September 2014, VOLUME134 /ISSUE 3

Mild Prematurity, Proximal Social Processes, and Development

  1. Hilary K. Brown, MSca,
  2. Kathy Nixon Speechley, PhDa,b,
  3. Jennifer Macnab, PhDa,
  4. Renato Natale, MDc, and
  5. M. Karen Campbell, PhDa,b,c
  1. Departments of aEpidemiology and Biostatistics,
  2. bPaediatrics, and
  3. cObstetrics and Gynaecology, The University of Western Ontario, London, Ontario, Canada


OBJECTIVE: To elucidate the role of gestational age in determining the risk of poor developmental outcomes among children born late preterm (34–36 weeks) and early term (37–38 weeks) versus full term (39–41 weeks) by examining the contribution of gestational age to these outcomes in the context of proximal social processes.

METHODS: This was an analysis of the Canadian National Longitudinal Survey of Children and Youth. Developmental outcomes were examined at 2 to 3 (N = 15 099) and 4 to 5 years (N = 12 302). The sample included singletons, delivered at 34 to 41 weeks, whose respondents were their biological mothers. Multivariable modified Poisson regression was used to directly estimate adjusted relative risks (aRRs). We assessed the role of parenting by using moderation analyses.

RESULTS: In unadjusted analyses, children born late preterm appeared to have greater risk for developmental delay (relative risk = 1.26; 95% confidence interval [CI], 1.01 to 1.56) versus full term. In adjusted analyses, results were nonsignificant at 2 to 3 years (late preterm aRR = 1.13; 95% CI, 0.90 to 1.42; early term aRR = 1.11; 95% CI, 0.96 to 1.27) and 4 to 5 years (late preterm aRR = 1.06; 95% CI, 0.79 to 1.43; early term aRR = 1.03; 95% CI, 0.85 to 1.25). Parenting did not modify the effect of gestational age but was a strong predictor of poor developmental outcomes.

CONCLUSIONS: Our findings show that, closer to full term, social factors (not gestational age) may be the most important influences on development.

  • preterm birth
  • child development
  • parenting
  • Abbreviations:
    adjusted relative risk
    confidence interval
    National Longitudinal Survey of Children and Youth
    Peabody Picture Vocabulary Test–Revised
    relative excess risk due to interaction
    relative risk
  • What’s Known on This Subject:

    Previous studies examining developmental outcomes associated with late preterm and early term birth have shown mixed results. Many of these studies did not fully take into account the role of the social environment in child development.

    What This Study Adds:

    Social factors, not late preterm or early term birth, were the strongest predictors of poor developmental outcomes at 2 to 3 and 4 to 5 years. The influence of mild prematurity may lose strength beyond the neonatal period.

    Preterm birth is defined as delivery before 37 weeks’ gestation. Although developmental risks associated with very preterm birth (<34 weeks) are well established,1 children born closer to term were traditionally assumed to be low risk.2 Recent research suggests that children born late preterm (34–36 weeks) and early term (37–38 weeks) may be at greater risk for poor developmental outcomes than full-term peers (39–41 weeks), prompting some experts to recommend expanding the definition of preterm to include births before 39 weeks.3 However, it is unclear to what extent these outcomes are associated with being born early (mild physiologic immaturity) or with factors associated with being born early (social risk factors).

    Studies have shown that, compared with children born at term, children born late preterm are at risk for developmental delay4 and low IQ.5 They perform worse on academic tests (receptive vocabulary, reading, mathematics),68 are more likely to have special education needs,7,9 and are at risk for cerebral palsy and attention deficits.10,11 A handful of studies have shown that children born early term may be at risk for low IQ12 and poor academic performance.6 On the other hand, several studies failed to find significant risks for poor developmental outcomes for late preterm13,14 and early term15 births.

    Evidence of rapid fetal brain development at 34 to 40 weeks’ gestation16 supports the argument that mild physiologic immaturity explains these risks. However, children do not develop in isolation.17 A large body of literature supports the importance of social factors, particularly parenting, in child development.17 Most late preterm and early term studies have downplayed the role of social factors.7,9 However, the intricacies of the social environment must be taken into account to delineate the effects of late preterm and early term birth on development.

