NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) and Institute of Medicine (US) Panel to Review the National Children's Study Research Plan. The National Children's Study Research Plan: A Review. Washington (DC): National Academies Press (US); 2008.

Cover of The National Children's Study Research Plan

The National Children's Study Research Plan: A Review.

Show details

3Priority Outcome and Exposure Measures

As stated in the previous chapter, the core hypotheses of the National Children’s Study (NCS) were intended to serve as guidelines for the selection of outcome and exposure measures. The elaborate NCS planning process (described in Chapter 1) led to seven priority outcome areas:

  1. pregnancy outcomes,
  2. neurodevelopment and behavior outcomes,
  3. child health and development outcomes,
  4. asthma,
  5. obesity and growth,
  6. injury, and
  7. reproductive development outcomes.

Environmental exposure factors include the natural and built environment and the psychosocial environment. They comprise a wide range of biological, physical, chemical, genetic, social, cultural, and geographical factors. The NCS will attempt to examine many different exposures and link them in dose-response relationships with multiple outcomes. The study’s geographical dispersion and the varied socioeconomic and demographic characteristics of the study population have important implications for the collection of exposure measures.

This chapter first discusses each priority outcome area, which necessarily includes some discussion of the kinds of exposures that are proposed to be associated with one or more outcomes. The chapter then reviews categories of exposure measures in more detail.


For each priority outcome area, the discussion summarily describes the proposed hypotheses regarding specific outcomes and associated environmental factors as presented in the NCS research plan. It then offers the panel’s assessment in terms of public health significance and soundness of concepts and methodology. Each area ends with one or more recommendations.

Pregnancy Outcomes (1)


The specific pregnancy outcomes identified in the NCS research plan are birth defects, prematurity, outcomes of artificial reproductive technology (ART), and outcomes of pregnancy when the woman has subclinical hypothyroidism (NCS Research Plan, Vol. 2, App. A-2, Pregnancy Outcomes). The NCS proposes to focus on altered maternal glucose metabolism and folate and vitamin supplementation as risk (or protective) factors for birth defects; the role of inflammation in the pathogenesis of prematurity; the association of ART with fetal growth restriction, prematurity, and developmental disabilities; and the relationship between maternal subclinical hypothyroidism and developmental disabilities.

Assessment: Public Health Significance

The outcomes of pregnancy clearly represent an important area for research to which the NCS could make significant contributions. If the outcomes proposed for the NCS, birth defects, prematurity, and the outcomes of ART (and subfecundity generally) are certainly of public health significance. Taken together, they account for up to 15 percent of all pregnancies. Moreover, prematurity and birth defects have proven difficult to predict and prevent (Centers for Disease Control and Prevention, 2007; Institute of Medicine, 2006). Although ART is responsible for a relatively small percentage of births (1-5 percent), it nevertheless contributes significantly to poorer birth outcomes in the United States. Thus, a strength of this section of the NCS research plan is its focus on significant public health problems.

The public health significance of maternal subclinical hypothyroidism is less clear. Limited studies suggest that unrecognized hypothyroidism during pregnancy may lead to poorer cognitive function in the child. Most cases of hypothyroidism represent autoimmune disorders, but the NCS investigators hypothesize that environmental exposures may act to disrupt the endocrine system and produce hypothyroidism, although this has not been demonstrated outside the laboratory or wildlife. The reference cited (Landrigan, Garg, and Droller, 2003) does not list the candidate exposures, and it is not clear what these would be.

Another factor potentially contributing to subclinical hypothyroidism is maternal depression with alteration of the hypothalamic-pituitary-adrenal (HPA) axis. Although maternal distress can result in alterations of hormonal function and is associated with adverse pregnancy outcomes, it is unclear that the operational pathway is through subclinical hypothyroidism. Maternal depression is certainly associated with poorer cognitive and especially behavioral outcomes in the child, but, again, the operational pathway seems more likely to be through poorer maternal-child interactions than subclinical hypothyroidism. While clinical hypothyroidism is associated with such complications of pregnancy as preeclampsia, no evidence is presented that subclinical hypothyroidism poses such a threat.

In addition, because the NCS data collection sites have been selected using equal probability sampling, the distribution of exposures to environmental agents that might result in subclinical hypothyroidism is unclear. There may not be sufficient variability, especially of high and low levels of specific agents, to permit detection of their effect. In sum, the question addressed by the posited relationship of subclinical hypothyroidism and child development outcomes represents a highly speculative chain of logic, and the importance of the problem is not entirely clear, especially since it is said to affect only 2 percent of births.

With regard to other hypotheses that could be worth evaluating in the NCS—perhaps in place of the proposed research on subclinical hypothyroidism—we make three suggestions. First, the NCS could expand the proposed study of maternal depression as a factor in adverse pregnancy outcomes: The research plan limits evaluation of its role to its effects on subclinical hypothyroidism. Second, the NCS could reconsider its decision not to obtain dental records to establish maternal periodontal disease (NCS Research Plan, Vol. 2, App. A-2, p. A2-16), or at least mount a substudy (see Chapter 2) to collect the information needed to examine the effects of maternal periodontal disease on prematurity and other adverse pregnancy outcomes.

Finally, in regard to public health significance, the current list of hypotheses does not directly address one of the most critical and enduring reproductive public health issues in the United States: the causes of racial and ethnic disparities in birth outcomes, especially the elevated rates of poor birth outcomes among African American women. The current hypotheses address a relatively narrow set of clinical concerns, although this extensive study of 100,000 births has the potential to help focus the country’s intellectual attention on addressing its most fundamental issues with respect to reproductive outcomes.

Assessment: Methodological Concerns

Pregnancies without a live birth One concern about the treatment of pregnancy outcomes is that the research plan provides insufficient detail to understand how pregnancies that do not end in a live birth are to be handled. The NCS preconception and early pregnancy sampling design means that it has the potential to be one of the most important sources of scientific information on fetal loss, a critical pregnancy outcome. Many pregnancies may end in very early miscarriages, even before pregnancy testing, which could affect the ability to detect an association of early termination with inflammatory factors. In addition, prenatal diagnosis may lead to termination of pregnancies in which the fetus is assessed to be severely affected or nonviable. The research plan suggests that information on such outcomes will be sought. It will be important to employ sensitivity in data-gathering so that accurate information is obtained on these matters. The extent to which autopsy and other diagnostic materials will be obtained to ensure accurate descriptions of outcomes is unclear.

Appropriateness of the NCS design Clearly, a 20-year, longitudinal cohortstudy with an equal probability sample is not required to study pregnancy outcomes themselves. Most, if not all, of the pregnancy outcomes under investigation will be evident within a year or two of birth or perhaps by early school age, so that a 21-year time frame is not required to study them. Nevertheless, following up children into the later years of childhood will help track whether the impact of the reproductive outcomes persists beyond early childhood and what risk factors determine persistence and severity.

Prepregnancy exposure measures Many of the important questions aboutthe effects of various exposures on pregnancy outcomes may require obtaining assays before pregnancy. For the hypotheses considered under birth outcomes and others in the research plan that involve birth defects, the period of concern is the first few weeks of pregnancy, when much of organogenesis occurs. In terms of the hypotheses about maternal glucose metabolism, it would seem that a more efficient design for that specific investigation would be a case-control study of women with established diabetes who are intending to become pregnant with careful attention to the periconception period alterations in glucose metabolism.

One specific hypothesis involves the role of folate and multivitamin supplementation in the prevention of birth defects. While the literature cited indicates that such supplementation may reduce the risk of a number of birth defects, it is not clear that it will affect specific defects due to impaired glucose metabolism. In addition, for many of these defects, supplementation must begin before pregnancy, which again raises the question of the size of the prepregnancy sample.

Relationships among outcomes The research plan does not explicitly consider the connections, or “crosswalks” among the outcomes selected. For example, preterm infants have twice the risk of birth defects as full-term infants, but these two outcomes are not well connected in the plan. Prematurity is considered as an outcome for ART, but ART is not considered an exposure for prematurity. It is not clear how causality will be attributed when outcomes may be associated.

