 3 components of the CRCT. Importantly, we also found that child race/ethnicity and maternal age at birth are associated with first-grade test performance irrespective of the severity of preterm birth. However, maternal (eg, age, education) and child social characteristics seem to play a greater role in test failure among children born moderately preterm than for those born on the margins of the prematurity distribution (EPT and LPT). There may be socioeconomic and sociocultural reasons for this finding. Extreme premature infants who survive to the first grade may have a more nurturing environment as a group than less-premature infants who are more likely to survive irrespective of their care. On the other side of prematurity, LPT infants have more iatrogenic prematurity owing to elective cesarean delivery, which may be more of an upper-class phenomenon. Thus, the infant gestational age groups do not start out on a level playing field with each other. The birth certificate variables are not nuanced enough to pick this up, particularly lacking information on income and the interaction of socioeconomic status with course of labor and delivery. Given our lack of detailed information about the trajectory of these children past their birth, we can only speculate about the potential reasons for these observations.
This study also provides evidence that preterm birth has long-term consequences even among those LPT infants who were formerly thought to be at negligible risk. This study adds to a growing body of literature suggesting that even LPT infants suffer disproportionate rates of clinical neurocognitive problems and are susceptible to early and long-term academic failure.3,10–15 Additional research is needed to identify the specific pathways through which late prematurity influences school performance. For example, to what extent do clinical interventions (eg, antenatal corticosteroid use) improve longer-term developmental outcomes by reducing the incidence of neonatal morbidity (eg, respiratory distress)? There is some precedent in the literature in this area. Patrianakos-Hoobler et al found that preterm survivors who experienced respiratory distress also experienced higher prevalence of neurologic impairment even 2 years after birth, with 23% of their sample labeled as “delayed” at 2 years of age.8 They found that developmental delays at age 2 years were highly predictive of the lack of school readiness at first grade, suggesting that neonatal and postneonatal morbidity may directly affect school outcomes such as test performance. Consequently, it is reasonable to suggest that reductions is neonatal morbidity might be associated with even longer-term developmental outcomes.
Our findings also appear to be similar to that of Lipkind et al.16 Their study examined third-grade math and English test scores in a single city, but the magnitude of the relationship between test performance and prematurity was similar to that observed in our study. Premature subjects in their study scored from 4% to 10% lower on the standardized than did their full-term counterparts.16 Unlike the Lipkind study, we examined first-grade test performance. Test failure at this grade level is inherently low. We expect that we will be able to explain even more variation in test failure when we are able to analyze our second- and third-grade data. Approximately 19% of third-grade students in Georgia failed the math CRCT exam in 2011 as opposed to ∼13% of first graders.20 In the sixth grade, failure rates are ∼24%. Given our findings, it is reasonable to believe that prematurity could have an even longer and more substantial impact on school achievement as a child progresses through the grade levels. This is important because stakes of state-level assessments (eg, grade retention) increase as a child moves up each grade level, as do failure rates.
- Accepted December 5, 2012.
- Address correspondence to Bryan L. Williams, PhD, Nell Hodgson Woodruff School of Nursing, Rollins School of Public Health, Emory University, 1520 Clifton Rd NE, Atlanta, GA 30322. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: No external funding.
- ↵Usher A. Update With 2009–10 Data and Five-Year Trends: How Many Schools Have Not Made Adequate Yearly Progress? Washington, DC: Center on Education Policy; 2011
- Combs-Orme T,
- Holden Nixon B,
- Herrod HG
- Gisselmann M,
- Koupil I,
- De Stavola BL
- Luu TM,
- Ment LR,
- Schneider KC,
- Katz KH,
- Allan WC,
- Vohr BR
- McGowan JE,
- Alderdice FA,
- Holmes VA,
- Johnston L
- ↵Lipkind HS, Slopen ME, Pfeiffer MR, McVeigh KH. School-age outcomes of late preterm infants in New York City. Am J Obstet Gynecol. 2012;206(3):222.e1–6
- ↵Yaffe D, Coley RJ, Pliskin R; Educational Testing Service PIC. Addressing Achievement Gaps: Educational Testing in America: State Assessments, Achievement Gaps, National Policy and Innovations (ETS Policy Notes, Volume 17, no. 1). Princeton, NJ: Educational Testing Service; 2009
- ↵Georgia Department of Education. Curriculum, instruction, and assessment. 2012. Available at: www.gadoe.org/ci_testing.aspx?PageReq=CI_TESTING_CRCT. Accessed January 6, 2012
- ↵Georgia Department of Education. 2010–2011 Report Card. 2012. Available at: www.doe.k12.ga.us/ReportingFW.aspx?PageReq=102&StateId=ALL&T=1&FY=2011. Accessed January 6, 2012
- Randall J, Cheong YF, Engelhard G. Using explanatory item response theory modeling to investigate context effects of differential item functioning for students with disabilities. Educ Psychol Meas. 2011;71(1):129–147
- Sanford EE
- Randall J, Engelhard G. Performance of students with and without disabilities under modified conditions. J Special Educ. 2010;44(2):79–93
- ↵Oken E, Kleinman KP, Rich-Edwards J, Gillman MW. A nearly continuous measure of birth weight for gestational age using a United States national reference. BMC Pediatr. 2003;3:6
- ↵Koerber K. Domestic migration flows for states from the 2005 ACS. Presented at the Population Association of America 2007 Annual Meeting, New York, NY
- ↵IRS. SOI Tax Stats - State-to-State Migration Database Files. Available at: www.irs.gov/uac/SOI-Tax-Stats—Free-Migration-Data-Downloads
- ↵US Census Bureau. Census 2000 Summary File 3. 2000. Available at http://www.census.gov/census2000/sumfile3.html. Accessed January 14, 2013
- US Census Bureau. American Community Survey, 2005 Summary Tables. 2005. Available at: www.census.gov/acs/www/data_documentation/data_main/.
- ↵Council TGH. Public education (PreK–12). The New Georgia Encyclopedia. 2009. Georgia Humanities Council, in partnership with the University of Georgia Press, the University System of Georgia/GALILEO, and the Office of the Governor. Available at: http://www.georgiaencyclopedia.org. Accessed September 1, 2012
- Copyright © 2013 by the American Academy of Pediatrics