    Theories of child development clarify the roles of these social factors. Bioecological theory18 distinguishes between proximal social processes and social context. Proximal social processes refers to ongoing child–environment interactions; in the early years, the most important is parenting (ie, interactions, effectiveness, and consistency).19,20 Social context refers to settings in which the child develops; in the early years, the most important is the home, described in terms of family structure, resources, and functioning. The concept of double jeopardy enhances bioecological theory by capturing the idea that children with both biological and social risk factors are at even greater risk for poor outcomes than those with only biological or social risks.21,22 (Escalona21 introduced this idea by showing greater cognitive decline among low birth weight infants in low versus high socioeconomic status households.) Parenting may be a more relevant effect measure modifier because parenting most directly affects child development23 and is modifiable.


    Our aim was to elucidate the role that gestational age (mild physiologic immaturity) plays in determining risks of poor developmental outcomes among children born late preterm and early term by examining the contribution of gestational age to these outcomes in the context of proximal social processes. The research questions, depicted in Fig 1, were as follows:

    FIGURE 1

    Study conceptual model.

    1. How does the risk of poor developmental outcomes among children born late preterm and early term compare with that of children born full term, after biological and social factors are controlled for?

    2. Do proximal social processes modify the effect of gestational age on poor developmental outcomes?


    Study Design and Setting

    This was a secondary analysis of the National Longitudinal Survey of Children and Youth (NLSCY), which followed a sample of Canadian children from 1994 and 1995 (Cycle 1) to 2008 and 2009 (Cycle 8). Data access was obtained through the Social Sciences and Humanities Research Council; ethics approval was not needed because respondents were not identifiable. For this study, 0- to 1-year-olds in Early Child Development cohorts of Cycles 2 through 6 were pooled and followed for 2 subsequent cycles (N = 15 099 at 2–3 years; N = 12 302 at 4–5 years) (Fig 2).


    The NLSCY excludes children living in institutions or on reserves and whose parents are members of the armed forces. Additional criteria used to define our study sample were born at 34 to 41 weeks1,24 and singleton gestation.25,26 Children were excluded if their respondent was not the biological mother at all cycles. (Questions about the perinatal period were asked only of biological mothers at child age 0 to 1 years to maximize validity of responses.) At a given cycle, <3% of children had a respondent who was not the biological mother.

    Data Sources

    The NLSCY sampled children from all 10 provinces through the Labor Forces Survey, which has a stratified, multistage design that uses probability sampling. Primary strata were defined by urbanicity; secondary strata were defined by income and population density. Clusters of dwellings were identified within strata, and dwellings were systematically sampled from clusters. For Early Child Development cohorts, 1 child per household was selected (exception: twins, Cycles 3 and 4). Data collection was by computer-assisted telephone and personal interviewing.


    Gestational age was determined by maternal report (at child age 0–1 years) of the days or weeks before or after the due date the child was born. Studies show accurate maternal recall of gestational age when questions are in relation to due date.27 Nevertheless, to maximize accuracy, we excluded children with implausible birth weight for gestational age (>4 SDs, for boys and girls separately).28,29 Gestational age was based on completed weeks (36 weeks, 6 days = 36 completed weeks),2 and children were classified as late preterm (34–36 weeks), early term (37–38 weeks), or full term (39–41 weeks).2