Statistical power The statistical power to address some of the proposedresearch topics is not clear. For example, the plan indicates that 1 percent of couples are exposed to ART, which would yield about 1,000 pregnancies (1 percent). As shown in the power tables (NCS Research Plan, Vol. 1, Sec. 10.2.3, Tables 10-1 and 10-2), this is a lower figure than required for many estimates. Clearly, there will be insufficient power to examine the effect of specific types of ART on the many different types of birth defects. The situation may be even more problematic with the restriction to singleton pregnancies, as the hypotheses propose. No estimate of the number of singleton ART pregnancies is given in the research plan. The proposed data collection effort does not appear to include the number of embryos implanted. This is an unfortunate omission, because a singleton pregnancy may have different implications if it results from the implantation of a single embryo rather than from the implantation of two or more embryos with spontaneous intrauterine demise or selective reduction.

The question of statistical power also affects the hypotheses regarding altered maternal glucose metabolism and birth defects. The research plan argues the importance of this question from the rise in obesity and type 2 diabetes. Even if the prevalence doubles from the plan’s estimates, only about 10 percent of women with altered glucose metabolism will have a child with a birth defect. The power calculations are based on all birth defects and all heart defects, yet “birth defects” is not a homogeneous group of conditions. Even such categories as “heart defects” comprise a large number of derangements of organogenesis. The ability to detect specific defects of a single organ, the heart, is considerably less than that suggested by the calculations for all defects. Thus, it is not clear that the study is adequately powered to explore the hypotheses under question.

Ascertaining birth defects Physical exams and digital photographs withspecific attention to dysmorphology are planned for birth and 6 months to identify birth defects; however, these methods will not pick up birth defects that are not associated with external stigmata. For example, the diagnosis of heart lesions not associated with other external physical signs might be missed without such specific examinations as EKGs and echocardiograms. Presumably, significant birth defects will be reported by the mother during subsequent interviews, but this is not clear. The accuracy of maternal reports in characterizing birth defects needs to be determined. Medical records would provide more accurate diagnoses, but the NCS only plans to abstract medical records at the time of delivery and neonatal examination for both mother and infant. Subsequent to the birth, medical and clinical event data will be collected by a personal health record only with the parent as the primary respondent (see Chapter 2).

The research plan proposes ultrasound examinations in the second and third trimesters; however, the second trimester ultrasound will be obtained only if the mother has not already had an early ultrasound for gestational age dating (NCS Research Plan, Vol. 1, Sec. 6.6, Table 6-1). Such early ultrasounds may vary considerably in the quality of the reading. In addition, such limited periodicity would not be sufficient to detect many instances of fetal growth restriction.

As noted above, the teratogenicity of an exposure is often dependent on the timing of the exposure during pregnancy. Experience of a teratogen in the first few weeks of pregnancy, when major organ development is occurring, is more likely to cause greater disruption than later in pregnancy. This argument would suggest that if, for example, impaired glucose metabolism serves as a teratogen, then it would be especially important to assess exposure in the first few weeks of pregnancy, when major organs are developing. In this regard, the proposed use of hemoglobin A1C as a measure will reflect the average blood glucose over weeks and not any fluctuations around the period of conception and organogenesis. Blood glucose measurements at various visits will reflect glucose metabolism at that visit and may not capture variations that occur as the woman’s metabolism adjusts to pregnancy.

Interventions The degree to which the study would ascertain ART interventions other than in vitro fertilization is not clear. Superovulating agents are not only more common, but also are frequently used by practitioners who do not specialize in the treatment of infertility. However, some of the diary and hormonal information collected by ART centers will be unavailable with these techniques.

Moreover, ART is but one of many interventional reproductive health service technologies that could be a focus of this large perinatal study. For example, there is critical debate and a serious lack of information about the sequelae of Caesarian births and the use of analgesics. Similarly, significant variations in outcomes among hospitals with neonatal intensive care units have been documented (Vohr et al., 2004). Attribution of developmental outcomes to ART or to any other prenatal intervention or exposure needs to account for this variation.


We have identified a number of issues and concerns with the NCS research plan proposals for assessing environmental influences on pregnancy outcomes. We offer two recommendations for steps that we judge to be of high priority for the NCS: one on the set of hypotheses that merit study and the other on the need for more specificity of the proposed research on pregnancy outcomes, which will be among the very first for which data are to be collected.

Recommendation 3-1: The NCS should consider replacing research on subclinical maternal hypothyroidism as a factor in adverse pregnancy outcomes with research on the effects of a broader set of maternal physical and mental health conditions, such as maternal depression, maternal perceived stress, and maternal periodontal disease.

Recommendation 3-2: The NCS should develop refined, detailed protocols for investigating all pregnancy outcomes, specifically a detailed protocol for obtaining information on various types of pregnancy loss, before beginning data collection at the Vanguard Centers, given that pregnancy outcomes are among the first outcomes to be examined; many outcomes lack clarity in measurement; and there are important questions regarding the adequacy of statistical power and the planned data collection (for example, the need for prepregnancy measurements of some exposures).

Although development of a detailed protocol will not help the statistical power issues for some outcomes as now specified, detailed protocols with more specific calculations on anticipated numbers of various exposures and subjects would certainly clarify what outcomes could be realistically studied, and whether changes to the research plan might be needed (e.g., not restricting assessment of the effects of ART to singleton births). Such protocols might also indicate subgroups for which more intensive study might be warranted, such as mothers with preexisting problems with glucose control. In addition, such protocols might lead to more focused outcomes (e.g., specific definitions with prevalence) and identify additional data collection required to ascertain these outcomes (e.g., echocardiogram data on congenital heart defects).

Neurodevelopment and Behavior (2) and Child Health and Development (3)

The Children’s Health Act mandate to “investigate basic mechanisms of developmental disorders” and “incorporate behavioral, emotional, [and] educational … consequences” of environmental influences encompasses an exceedingly broad range of developmental outcomes. Moreover, while developmental disorders are a clear study priority, environmental influences can also affect the much broader spectrum of age-normative developmental functioning. Given resource and burden limitations, the NCS faces difficult choices regarding the type and nature of its measurements of disorders and normative developmental outcomes.

To organize discussion of the issues and because the two domains are intertwined conceptually and in NCS planning, this section first discusses the specific plans for each domain and then provides a combined assessment.

Description: Neurodevelopment and Behavior Outcomes

The NCS proposes to focus on identifying specific developmental, behavioral, or mental health disorders, including sensory, motor, and learning disabilities, autism spectrum disorders, attention deficit–hyperactivity disorder (ADHD), anxiety disorders, depression, and schizophrenia and relating them to specific environmental exposures. The NCS will examine four broad hypothesized relationships (meta-hypotheses, NCS Research Plan, Vol. 2, App. A-1, p. A1-2; see App. A-2, Neuro/Behavior, for specific hypotheses within each meta-hypothesis):

Repeated, low-level exposure to nonpersistent pesticides … in utero or postnatally increases risk of poor performance on neurobehavioral and cognitive examinations during infancy and later in childhood….

Prenatal infection and mediators are risk factors for neurodevelopmental disabilities such as cerebral palsy and autism….

Exposures to adverse psychological, chemical, and physical environments and other stressors during vulnerable periods of pregnancy and early childhood can interact with genotype to cause or modulate behavioral problems in childhood….

Prenatal infection and mediators of inflammation during pregnancy and the perinatal period are associated with increased risk of schizophrenia.

According to the research plan (Vol. 1, Ch. 8, pp. 8.5-8.11), the NCS will rely on a combination of screening instruments and diagnostic information to identify developmental and mental health disorders. However, as stated earlier, the National Institute of Child Health and Human Development (NICHD) study staff have indicated that resources are not available at this time to abstract medical records except at birth (see Chapter 2).

