It is well established that children born small for gestational age or preterm (<37 weeks) have lower cognitive ability than their appropriate-for-gestational-age or term counterparts (4–13). It is therefore often assumed that children born at term are homogeneous with respect to long-term cognitive development. Recently, studies have reported a positive association between birth weight and cognitive ability even among children with birth weight within the normal range (14–16), although residual confounding by unmeasured family characteristics may have biased the results (17, 18). Although there are many studies on cognitive ability variations by birth weight, studies to examine the variation by gestational age are very limited. To our knowledge, there are only 2 such studies, but 1 did not adjust for important confounders including socioeconomic position (19) and the other was based on men only (20). Moreover, neither of these studies focuses on variations in cognitive ability by each week of gestation among children born at or beyond 37 completed weeks of gestation.

In this study, we took advantage of a 6.5-year follow-up of Belarusian children participating in a large, randomized trial known as the Promotion of Breastfeeding Intervention Trial (PROBIT) to examine whether child cognitive ability is associated with gestational duration and birth weight for gestational age among healthy, normally grown term births.

Table 1 presents the characteristics of children by gestational weeks. Of the total sample of 13,824 children, 3.4% were born at 37 weeks, 15.2% at 38 weeks, 30.3% at 39 weeks, 43.1% at 40 weeks, 6.7% at 41 weeks, 1.2% at 42 weeks, and 0.1% at 43 weeks.

Figure 1 shows the mean full-scale IQ scores by gestational age. The mean IQ increased with each completed week of gestation from 37 to 40 weeks and decreased among those born postterm. As no statistically significant differences in full-scale IQ were observed among children born at 39, 40, and 41 weeks, we combined these 3 gestational ages as a reference group. As no differences were observed in the magnitude of associations by sex (all P_interaction > 0.3), we present sex-adjusted results.

Table 2 presents the crude and adjusted mean IQ differences by gestational week. Compared with the score for children born at 39–41 weeks, the full-scale IQ score was 2.6 points (95% CI: −3.7, −1.4) lower in those born at 37 weeks and 0.5 points (95% CI: −1.1, −0.01) lower at 38 weeks. A graded relation was also found among children born postterm compared with those born at 39–41 weeks: 1.4 points (95% CI: −3.5, 0.6) lower at 42 weeks and 5.8 points (95% CI: −14.0, 2.5) lower at 43 weeks. These associations were attenuated with adjustment for potential confounding factors, mainly owing to differences in maternal age and family socioeconomic position. However, the overall pattern remained unchanged. After controlling for all potential confounding factors, we found that the full-scale IQ was lower by 1.7 points (95% CI: −2.7, −0.7) in children born at 37 weeks and by 0.4 points (95% CI: −1.2, 0.2) at 38 weeks compared with those born at 39–41 weeks. In the fully adjusted model, the IQ was lower by 0.5 points (95% CI: −2.6, 1.6) in children born at 42 weeks and by 6.0 points (95% CI: −15.1, 3.1) in those born at 43 weeks.

For fetal growth (birth weight for gestational age), 9.1% of children were born small for gestational age, and 7.6% were born large for gestational age. The full-scale IQ increased steadily with fetal growth. The cluster-adjusted mean full-scale IQ was 105.0 among children born small for gestational age, 106.4 among those in the 10th–50th percentile, 107.4 at the >50th–90th percentile, and 107.8 among those born large for gestational age. Compared with children who were born large for gestational age, those born small for gestational age had the lowest score of −2.7 (95% CI: −3.5, −1.9), followed by −1.6 (95% CI: −2.5, −0.7) for those at the 10th–50th percentile and −0.8 (95% CI: −1.7, 0.1) for those at the >50th–90th percentile after controlling for gestational age and maternal and family characteristics (Table 3).

The interaction between gestational age and fetal growth was not statistically significant (P = 0.13). In addition, the mean differences in full-scale IQ associated with gestational age and fetal growth were essentially unchanged with further adjustment for breastfeeding status.