    Developmental outcomes were described in terms of developmental delay and receptive vocabulary delay. Developmental delay was measured at 2 to 3 years using the Motor and Social Development Scale, which was developed by the US National Center for Health Statistics.30 The parent responds to 15 yes/no task performance questions (which vary depending on the child’s age), and the “yes” responses are summed. Scores were standardized by 1-month age groups (M = 100, SD = 15), and children scoring ≥1 SD below the age-standardized mean were classified as having a delay.31 The Motor and Social Development Scale has good construct validity; scores are correlated with the Bayley Scales of Infant Development (3rd edition),32 and high scores are predictive of fewer behavior problems.30,32 Receptive vocabulary delay was measured at 4 to 5 years using the Peabody Picture Vocabulary Test–Revised (PPVT-R). In the NLSCY, all PPVT-R assessments are conducted in person with a trained tester who presents a series of pictures and states a word for which the child must choose the correct picture. There are 175 items of increasing difficulty.30 The number of correct responses is computed, and an age-standardized score is based on 1-month age groups. Children scoring ≥1 SD below the age-standardized mean were classified as having a delay. The PPVT-R performs well (split-half reliability coefficient = 0.80).33

    Social factors were classified as proximal social processes (parenting) or social context variables. Parenting was measured using the Parenting Scale. This adaptation of the Parenting Practices Scale34 assesses patterns of parent–child interactions. Subscale Cronbach’s αs are 0.68 (parenting interactions [5 items; eg, “How often do you and this child talk or play with each other, focusing attention on each other for five minutes or more, just for fun?”]), 0.60–0.63 (parenting effectiveness [7 items; eg, “How often do you think that the kind of punishment you give this child depends on your mood?”]), 0.65–0.72 (parenting consistency [5 items; eg, “When you give this child a command or order to do something, what proportion of the time do you make sure he/she has done it?”]), and 0.52–0.56 (rational parenting [4 items]; excluded because of poor performance).30 Questions vary depending on the child’s age. The Parenting Scale shows good construct validity; it is correlated with family structure and socioeconomic status.30 For each subscale, the standardized average across periods of data collection was taken to reflect the “average exposure” of the child; the “worst” 10% was considered to be poor parenting. (Averaging measures has the added benefit of producing more reliable estimates.)

    Several variables were assessed for their roles as confounders: perinatal variables (smoking during pregnancy, alcohol use during pregnancy, placental ischemia and other hypoxia [maternal hypertension, small for gestational age], other biological determinants of preterm birth [maternal diabetes, other medical conditions during pregnancy], delivery mode); social context as described in terms of family structure (maternal partnership status, number of siblings), family resources (family income adequacy, maternal education, maternal age, maternal health, maternal mental health), and family functioning; and other covariates (child gender). Neonatal special care and breastfeeding were not included in multivariable analyses because they are on the causal pathway.


    SAS 9.335 (SAS Institute, Inc, Cary, NC) was used for all analyses. Descriptive analyses included frequencies and percentages. Univariable modified Poisson regression (with PROC GENMOD)36 was used to assess unadjusted associations between covariates and outcomes before multivariable analyses.

    To address the first research question, we directly estimated adjusted relative risks (aRRs) by using multivariable modified Poisson regression.36 Parsimonious models were built with blockwise entry of variables according to conceptual categories: perinatal variables, gestational age, family structure, family resources, family functioning, proximal social processes, and other covariates. A P value of <.20 was used to retain covariates at each step, and 95% confidence intervals (CIs) were used in final models to reflect clinical significance.37

    To address the second research question, we explored additive interaction between gestational age and parenting subscales by calculating the relative excess risk due to interaction (RERI = relative risk [RR]11 – RR10 – RR01 + 1).38 (Values were obtained from main effects and interaction terms added to the final models for the first research question.) We calculated CIs by using the method of variance estimates recovery technique.38 (Note that for RERIs, 0 indicates no excess risk.)

    To account for the NLSCY’s complex sampling design, we used longitudinal weights, standardized to maintain the original sample size. Because statistical packages with bootstrapping capabilities have not been developed for modified Poisson regression, we took the sampling design into account by controlling for province and urban or rural status. Because the dataset included 5 pooled cycles, a “time” variable was entered into the models to control for NLSCY cycle of entry.


    Of 18 642 eligible children, 0.7% were excluded because of implausible birth weight for gestational age. The sample size was N = 15 099 at 2 to 3 years (18.5% lost to follow-up or excluded) and N = 12 302 at 4 to 5 years (33.6% lost to follow-up or excluded) (Fig 2). Table 1 summarizes descriptive statistics at both ages.