Sensory, motor, and learning disabilities Some sensory and motor difficulties are evident very early in the child’s life. Learning disabilities, however, are often not identified until children enter school. Routine infant hearing screening is recorded in the hospital chart at birth, which will be abstracted by the NCS. Screening for sensory and motor disabilities will begin before the neonate has been discharged from the hospital by using the Network Neurobehavioral Scale to assess the infant’s neurological status.

The NCS plans to track children’s developmental status during infancy with regard to cognitive, motor, and language delays using multiple assessment strategies. At 12 months, the NCS will administer three of the Bayley III Scales of Development: Cognitive, Motor, and Language to all enrolled children to assess the achievement of developmental milestones in these domains. Actual diagnosis of learning, sensory, and motor disabilities will be confirmed whenever possible through the child’s medical records, including the diagnoses and treatment plans of their medical providers. The child’s health care visits will be reviewed at every contact with the parents, including both in-person contacts at 6 and 12 months and phone contacts at 3, 9, 18, and 24 months, and they will continue to be assessed regularly after that. Throughout childhood and adolescence, the child’s developmental status with regard to cognitive, language, and motor functioning will continue to be assessed periodically through direct testing by the NCS and diagnoses confirmed whenever possible through health care providers.

Autism spectrum disorders These disorders are not generally diagnosed until the child’s second year or later. The NCS will begin screening for autism spectrum disorders when the child is 18 months old and continue to screen for symptoms periodically through the toddler and preschool period by using the Modified Checklist for Autism in Toddlers (M-CHAT), a parental report instrument presumably provided over the telephone. The M-CHAT, however, is a screen for risk of autism; it does not yield a diagnosis of autism spectrum disorders. For diagnostic information, the NCS proposes to use diagnostic assessments conducted by the children’s health care providers and abstracted from medical records whenever possible.

Behavioral, attention, and mood disorders These disorders are rarely diagnosed in infants. At 12 months the parent will be asked to complete the Brief Infant-Toddler Social and Emotional Assessment (BITSEA), a screening instrument that assesses risk for mood problems, behavior problems, and self-regulatory deficits. The BITSEA, or an age-appropriate modification of the BITSEA, will be repeated through the toddler and preschool period to track any problems over time. As the children become older, other similar screening instruments will be used, such as the widely used Strengths and Difficulties Questionnaire, which assesses conduct problems, emotional problems, hyperactivity and inattention problems, and relationship problems, and can be completed by parents, teachers, and in the teen years by the adolescents themselves.

Early diagnoses of disorders will be confirmed whenever possible through the children’s health care providers’ records. Later in childhood, measures and diagnostic interviews, such as the Preschool Age Psychiatric Assessments (PAPA) interview or the National Institute of Mental Health Diagnostic Interview Schedule for Children (NIMH-DISC-IV), may be used to supplement diagnostic information from children’s health care providers and ensure diagnostic information on children who do not visit health care providers regularly.

Schizophrenia This psychotic disorder, believed to have both genetic andenvironmental etiology, will also be studied in the NCS. Schizophrenia, however, is rarely diagnosed before late adolescence or early adulthood. The research plan identifies no specific screening or diagnostic tools, stating that screening and diagnostic procedures for schizophrenia are likely to continue to evolve before the NCS children reach the life stage when schizophrenia is usually diagnosed.

Description: Child Health and Development Outcomes

Normative child health and development is concerned not with disorder or symptoms of disorder, but with individual differences in trajectories of normal, healthy adaptation over time. The NCS proposes to examine cognitive and language development and also social and emotional development. The research plan spells out five broad meta-hypotheses (NCS Research Plan, Vol. 2, App. A-1, p. A1-3; see App. A-2, Health/Development, for specific hypotheses within each meta-hypothesis).

The first meta-hypothesis links family resources and processes to the structure and quality of children’s home, child care, school experiences, and economic opportunities, which, in turn, affect developmental and health trajectories. The second hypothesis links geographic area of residence to exposure to social, physical, psychological, and environmental factors that adversely affect the risk of health problems and decrease access to protective resources. The third hypothesis links media use and content (TV, video, electronic games, Internet, mobile devices) to developmental trajectories from prosocial to antisocial behavior. The fourth hypothesis links interactions between children and families and the formal child care, school, and religious institutions in their communities to cognitive, social, and emotional development.

Cognitive and language development will be tracked throughout childhood using the procedures and instruments outlined for sensory, motor, and language disabilities above. The intent in this instance will be to identify normal development.

Under the rubric of social and emotional development, the NCS proposes to cover several domains of child functioning, both intrapersonal and interpersonal, including temperament, mother-child interaction, and relationship skills. The research plan posits that assessing temperament early in development is important, as temperamental qualities not only exert direct influence on children’s adjustment, but also influence parental reactions to the infant’s signals and needs and thus affect subsequent development indirectly.

When the infant is 6 months old, the NCS proposes to collect maternal reports of child temperament using three subscales of the Rothbart Infant Behavior Questionnaire-Revised (IBQ-R), including activity level, fearfulness, and positive anticipation of and approach to novelty. Also at 6 months, the NCS will conduct its first videotaped observation of mother-child interaction. At 12 months, the child’s social and emotional development will be assessed using parental reports on the BITSEA. During the toddler and preschool years, the same constructs will be assessed again, using the same procedures when appropriate, or using assessments that are age-appropriate measures of these constructs, such as the Strengths and Difficulties Questionnaire, which assesses prosocial behavior and relationship skills. As the child ages and begins to spend time in the broader social contexts of school and community, assessments of developmental trajectories in social and emotional competence will be tailored to include these experiential changes.

Assessment: Public Health Significance and Conceptual Concerns

Both the domain of neurodevelopment and behavior disorders and that of normal child health and development are clearly important areas to which the NCS could make significant contributions. However, the research plan has serious conceptual limitations in each domain that warrant concern.

Neurodevelopment and behavior In this domain, it is not clear by whatprocess and reasoning certain specific outcomes were selected for study and not others. In this domain, the NCS proposes to investigate sensory, motor, and learning disabilities, autism spectrum, ADHD, depression, anxiety, and schizophrenia. One could also add such outcomes as eating disorders, bipolar disorder, suicide, substance abuse, binge drinking, and high-risk sexual behaviors—all high-risk behaviors of public health significance that are common in adolescents and many of which are likely to be influenced by developmental and environmental conditions prevailing much earlier in childhood. It is not the case that these disorders are excluded because they do not become manifest until adolescence, since schizophrenia is included and it is not typically diagnosed before young adulthood. The research plan does not indicate why the specified outcomes were chosen and other important outcomes were not. Clearly, not all disorders can be studied in the same depth, and it is important to have a clear rationale for selecting the specific ones to study.

The hypotheses that propose to link environmental exposures to adverse neurodevelopment and behavioral outcomes reflect a biomedical approach that appears limited in its ability to contribute to understanding of this complex area. For example, the hypothesis to link repeated, low-level exposure to nonpersistent pesticides to poor neurobehavioral and cognitive skills is important because of widespread levels of exposure and the likelihood that children are more sensitive to exposures than adults. However, all other factors that could mediate a relationship of pesticide exposures to adverse outcomes are lumped into “potential confounders” without a clear discussion of likely pathways of interactions that should be explored, or of likely dose-response relationships. The only exception comprises two specific hypotheses in this area that propose to examine the mediating effect of genetically decreased paraoxonase activity (an enzyme that protects against the adverse effects of low-dose organophosphate pesticide exposures).

Similarly, the proposal to study depression and other mood disorders reflects an important area of investigation. The NCS takes a largely biomedical approach to these outcomes. Specific hypotheses focus on the role of genetically based levels of neurotransmitters, such as serotonin, norepinephrine, and dopamine, as they interact with “life stress” to the exclusion of other influences and pathways that may be operative. For example, negative attributional style is widely hypothesized to be relevant to the development of depression, but there are no plans to assess this construct (Gibb et al., 2006). (Exposure to prenatal tobacco smoke and “maternal insensitivity” are posited as mediating factors in the relationship of dopamine to ADHD.)