We also examined whether length and head circumference at birth (standardized for sex and gestational age), as other measures of fetal growth, were associated with full-scale IQ. Although both measures were positively associated with full-scale IQ, the associations were small and disappeared after adjustment for birth weight. The mean differences in full-scale IQ per standard deviation were 0.34 (95% CI: −0.05, 0.72) for birth length and 0.35 (95% CI: −0.06, 0.76) for birth head circumference, but associations with birth weight for gestational age were unchanged after adjustment for birth length or head circumference.

The patterns of association found for verbal and performance IQ were consistent with that observed for full-scale IQ. For verbal IQ, compared with children born at 39–41 weeks of gestation, children born at 37 weeks scored 1.2 points (95% CI: −2.1, 0.3) lower, 0.5 points (95% CI: −1.0, 0.1) lower at 38 weeks, 1.0 points (95% CI: −2.7, 0.7) lower at 42 weeks, and 4.5 points (95% CI: −19.2, 10.2) lower at 43 weeks after adjustment for potential confounders. For performance IQ, the adjusted mean differences were −1.8 (95% CI: −3.1, −0.5) points for 37 weeks, −0.3 (95% CI: −1.0, 0.4) points at 38 weeks, 0.4 (95% CI: −2.1, 2.9) points at 42 weeks, and −6.0 (95% CI: −10.2, −1.7) points at 43 weeks. The adjusted mean differences in verbal IQ from children born large for gestational age were −2.4 (95% CI: −3.5, −1.2) among those born small for gestational age, −1.7 (95% CI: −2.6, −0.8) among the 10th–50th percentile, and −0.8 (95% CI: −1.7, 0.1) among the >50th–90th percentile. For performance IQ, the corresponding values were −2.4 (95% CI: −3.5, −1.3), −1.1 (95% CI: −2.0, −0.2), and −0.6 (95% CI: −1.5, 0.3), respectively.

Although academic performance ratings tended to be lower among children born at early term (37–38 weeks) and postterm (42–43 weeks) compared with those born at 39–41 weeks, the differences were small and statistically nonsignificant. For example, the adjusted mean rating differences in math were −0.03 (95% CI: −0.11, 0.05) at 37 weeks, −0.02 (95% CI: −0.06, 0.02) at 38 weeks, 0 (95% CI: −0.13, 0.12) at 42 weeks, and 0.06 (95% CI: −0.53, 0.65) at 43 weeks. The adjusted mean differences in writing were −0.01 (95% CI: −0.09, 0.06) at 37 weeks, −0.05 (95% CI: −0.09, −0.01) at 38 weeks, −0.01 (95% CI: −0.11, 0.13) at 42 weeks, and −0.23 (95% CI: −0.80, 0.34) at 43 weeks. The adjusted mean differences in math by fetal growth were −0.14 (95% CI: −0.21, −0.07) among children born small for gestational age, −0.08 (95% CI: −0.14, −0.02) among the 10th–50th percentile, and −0.02 (95% CI: −0.07, 0.04) among those in the >50th–90th percentile. The corresponding values for writing were −0.20 (95% CI: −0.27, −0.13), −0.10 (95% CI: −0.15, −0.03), and −0.03 (95% CI: −0.09, 0.03), respectively.

Our results are based on a large sample of healthy term births, permitting us to estimate gestational week-specific associations with cognitive ability. The effects estimated in our study are probably generalizable to other developed country settings, since Belarus resembles Western developed countries with respect to basic health services, sanitary conditions, and readily accessible health-care facilities, even in rural areas. Belarus has the lowest infant and child mortality rates among the Commonwealth of Independent States, with an infant mortality rate of 7 per 1,000 livebirths in 2002.