    TABLE 1

    Weighted Sample Characteristics for Developmental Delay Analysis (2–3 y of Age; N = 15 099) and Receptive Vocabulary Delay Analysis (4–5 y of Age; N = 12 203)

    Research Question 1

    The rate of developmental delay in 2- to 3-year-olds was 14.2% (16.7% in late preterm, 14.3% in early term, and 13.9% in full term). In unadjusted analyses, children born late preterm appeared to have greater risk for developmental delay (RR = 1.26; 95% CI, 1.01 to 1.56). However, after we controlled for confounders, children born late preterm (aRR = 1.13; 95% CI, 0.90 to 1.42) and early term (aRR = 1.11; 95% CI, 0.96 to 1.27) were not at greater risk for developmental delay than those born full term (Table 2).

    TABLE 2

    Weighted Univariable and Multivariable Statistics for Developmental Delay (2–3 y of Age)

    The rate of receptive vocabulary delay in 4- to 5-year-olds was 13.0% (13.1% in late preterm, 13.9% in early term, and 12.7% in full term). After we controlled for confounders, children born late preterm (aRR = 1.06; 95% CI, 0.79 to 1.43) and early term (aRR = 1.03; 95% CI, 0.85 to 1.25) were not at greater risk for receptive vocabulary delay than those born full term (Table 3).

    TABLE 3

    Weighted Univariable and Multivariable Statistics for Receptive Vocabulary Delay (4–5 y of Age)

    Research Question 2

    Additive interactions between gestational age and parenting subscales were tested (Table 4). RERIs show the excess risk that is attributable to interaction. For both outcomes, because all RERIs were close to 0, there was no evidence of excess risk due to interaction for any of the parenting subscales for either late preterm birth or early term birth.

    TABLE 4

    Additive Interactions Between Gestational Age and Parenting

    Sensitivity Analyses

    Results for the first research question were unchanged when neonatal special care and breastfeeding were added to the multivariable models (data not shown).

    It is possible that null findings for the second research question could be explained by lack of power caused by limiting “poor parenting” to 10% of scores. We reran analyses using 25% as a cutoff. Results remained unchanged, with RERIs near 0 (data not shown).

    To test the validity of our gestational age variable, we examined the association between late preterm and early term birth and poor neonatal outcomes39,40 in our data. Compared with children born full term, there was greater risk for neonatal special care for children born late preterm (aRR = 3.71; 95% CI, 3.15 to 4.38) and elevated but not statistically significant risk for children born early term (aRR = 1.16; 95% CI, 0.98 to 1.37) (Table 5).

    TABLE 5

    Association Between Gestational Age and Neonatal Special Care in 2- to 3-y-Old Sample (N = 15 099)


    There was elevated risk for developmental delay among children born late preterm compared with those born full term (16.7% vs 13.3%). Although this unadjusted association is an important finding, it was no longer statistically significant in adjusted analyses. Moreover, there was no evidence of elevated risk for developmental delay among children born early term or for receptive vocabulary delay among children born late preterm or early term. Although these findings contrast with those of some previous studies, several others also found no association.1315

    Despite a null adjusted main effect, there could be significant risks associated with mild prematurity in families with proximal social risks (poor parenting).21,22 This was not the case in our study, in contrast with previous research suggesting an interaction between mild prematurity and social factors.4143 However, the main effects for parenting showed a strong association with developmental outcomes, even after we controlled for social context variables. This finding is consistent with previous literature showing that a lack of positive involvement, punitive discipline because of parenting ineffectiveness, and inconsistency are associated with poor developmental outcomes.19,20

    Consistent with previous research,39,40 we found a strong association between late preterm birth and neonatal special care (Table 5). This finding gives us confidence in the validity of our gestational age variable. It is possible that the impact of mild prematurity loses strength beyond the neonatal period, particularly once social factors are taken into account. The relative importance in childhood of gestational age and proximal social processes such as parenting is reflected in the size of their RRs, which were much larger for parenting than for gestational age. Population attributable fractions for parenting (interactions, 3.99%, 3.03%; effectiveness, 1.27%, 1.18%; consistency, 2.89%, 4.53%) were also slightly larger than those for gestational age (late preterm, 0.93%, 0.43%; early term, 3.06%, 0.85%) (for developmental delay and receptive vocabulary delay, respectively). (However, all population attributable fractions for these main variables were small, showing that many other factors explain the complexity of child development.)