Positive health and development outcomes In this domain, the discussionin the research plan does not define clearly how the NCS understands the key constructs of what constitutes “health” and “normal development” nor the relationship of normal variation to diagnosable problems. Moreover, the specific hypotheses in this area not only are very general regarding developmental processes and the factors that are assumed to promote healthy development, but also are not well integrated into an overall concept of healthy development. For example, under the first meta-hypothesis, the NCS proposes to look separately, and also interactively, at the effects of the nature and stability of family structures, families’ social networks, family socioeconomic status, family members’ health status, parents’ promotion of healthy lifestyle behaviors, and parental monitoring of children’s activities at home and in their neighborhoods. Similarly, under the fifth meta-hypothesis (added in September 2007), the NCS proposes to look separately at the effects of parenting behavior, child care experiences, access to and use of health and social services, and such child characteristics as intelligence and temperament, on healthy child development. Pathways for the separate and interactive effects of these factors are not clearly specified, nor are the factors always clearly defined.

Assessment: Methodological Concerns

In the neurodevelopment and behavior domain, once disorders are selected for study, they must be precisely coupled with procedures to screen for them and subsequently to diagnose them in a rigorous fashion. The current plans for achieving this goal do not appear to meet this standard. As noted above, the research plan identifies several recognized tests for screening for the disorders studied, but these screening instruments are not useful as diagnostic instruments. The NCS appears to rely mostly on health provider records whenever possible and on mothers’ reports for diagnoses. Yet, as we have noted previously, plans for the use of pediatric medical records, except at birth itself, appear to be on hold because of resource constraints. Discussion of medical records with parents, which the research plan mentions, is not likely to yield reliable information. Nor does the plan address the adequacy of medical records themselves, which may be insufficient for diagnostic purposes even if they are obtained. Reliable diagnosis using Diagnostic and Statistical Manual criteria, subsequent to positive findings from screeners or medical records, may take extensive interview time and necessitate the gathering of denser observational data than is currently contemplated. Moreover, access to medical records information will depend heavily on the participant’s type of insurance for mental health services. Developmental disabilities fall under mental health codes, so that funding may be limited and professionals willing to take such funding not readily accessible.

Moving to assessment of normal, healthy functioning, there is little attention to evaluating normal variation in such aspects as attention, working memory, executive function, language ability, attachment, and peer relations. The outcome measures discussed partially overlap with those listed in the neurodevelopment and behavior section and include:

  • 3 Bayley scales;
  • CDI (Communicative Development Inventory) at 12 months;
  • intelligence tests—perhaps Woodcock-J scales at later ages or Bayley or Kaufman scales;
  • unspecified assessments of executive function and attention;
  • Rothbart IBQ-R measures of temperament;
  • videotaped mother-child interaction at 6 months; and

This list appears ad hoc and not well informed by contemporary child development research or well focused on specific hypotheses or domains of interest. For example, Achenbach’s measures of psychosocial functioning seem to be missing, even though they are the usual ways of looking at internalizing and externalizing behavior across a normal range. As another example, there is no plan to use the Strange Situation at 12 months to evaluate attachment security, even though there is a vast literature on attachment theory and on the prospective implications of Strange Situation classifications. Other examples of missing measures include the Continuous Performance Test (also relevant to ADHD diagnosis) and measures of verbal and visual-spatial working memory. Better measures of infant cognition would include the Fagan test, Rose’s cross-modal matching paradigm, or Haith’s visual evoked potential (VEP). These tests are known to predict later IQ better than the Bayley and to be sensitive to the effects of environmental contaminants when the Bayley is not (Jacobson and Jacobson, 1996). A final example is that the CDI will not, at 12 months, provide a good assessment of language acquisition. It needs to be repeated at subsequent ages and coupled with a more intensive battery at about 4 years, using, for example, the Reynell or the deVilliers test (Seymour, Roeper, and deVilliers, 2003a, 2003b).

For both outcome sections, there is no information on the reliability of tests and procedures, the length to administer them, their validity and especially their discriminant validity, when and how often they will be given, and whether there is a need to worry about repeated testing effects. These are basic research issues that should have been discussed. Finally, the research plan needs to address ethical issues regarding referral for treatment when disorders are discovered and what to do when such treatment is not covered by insurance (see Chapter 5).

In summary, in measuring neurodevelopment and behavior and child health and development outcomes, there is nearly exclusive reliance on screeners, brief questionnaires, and psychometric tests, chosen in a way that does not seem closely related to hypotheses regarding domains of interest. Direct behavioral assessments of children are very limited; instead, measures are largely being derived from reports by parents rather than from validated behavioral tests that are available. As noted in Chapter 2, there are infrequent home or clinic visits after age 2, which seriously limits the ability to observe and measure normal, or pathological, behavioral development.

We recognize that a number of well-validated instruments that are not proposed in the research plan may be too burdensome for respondents and too costly to administer given the wide range of exposures and outcomes that the NCS is striving to measure. As multifaceted as the NCS is, it cannot cover all domains in the same degree of detail, and it may be that the domains of neurodevelopment and behavior and child health and development are ones that must be left less well covered than other domains. To the extent that is the case, it is incumbent upon the NCS to carefully review the specific outcome measures that are proposed to be sure they are the best possible measures within the constraints of resources and respondent burden. In this regard, the NCS could make more extensive use of measures used in the National Childcare Study (updated as appropriate), as a large group of inventive and dedicated behavioral scientists faced the same basic constraints as the NCS when it developed the set of cognitive and behavioral measures it employed.


Our discussion has identified a large number of concerns. Most important is the need for the NCS to justify the selected outcomes of interest and related exposure measures, given the wide range of possible outcomes the NCS could study in the two domains of neurodevelopment and behavior and normal health and functioning. The NCS needs to indicate clearly why some outcomes were chosen and others not, by using the criteria of the research plan itself, which include the current state of knowledge, the need for a large sample size to answer key questions, and the public burden of particular disorders. Despite the wide mandate of the Children’s Health Act, it is not possible to assess all normal variation in healthy functioning as well as all possible disorders in a rigorous fashion. One strategy might be to select a smaller number of disorders and orient screeners, normal assessments, and diagnostic interviews around those areas of functioning. That may allow for more principled decisions about ages at assessment, laboratory versus home visits, and other methodological issues.

Recommendation 3-3: The NCS should develop a clearer rationale for the selection of specific neurodevelopment and behavior disorders to be considered in the study and a clearer conceptual basis for the assessment of normal child health and development trajectories and outcomes. Clarity is needed to guide the choice of outcome measures and exposure measures and the frequency and types of contacts (at the home, in clinics) with study participants in order to obtain the best information possible within resource and burden constraints.

Asthma (4)


The NCS proposes to examine six hypotheses about risk factors for childhood asthma (NCS Research Plan, Vol. 2, App. A-2, Asthma). Broadly, the hypotheses assume (1) prenatal maternal stress as a risk factor, mediated by genetic and environmental factors that influence immune development and lung growth/airway inflammation in early life (for example, prenatal exposure to tobacco and air pollution); (2) indoor and outdoor air pollution and aeroallergens as risk factors, mediated by genotype; (3) antioxidants in diet as protective factors; (4) social environmental factors that influence exposure to physical environmental risk factors, psychosocial stress, and adverse health-related behaviors, thereby explaining, at least in part, disparities in asthma prevalence and severity by race and socioeconomic status; (5) early exposure to common indoor aeroallergens and microorganisms as protective factors; and (6) significant gene-environment, gene-gene, and genotype-phenotype interactions as risk factors.

Assessment: Public Health Significance

Asthma is of high public health importance in terms of both the burden of this chronic disease and its documented increase in incidence. Asthma continues to be the most common chronic disorder among children in the United States, affecting 9.9 million children (or 14 percent of all children) under age 18 (Bloom and Cohen, 2007). Asthma is one of the leading causes of school absenteeism, accounting for 12.8 million missed school days in 2003 (Akinbami, 2006). It is appropriate, and responsive to the Children’s Health Act of 2000, for the etiology of asthma to be a major focus and aim of the NCS.