Although we adjusted for a wide range of important maternal and family characteristics, residual confounding by unmeasured family characteristics cannot be excluded. For example, child cognitive development is strongly determined by maternal cognitive ability (27), which was not measured in our study. Some studies using family-based analyses comparing siblings within and between families suggest residual confounding in the well-known positive association between birth weight and cognitive ability in children (17, 18), although results are not consistent across such studies (16). As family socioeconomic position partly confounded the association observed in our and other studies (15, 17, 18), differences in unmeasured indicators of family socioeconomic position may further confound the observed association. However, Belarus is one of the countries with the lowest degree of income inequality, as indicated by their Gini index of 29.7 compared with 40.8 in the United States (one of the highest inequality countries) and 25.0 in Sweden (one of the lowest inequality countries). Thus, the potential residual confounding by socioeconomic factors may not be as important as in other settings.

It should also be noted that the association between gestational age and IQ observed in the present study could be confounded by the underlying causes of earlier birth or suboptimal fetal growth, causes that might themselves lead to suboptimal brain development. Nonetheless, after restricting our analysis to children without delivery or postpartum maternal or neonatal complications identified by maternity hospital records, we found nearly identical results as in our main analysis, although pregnancy complications such as gestational diabetes were not recorded.

It is worth noting the clustering of mean IQ scores across polyclinics, as denoted by the high intraclass correlation coefficient of 0.31 (28). Since IQ was measured by a single pediatrician for all children at 24 of the 31 polyclinics (2 pediatricians shared the work in 7 of the busiest polyclinics), the high intraclass correlation coefficient suggests that, despite our efforts to standardize measurements across pediatricians, IQ assessment varied across pediatricians, probably due to differences in strictness or leniency in scoring or timing of responses, rather than true geographic differences in IQ (29). However, neither gestational age nor birth weight was measured by the pediatricians, neither substantially differed by polyclinic, and neither covaried with IQ across polyclinics. Thus, the issue here is not confounding but loss of statistical power and consequently wide confidence intervals (29). We also note that the mean IQ in our study is higher than the English/American norms for the test (i.e., mean of 100). This probably reflects our inclusion criteria, which restricted the study sample to healthy infants born at term with birth weights of ≥2,500 g. The high mean IQ scores might also reflect the study pediatricians’ permissive scoring of children's responses or time allotted to complete tasks or answer questions. High literacy rates in Belarusian parents, however, might result in truly higher IQ in their children. Finally, the Russian version of the WASI used in our study has not been formally validated by comparison with the Russian version of the full Wechsler scales or other instruments. Despite the lack of such validation, however, the WASI scores we obtained were strongly associated with parental socioeconomic factors and by parental education in particular. The scores were also positively correlated with children's academic performances (correlation coefficients ~ 0.3). As shown in Table 3, other factors considered in our study were also associated with IQ scores in the expected directions.

The positive association that we observed between fetal growth and cognitive ability in children born at term is consistent with results from other studies (14–16). However, studies among full-term births (30, 31) are less common. The present study adds to the existing literature suggesting a positive association between fetal growth and IQ across the entire distribution of fetal growth, even among children born at term and after controlling for important confounding factors.

We have recently examined variations in cognitive ability by fetal growth comparing siblings with nonsiblings and found that the positive association disappeared in within-sibling analyses (18). As noted earlier, the positive association observed in the present study could thus be due to residual confounding by unmeasured family characteristics. Alternatively, the disappearing effect of fetal growth in our sibling-based study could be explained by the potential biases due to relatively poorly measured birth weight and gestational age, which were based solely on maternal recall within a 2-year time window. Variations in measurement of cognitive ability might also explain inconsistent results.

Consistent with our findings, lower cognitive ability among children born postterm (≥42 weeks of gestation) has been reported in a few previous studies (19, 20). Our study further demonstrates that cognitive ability increases with each additional week of gestation between 37 and 41 weeks and in both sexes. This pattern of association across gestational weeks among term births is consistent with recent reports of increased risks of infant mortality and morbidity in late preterm (34–36 weeks) births (32, 33) and even at 37 and 38 weeks of gestation (3). Chyi et al. (34) have recently shown that healthy, late-preterm children without significant neonatal complications have poorer reading and math scores compared with those born at term. Our results suggest that mild cognitive deficit can occur even among infants born at early term compared with those born at 39–41 weeks. The gradient observed suggests that each additional week of gestation is beneficial for brain development, at least up to 39 weeks (35).