    Strengths and Limitations

    A major strength of our study was the extensive coverage of information on factors that influence child development. In contrast with previous research,7,9 we were able to account for parenting and other important social risk factors. Moreover, the longitudinal nature of the data allowed us to capture aspects of the social environment at >1 time point (eg, changes in family income adequacy and maternal partnership status).

    A limitation was that NLSCY data were mostly by maternal self-report. Although we took steps to maximize the validity of the gestational age variable, it is possible that null findings are partially due to misclassification. Other perinatal variables may have been overreported or underreported,44 but this is expected to be minimal because all perinatal questions were asked soon after birth. It is possible that maternal report of developmental delay was distorted by the mother’s health or socioeconomic characteristics.45 However, parental concerns are considered to be a valuable component of clinical assessments of development.46

    Bias could have been introduced if we falsely considered variables to be mediators or confounders. We did not exclude or control for sensory impairments, disabilities, or chronic health conditions because this could result in adjusting away part of the association between gestational age and developmental outcomes (if such conditions are outcomes of mild prematurity). However, because we wanted to isolate the effect of mild prematurity, we controlled for biological determinants of late preterm and early term birth (which could harm the fetus)47 and social factors (which are associated with preterm birth48 and child development18). Although there may be a reciprocal relationship between child and parental behavior,49 we considered parenting to be a confounder because parenting skills are a proximal representation of the social environment.23 This conceptualization is consistent with previous research20,50; moreover, the relationship between gestational age and developmental outcomes was nonsignificant even before parenting was entered into the model. We were unable to exclude children with congenital anomalies, because no question in the NLSCY asks about such conditions. However, congenital anomalies, some of which are not survived past infancy, account for <2% of births.51

    Our study may be limited by issues related to generalizability. There was loss to follow-up; nonrespondents were more likely to have social risk factors, including single-parent families, income inadequacy, and low maternal education. Data collection began before 2000; the incidence of late preterm birth has increased in recent years,52 and social conditions have shifted over time. Although it is possible that frequencies of factors under study are not entirely generalizable, our goal was causal inference, not prevalence estimation. According to Rothman and colleagues,53 threats to external validity do not affect internal validity; therefore, associations remain valid.

    Recommendations for Future Research

    Future research could build on this study by performing a similar analysis (with full consideration of social factors) in a sample for which there is a measure of gestational age based on first-trimester ultrasound.54 There appears to be a dichotomy between clinical samples with gold standard measurement of gestational age (but poor attention to social factors)11,12 and population-based surveys with adequate representation of social factors (but maternal report of gestational age).4 Although it is difficult to measure all variables with precision, there is a need for studies in samples that can adequately address both biological and social factors.


    Although there was slightly elevated unadjusted risk of developmental delay associated with late preterm birth, findings from multivariable models suggest that social factors, not gestational age, are the most important predictors of outcomes beyond the neonatal period in births close to full term.


      • Accepted June 2, 2014.
    • Address correspondence to M. Karen Campbell, Department of Epidemiology and Biostatistics, The University of Western Ontario, Kresge Building Room K201, London, Ontario, Canada, N6A 5C1. E-mail: Karen.Campbell{at}
    • Ms Brown was primarily responsible for the design and analysis of the study and the writing of the manuscript, under the supervision of Dr Campbell; Drs Speechley, Macnab, Natale, and Campbell provided guidance on the conceptualization and design of the study and provided regular feedback on drafts of the manuscript; and all authors approved the final manuscript as submitted.

    • FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.

    • FUNDING: Supported by a Canadian Institutes of Health Research Doctoral Research Award (GSD 104470) to the first author.

    • POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.