The major strength of the proposed NCS in the area of asthma is the prospective cohort design, which will allow for in-depth investigation of early life risk factors for the development of asthma. There are only a few large prospective cohort studies on potential risk factors that contribute to the increased incidence of asthma. Alternatively, epidemiological studies on risk factors for the exacerbation of asthma are numerous. Therefore, the major strength of the NCS is the ability to study asthma incidence beginning with prenatal exposures. Follow-up into the later years of childhood is necessary to understand whether asthma symptoms persist beyond early childhood and what risk factors determine persistence and severity.

The proposed measures of asthma outcome are valid and feasible to collect. They include questionnaire data, physician-confirmed diagnoses, and such objective tests as spirometry and peak flow measures.

Assessment: Methodological Concerns

Issues of statistical power arise for research on asthma as for other outcomes. For instance, it is not clear whether there will be, for analysis purposes, (1) a sufficient number of children from a variety of housing conditions, such as inner city housing developments compared with more rural housing or (2) a sufficient number of children from a variety of geographic conditions that contribute to differences in environmental exposures, diet, and other factors. Insufficient numbers of children on these dimensions could weaken the ability of the NCS to estimate robust associations that suggest causal relationships (see Chapter 4 for further discussion).

Throughout the asthma section, the research plan refers to the collection of biological samples to measure biomarkers of exposure and response. It is not clear, however, whether the study has adequately considered the appropriate timing (during pregnancy and childhood) of the collection of these samples. An important aim of the study should be to determine whether the difference in timing of an exposure affects the risk of the development of asthma. Given the infrequency of follow-up after birth proposed for the NCS (see Chapter 2), it is unlikely that these questions of timing will be satisfactorily addressed.

The research plan proposes to measure exposure to indoor and outdoor aeroallergens, which is appropriate given the results of a recent study by researchers at the National Institute for Environmental Health Sciences, which found that more than one-half of the current asthma cases in the country can be attributed to allergens (Arbes et al., 2007). The research plan needs to provide more detail, however, on how aeroallergen exposure will be assessed and which aeroallergens the study will focus on. Aeroallergen exposure assessment is not simple and straightforward, and detailed protocols are necessary. Dust samples will be collected to measure mold, endotoxin, and common environmental antigens. It is not clear whether or how environmental chemicals, such as pesticides, and metals will be measured in the dust samples.

Accompanying the increase in asthma incidence among children has been a parallel increase in children’s allergies, including severe food allergies. It would be straightforward to collect information on food allergies among the NCS cohort and to explore potential risk factors ranging from maternal diet to early life exposure to allergens. Furthermore, data suggest that children with certain food allergies, such as to eggs, have a higher risk of developing asthma. This hypothesis could readily be explored.

Recommendation 3-4: The NCS should develop a clearer rationale for its hypotheses about factors that may increase the incidence of asthma. These should focus on prenatal and early life risk factors.

Obesity and Growth (5)


The NCS proposes to investigate several ways in which environmental factors may increase the risk of obesity and insulin resistance (type 2 diabetes) in children (NCS Research Plan, Vol. 2, App. A-2, Obesity/Growth). The NCS has chosen four biomedically oriented meta hypotheses to investigate: (1) impaired maternal glucose metabolism during pregnancy is directly related to the risk of childhood obesity and insulin resistance; (2) intrauterine growth restriction is associated with risk of central-body obesity; (3) breast milk compared with infant formula feeding is associated with lower rates of obesity; and (4) consumption of a high glycemic load diet is associated with obesity. In addition, the study proposes to relate the development of type 1 diabetes to an interaction between genetic susceptibility, early exposure to viral infections, and early exposure to cow’s milk protein or other dietary components. The NCS research plan does include getting growth measures as direct observations. In addition, the NCS will be obtaining related blood samples for lipids and glucose metabolism abnormalities.

The NCS research plan, however, does not define precisely or provide rationale for what it means by obesity. Recent clinical recommendations focus on specific cutoff values for body mass index (BMI). However, there is controversy as to what standards should be used for BMI (Cole et al., 2000) and whether all standards apply to specific groups of children. In addition, other measures of excess fat (e.g., waist circumference, bioelectrical impedence) are available (although again, there is controversy as to the information added by such measures) (Krebs et al., 2007). The NCS should be specific as to why certain measures are preferred. There is also reference in the appendix to “body habitus,” but this is likewise not defined. Exercise is not mentioned in the hypotheses on obesity and growth. It could be asked directly of parents, and it could also be examined in terms of the availability of recreational facilities in geographical areas. As with other outcomes, no mention is made of diagnostic assessments to eliminate causes of obesity that are not of interest to the NCS.

Assessment: Public Health Significance

Childhood obesity has been increasing steadily, particularly during the past two decades (Ogden et al., 2002, 2006). This rising rate of obesity is a major health concern, both because of its impact on childhood health and because of the strong association between obesity and cardiovascular disease risk, hypertension, and type 2 diabetes that begins in childhood and continues into adulthood (Institute of Medicine, 2007). Numerous studies over the last decade have indicated a large variety of interrelated risk factors for childhood obesity (Barlow, 2007; Hawkins and Law, 2006; Skelton, DeMattia, Miller, and Olivier, 2006). It certainly deserves to be a priority outcome of the NCS.

Assessment: Conceptual and Methodological Concerns

The large sample size of the NCS provides an important opportunity to evaluate such hypotheses as the effects of impaired maternal glucose metabolism and intrauterine growth restrictions on childhood obesity. In general, however, the hypotheses selected by the NCS appear too narrowly focused to result in new information that could ultimately reduce the epidemic in childhood obesity and curtail its adult sequelae. There are too many additional factors that play an important, if not essential, role in the development of obesity, including a range of social and psychological factors acting at both the individual and collective levels.

The current hypotheses selected focus on only a very small part of the complex pathways to the development and chronic condition of obesity. A more integrated and systematic examination of a wider range of factors, many of which will be collected as a part of the NCS, would be likely to have more concrete benefits for improving children’s health. Such an examination should elucidate the web of physical, behavioral, social, and genetic factors that act additively or interactively to determine intrauterine growth retardation, breastfeeding versus infant formula feeding, and children’s diet, along with the role of these factors during childhood development. A systems approach to understanding this complex chronic condition could be one alternative.

Although it is valid to test hypotheses about particular mechanisms of action in epidemiological cohorts, one must remember that association is not causation. For the hypothesis about glycemic load, the difficulty of inferring causation is magnified by measurement issues. There are serious concerns about the ability of the NCS to measure the consumption of high glycemic load diet using food diaries. Moreover, the time frame for sampling information about the child’s diet is neither well motivated nor connected to the hypotheses to be addressed.

The research plan raises other questions. First, it clearly delineates a work-up to rule out other reasons for obesity. However, the research plan does not indicate what maneuvers will be done to ensure that the obesity detected is due to diet or physical activity and not to metabolic problems. Second, there are questions about what will be done when the obesity is detected. Will treatment plans be put into effect? How will that alter the “natural history” portion of the study?

One last major concern is the tacit, but evident, focus on the mother as the source or cause of obesity in her children. While there is little debate that maternal factors play a role, many other factors and circumstances can be expected to have an impact as well.


Recommendation 3-5: The NCS should reevaluate its main hypotheses to be addressed in the study of childhood obesity and consider adopting a broader approach that incorporates social and psychological factors as well as biogenetic ones. Such an approach would help the study identify the constellations of key factors and their interrelationships that are important to understand in order to develop the most effective public health measures to reduce childhood obesity.

This is another situation for which consultants could be useful to assist the Vanguard Centers in developing and testing protocols that could lead to improved instruments.

Injury (6)


The NCS singles out repeated mild traumatic brain injury (rMTBI) for attention in the broad field of unintentional childhood injury, positing that it has a cumulative adverse effect on neurocognitive development (NCS Research Plan, Vol. 2, App. A-2, Injury). Under the category of injury, the NCS also proposes to examine the relationship of biological, physical, and psychosocial components of the environment and environment-gene interactions on early onset and continuation of antisocial physical aggression, as well as the relationship of genetic, family structure, and neighborhood and community factors interacting with traumatic life events on the risk of anxiety disorders.

Assessment: Public Health Significance

Childhood injuries are a major source of child (and adult) morbidity and the principal source of mortality in childhood. Unintentional injuries are the leading cause of death and hospitalization among children and teens in the United States, with the highest rates recorded among minority populations. In 2004, nearly 20,000 (44 percent) of the 45,000 deaths of people ages 1-24 were caused by injuries (Bloom and Cohen, 2007:Table 32). The most serious of the nonfatal injuries is traumatic brain injury. The NCS has the capacity to provide critical new information about risk factors (intrapersonal and environmental) and development sequelae of childhood injuries. This measurement focus is responsive to the Children’s Health Act of 2000. Each of the three proposed research hypotheses addresses important new areas for injury research that are particularly amenable to longitudinal designs, and each represents a distinct facet of injury research: identification and sequelae of new injury types (rMTBI); longitudinal antecedents of injury-inducing behavior (factors resulting in childhood or adolescent-onset aggression); and long-term sequelae of injury experiences (antecedents and resilience to traumatic life events in childhood).

Assessment: General Concerns About Data Collection

The NCS has thoughtfully addressed some of the measurement issues for the area of injuries. In particular, the research plan effectively discusses the injury threshold issue—for example, the determination of the severity to be counted as an “injury.” It proposes a variety of thoughtful data collection methodologies, including initially parent and later child self-reports, health visit logs, and activity diaries to measure the sequelae of injuries that result in the limitation of activities. However, details about the content or implementation of the health visit logs or activity diaries are notably absent—and their implementation in a large national sample could be problematic. Variations in local data collection methods and content could limit the acquisition of national injury data. The relative infrequency of data collection during middle and later childhood years is problematic for surveillance and case identification, especially for less serious injury experiences. No mention is made about using or linking to existing injury databases to further inform the proposed injury data collection efforts.

Assessment: Quality of Hypotheses

The three NCS research hypotheses represent the range and kinds of innovative injury research available from the NCS. However, the quality of the three specific hypotheses is uneven. One of the three proposed injury foci—rMTBI—addresses a newly recognized type of injury and its cumulative adverse effects on subsequent neurocognitive functions. It is unclear, however, if the NCS is appropriately designed or has appropriate data collection capacity to study this specific injury. MTBI is a relatively rare injury, estimated to occur in 7 per 1,000 children visiting the emergency room per year. However, as the NCS research plan notes (Vol. 2, p. A2-280), the multiplicative probability of a repeat emergency room visit among the NCS sample is exceptionally small, with an estimated 100 cases over the entire 20 years of the NCS. Moreover, the NCS provides no direct measure of the mild brain injury itself (via neuroimaging or other means). As proposed, this research hypothesis will be based only on (unspecified) parent reports about each head injury, with perhaps a medical records examination to verify a reported condition. It is also likely that many of these children will not seek medical care. The infrequency of data collection in middle and late childhood years suggests that this research will have to rely on retrospective information. As currently constructed, the NCS does not appear to be an appropriate locus to analyze this hypothesis. Its stated rationale, “the fact that little is known about rMTBIs is a compelling justification for this study,” is itself not a sufficient justification.

The second research hypothesis examines the developmental antecedents and pathways of early preadolescent onset compared with adolescent-onset aggressive behavior, with its assumed associated risk for injuries, especially intentional injuries. The NCS could help in disentangling the multiple sources of aggression seen in older adolescents and thereby offer insights into more effective targeted interventions. The current research seems, however, to focus more on the classification of aggression pathways and less on the roles of psychosocial and other interventions (ameliorative or harmful) in affecting the developmental trajectories. Moreover, the links between increased aggression and increased injuries is assumed but would be worthy of study.

The third hypothesis—antecedents and resilience to traumatic life events in childhood—seems almost only metaphorically linked to the injury section of the NCS research plan; perhaps conceptually reflecting on response to trauma. The hypothesis focuses primarily on anxiety disorders and the child’s response to cumulative exposures to potentially traumatic events—primarily influential life events, but also including physical abuse and injury exposures. While this hypothesis is appropriate for the NCS, it would be much better situated in the child health and development section of the research plan. While the emerging literature on posttraumatic stress disorder (PTSD) underlies this topic, as written the research plan provides little discussion of how injuries (especially intentional injuries) relate to subsequent anxiety disorders, or their relative contribution compared with other life traumas.

Environmental causes of childhood injuries, and clinical responses to injuries, are major issues in discussions of injury policy. Neither is well developed in the research plan. The NCS proposals for the measurement of home and community environments are not well articulated—an issue that is important for other research topics as well (see section on “Exposure Measures” below). Clinical treatment (or nontreatment) of an injury, as a modifier of the injury’s long-term sequelae, is never mentioned.


Recommendation 3-6: The NCS should consider replacing research on repeated mild traumatic brain injury (rMTBI) with more nuanced research on other injury-related topics, such as environmental factors in childhood injuries and the effects of clinical response to injury (treatment or nontreatment).

Hormonally Active Agents and Reproductive Development (7)


The NCS proposes to address the effects on the reproductive system from exposure to hormonally active agents (HAAs) in the environment. The study will examine the direct effects of prenatal exposure to phthalates on hypospadias in boys (abnormally placed urinary opening); prenatal exposure to polybrominated diphenyl ether on hypothyroidism, which, in turn, can lead to altered reproductive development; and exposure to bisphenol A on the risks of childhood obesity. The study will also examine the effects of environmental exposures to HAAs on later sexual maturation, including the effects of early childhood exposure to bisphenol A on early onset of puberty in girls; early childhood exposure to phthalates on the development of polycystic ovarian (masculinizing) syndrome (PCOS) in adolescent girls; and exposure to lead in early childhood and at the time of sexual maturation on delayed onset of puberty in girls. The study proposes to examine potential interaction between genetic variations and environmental exposures to HAAs in analyzing the proposed outcomes.

Assessment: Public Health Significance

Exposure to hormonally active agents and their potential effects on reproductive outcomes is an important issue for children’s health. Exposure to such agents is widespread, and the potential outcomes are widely observed. Documentation of the associations between HAAs and reproductive outcomes in humans, however, is limited, and speculation about the potential effects is limited to inferences drawn from cross-sectional and retrospective studies or from animal studies. Elucidating the effects of such agents requires a large, longitudinal sample. It should begin before conception in order to investigate the effects of early developmental exposure. The NCS design is therefore well suited for these research questions.

The NCS research plan selects several conditions to illustrate the effects of HAAs: hypospadias, hypothyroidism, obesity, early (and late) onset of puberty in girls, and polycystic ovary syndrome. Cryptorchidism (undescended testes) is mentioned in the section on prevalence/incidence, although it is not clear whether this is meant in isolation or in conjunction with hypospadias. Specific environmental exposures to be examined include polybrominated diphenyl ether, phthalates, bisphenol A, and lead. The focus on these outcomes and exposures represents a strength of the NCS.

In addition to the chemicals listed above, there are many other hormonally active compounds with widespread human exposure. They include, among others, UV sunscreens and parabens (used as a preservative in foods and cosmetics). An archive sample of urine and blood could be collected for future analysis of these classes of compounds. In addition, information could be collected on dietary consumption of phytoestrogens during pregnancy. A urine archive could also be used to measure phytoestrogens.

Assessment: Methodological Concerns

For none of the conditions identified does the NCS research plan provide a clear definition of the outcome. Hypospadias and PCOS can vary significantly in severity, and it is not clear that all degrees of severity are equally important. PCOS may also be difficult to detect in adolescents and some ethnic groups (Norman et al., 2007); it may not become evident until childbearing is attempted. There is controversy about which definition of PCOS to use and which diagnostic procedures to implement (Norman et al., 2007; Toledano and Nelson, 2007). Moreover, other causes of hyperandrogenism must be eliminated, and this diagnostic issue is not raised in the research plan. The issue of premature puberty in girls and changing secular trends remains controversial, may differ for different ethnic groups, may not actually be associated with changes in the onset of menses in girls, and has primarily been associated with obesity (Himes, 2006). Thus, the exact targets of analysis are unclear.

Missing from the list of outcomes is the assessment of pubertal onset in boys, which can be easily evaluated and should be included in the research plan. At a minimum, Tanner staging should be used to assess pubertal developments in both boys and girls (Tanner, 1962). The research plan indicates that the NCS will refine its approach to ascertaining the onset of puberty, and there is time do so. It is not clear from the research plan if children will have a physical examination as part of the study. However, for defects present at birth or early childhood, there is not that luxury of time.

Of relevance for statistical power considerations are prevalence estimates for the specified outcomes, which range from 0.8 percent for hypospadias, through 6 percent for PCOS (women of all reproductive age ranges, not just adolescents), to as high as 52 percent for 7-year-old and for 8-year-old African American girls showing signs of precocious (early onset) puberty (Herman-Giddens et al., 1997). No prevalence is given for hypothyroidism. For most outcomes, the NCS research plan notes that only half the sample can be used (either male or female), thus reducing power, and prevalence may be lower depending on the ethnic group. The expected NCS sample sizes in the research plan for estimating power are derived from studies with relatively high levels of exposure. The plan argues that the power should be sufficient to detect differences if the prevalence of exposure is higher than the studies in the literature. However, this remains an assumption.

While recognizing that there is widespread general population exposure to most of the hormonally active compounds of interest, it is not clear that the NCS sample design will result in an optimal distribution of exposures or outcomes. The reason is that relatively high exposures may be needed to detect associations (Toledano and Nelson, 2007). Specifying one or more substudies that enriched the sample with high-exposure population groups could be valuable.


Recommendation 3-7: The NCS should develop refined and detailed protocols for studying reproductive development outcomes, which, as presented in the research plan, often lack clarity in measurement and research design. Outcomes that are measured at birth for which there is little time to refine research protocols require immediate attention. The NCS should use results from the Vanguard Centers, such as estimates of the prevalence of specific reproductive development outcomes, to assist in protocol development, and it should consider the usefulness of substudies of high-exposure population groups.


In this section, we provide assessments of the proposed measures of exposures. These measures include the demographic and socioeconomic characteristics of parents, the chemical environment (persistent chemicals, nonpersistent organic compounds, air pollutants), the physical environment (housing and neighborhood conditions), psychosocial exposures, biological exposures, and genetic markers. (See NCS Research Plan, Vol. 2, App. F-1, for details of planned interviews and other methods to collect various kinds of exposure data from prebirth through age 24 months.)1 We then note important types of exposure data that are missing and should be added to the NCS research plan when resources permit. We conclude with a discussion of the desirability of facilitating linkages of the NCS to other data sources that could enrich data on exposures.

Demographic and Socioeconomic Measures

The NCS properly recognizes the need to collect a variety of interview-based demographic data for participating children and their parents, including age, gender, race, and ethnicity, together with measures of socioeconomic status (SES), such as education, occupation, home ownership, income, and family structure, for the parents. These data are needed as a means to classify population groups of interest, as controls for the confounding influences of factors associated with demographic and social position in causal analyses, and as dimensions for the focus on health and developmental disparities that is mandated by the Children’s Health Act of 2000.

By and large, the research plan lays out a cogent strategy for gathering these kinds of data, relying on widely used instruments from the decennial census, the American Community Survey (ACS), and the Survey of Income and Program Participation (SIPP). Our discussion of these measures focuses on potential problem areas.

Education, Employment, and Income

The research plan calls for measurement of educational level, employment status, and income of the mother and father at the initial face-to-face visit using standard questions from SIPP, Census 2000, and the ACS, with updating at the in-person visits during pregnancy and following the birth of the child. With regard to education, we applaud the recognition that parents, particularly younger parents, often acquire more education after the birth of children, and that the NCS will need to ask about completed schooling periodically over the course of the study.

Income and poverty status are key constructs of socioeconomic status, and, given their volatility, it is important to measure both recurrently. The research plan details how this would be done between prepregnancy and age 1 but not in the 24-month phone interview. The proposed SIPP income question battery may be too detailed to be worth the interviewing time; the Current Population Survey (CPS) or other surveys (for example, the Panel Study of Income Dynamics) provide validated shorter alternatives.

It is unclear what “Supported by income” in the measures grid means. Money mingles within households, and it is sufficient to get a good accounting of household-level income and composition.

Maternal employment will obviously vary a great deal surrounding the birth of the child, with relative high employment rates during the early stages of pregnancy, near-zero rates around the birth, and then increasing rates during the months following the birth. A key design issue is whether measures of employment (and income) will be gathered as “time of interview” variables or in an event-history format (for example, month by month) that would provide a “motion picture” of parental employment across the survey period.

Given the constraints on interviewing time, it is probably not worthwhile to invest in a detailed 21-year motion picture of all aspects of maternal employment. But it may well be worthwhile to invest interview time in a month-by-month employment history checklist (for example, whether employed or not), beginning with the month prior to the enrollment interview and extending (through retrospective questions in subsequent waves) at least until the child’s first birthday and perhaps throughout childhood. More details about month-by-month employment (for example, wages, work hours) may well require too much interviewing time relative to their analytic value, but wages and work hours questions modeled after the CPS should probably be asked about the week prior to each interview.

In addition to employment status, work hours, and wages, the research plan should include occupation. It is not clear whether and how the NCS intends to generate data on occupation, which is an important indicator of socioeconomic status. Questions about occupation should be asked (for example, using questions from the CPS) at the prepregnancy interview and then again at, say, the child’s first birthday as well as in later interviews.

Immigrant Status

It is not clear what immigration measures are planned, other than a standard “country of origin” question. Given health differences between immigrant and nonimmigrant children, the cultural implications of country of origin, and the difficulties of accessing health care for undocumented immigrants, it is vital to ascertain immigrant generation, languages spoken (primary, secondary), acculturation, and, if possible, the legal status of parents.


The NCS measures of demographic and socioeconomic variables are generally well planned. The highest priority is to include more information relative to immigration status.

Recommendation 3-8: The NCS should add to its well-planned battery of demographic and socioeconomic measures questions on immigrant generation, languages spoken, and, if possible, the legal status of the parents and child.

Chemical Exposure Measures

As described in the research plan, the primary purpose of the NCS exposure assessment is to enable the epidemiological analysis of associations between priority exposures and priority outcomes that are described in the study hypotheses. The study will archive environmental chemical (and biological) samples for potential future analysis of other exposures that are not deemed a priority in the current proposal. Broad classes of environmental chemical exposures include, among others, metals, modern chemicals, pesticides, air pollution, and drinking water disinfection by-products.

General Comments

The overall goals of exposure assessment strategies in epidemiological studies are to identify populations with a wide range of exposures to maximize power and to identify populations with high exposure levels, because associations are more likely to be detected in such a setting. In addition, measurements need to be accurate and precise. It is well recognized that in environmental epidemiological studies, exposure assessment is expensive and there are numerous limitations.

One concern, which applies broadly for the NCS, is whether the national equal probability sampling strategy will enroll sufficient numbers of children living in different conditions across the United States to ensure a range of exposures to environmental chemicals. For some contaminants, such as air pollution and pesticides, there are segments of the population with high exposures, and the NCS could usefully consider ways to enrich the study population with these segments through substudies or adjunct studies. Doing so would ensure that the NCS has large enough samples of children with the conditions of concern for the main hypotheses, as well as large enough samples to understand the causes and consequences of health disparities. The NCS could also usefully consider a plan to reassess the sampling scheme as recruitment progresses in relation to the exposure variability among participants (see Chapter 4).

The NCS exposure assessment strategies, as described in the research plan, strive to measure the magnitude, frequency, and duration of exposure. The study should also strive to measure exposures at the relevant etiological exposure window, which generally requires repeated measurements for later sophisticated statistical analysis. The research plan makes some general statements on the matter, but detailed plans and the rationale for the timing of data collection need to be provided to ensure that this critical—and difficult to accomplish—aspect of the exposure assessment strategy is well specified.

The research plan mentions the need for a validation study of exposure measurement in subsamples of the cohort, but it does not describe how this will be carried out and incorporated in the overall analysis. The plan should provide further details, which will of course vary depending on the characteristics of the exposure of interest.

Figure 9-1 in the research plan (NCS Research Plan, Vol. 1, p. 9-4) is used to illustrate that questionnaire data, observations, environmental measures, and biological measures will be combined to assess exposures. However, the figure alone is not detailed enough to provide a description of how this will be accomplished. The plan needs to specify how data from these multiple sources will be combined into exposure metrics for the priority candidate exposures outlined in the proposal.

Finally, the NCS should give consideration to collecting biological samples for archives that can be later analyzed for hormonal activity. Bioassays exist to measure total estrogenicity or antiandrogenicity of a biological sample. This approach is less expensive and potentially more relevant biologically than standard chemical analysis for many compounds.

Persistent Chemicals

The research plan states that the assessment of maternal persistent organic chemical (POC) burden, which is an indirect measure of fetal exposure, can be obtained from maternal blood taken before or during pregnancy or obtained from maternal blood, milk, or adipose tissue taken soon after parturition. Until recently, this approach seemed reasonable for POCs. However, Bloom et al. (2007) showed that POC levels vary across pregnancy, which indicates that the timing of the POC sample collection should be reconsidered.

The research plan further mentions that POCs will be measured in urine, but urine is traditionally not used as a medium to measure POCs. The study needs to make clear how such measurements will be conducted.

The statement on page 9-4 of the plan (NCS Research Plan, Vol. 1) that “the purpose of testing at multiple time points is to determine when the exposure(s) occurred” is not entirely relevant for POCs. It is difficult to determine when exposure occurred for POCs unless there is a single acute high-exposure source. The paragraph goes on to state that dust samples will be collected from homes to help distinguish between past high-level exposures and recent lower level exposures, but it is not clear how the use of dust samples will accomplish this. Furthermore, it is likely for POCs that there are low-level, continuous exposures rather than past instances of high-level exposures. The research plan needs to provide further specificity on the assessment of POC exposure, including the use of dust samples and the timing of collection.

In Table 9-1 (NCS Research Plan, Vol. 1, p. 9-5), modern chemicals, such as parabens and UV sunscreens, need to be added to the list for analysis of archival samples. These chemical classes may alter endocrine signaling.

Assessing the use of clothing treated with flame-retardant chemicals by questionnaire, as intended, is likely to be difficult, since parents may not know which clothing articles have been treated. In addition, there are dozens of flame retardants, and it is therefore difficult to determine which specific ones are used in specific articles of clothing. It would be useful if there were pictures or product names to help the parents identify the relevant articles of clothing. Given the high prevalence of exposure to brominated flame retardants among the U.S. population, identifying sources of exposure is a worthwhile aim of the NCS.

Nonpersistent Organic Compounds

The NCS plans to measure volatile organic chemicals (VOCs) from samples of soil, water, dust, and air. Water collections are to be made at homes served by a community water supply. It is not clear whether samples will be taken from homes served by wells. If not, there may be missing data related to potential sources of exposures. For example, well water is more likely to be from rural homes with different exposure patterns for pesticides compared with city dwellings.

For semivolatile organic chemicals and nonpersistent nonvolatile organic chemicals, the research plan recognizes that these compounds are rapidly metabolized and therefore multiple measurements are needed. However, the specific timing of sample collection, the number of samples collected, and the statistical methods used to analyze these results need to be clarified. Furthermore, it is not clear how these multiple (repeated) exposure data points will be incorporated into the study and used in the testing of the priority hypotheses.

The NCS presents a unique opportunity to improve understanding of the use of biological markers to estimate historical exposure for nonpersistent chemicals. For example, much concern exists regarding the potential health effects of organophosphate pesticides and bisphenol A, chemicals with short half-lives. However, there is limited information about whether prior exposure to such toxicants can be estimated by identifying signatures using “omics”2 or through the development of novel technological exposure assessment approaches. This opportunity deserves consideration in the design and execution of the NCS.

For nonpersistent pesticide analysis, to minimize failure to detect exposures, the NCS plans to collect environmental samples only when a recent pesticide application was reported or in agricultural areas. It is not clear how “recent application” will be defined (e.g., days, weeks). The definition of “recent” should be partially dependent on the specific pesticide used, since pesticide half-lives vary in the home environment. The NCS notes that water and soil samples for pesticides will be collected only in rural areas, yet there are data showing high levels of indoor pesticides in urban areas, specifically in public housing and apartment complexes due to the use of insecticides for pest control.

The research plan states (NCS Research Plan, Vol. 1, p. 9-7) that the parent will answer questions on the use of teething rings and pacifiers by the child. It will be important to collect product names, since the presence and percentage of chemicals, such as phthalates, present in these products may vary widely.

Air Pollutants

The research plan states that personal sampling, using a portable monitor, detector tube, or other technology, is burdensome, and therefore indoor air sampling will be performed to measure exposures to various pollutants. However, personal sampling data are usually superior to area sampling data for assessing individual exposures. Statistical methods could be used to relate indoor and personal sampling results from a subsample of subjects, and the results could be applied to the larger cohort. The NCS plans to collect particulate matter less than 10 microns. Currently, there is much more interest in particulate matter less than 2.5 microns and in ultrafine particulate matter, because these smaller particles are more strongly associated with risk in air pollution studies, so air samples collected in the NCS need to allow for a 2.5 micron and ultrafine cut point measurement.

The research plan states that biomonitoring has a limited role for assessing exposure to air pollutants. Yet urine samples are often used to measure exposure to some of the metals found in ambient particles, and an aliquot of urine could be archived for potential measurement of metals relevant to ambient air pollution exposure.

More generally, the NCS needs to reconsider its plan not to collect biological samples for measuring VOCs and other pollutants, even though most media (expired air and urine) are easy to collect and future analytical needs may outweigh the time for collection. Granted such biospecimens have short half-lives, but they are relevant for assessing personal exposure.

The NCS should consider incorporating the collection of novel biological media that may be of potential value for exposure assessment. These data include meconium and the shedding of primary teeth. The former has been demonstrated to be of potential value for measuring organophosphate exposure, and the latter has been well demonstrated to be of significant value for estimating cumulative exposure to lead in children.

Finally, there are few details or protocols for data and biological sample collection from fathers. However, there is a growing literature identifying paternal exposures as of potential importance to offspring, possibly through epigenetic mechanisms (e.g., the impact of paternal age on the risk of offspring schizophrenia). It is recommended that the NCS consider collection of biological samples and data from fathers, when appropriate.

Other Concerns

Recent experimental evidence suggests that there are transgenerational effects of some modern chemicals, possibly through epigenetic reprogramming. Such considerations raise the question as to whether the full effects of some chemical exposures can be detected in the proposed time frame of the NCS. The development of more than a two-generation cohort (parent-child) is beyond the scope of the NCS, but investigators in the NCS could consider the possibility of following subsamples of participating children into their adult years so that a multigenerational study would be feasible (for example, mother-child-grandchild), if results from the current study warrant such assessments.

Projected response rates for the NCS are only 75 percent at the stage of enrollment. Because nonparticipation in the study might be related to characteristics that predict environmental chemical exposures, the NCS needs to develop methods to measure such relationships and their effects on analysis and to make appropriate adjustments to the data (see Chapter 4).

There are important ethical and methodological issues involved in the issue of whether and how to report environmental chemical exposure results to study participants (see Chapter 5). Some exposure measures are currently not interpretable as to the level of potential risk. For some exposures, it is not even clear what is considered high or low exposure. Other exposures, such as to high lead or mercury levels, are known to carry increased risks of adverse health effects. If study participants are told of such results, one hopes that they would alter their lifestyle, home environment, or diet to reduce exposure and thus potential risk. The NCS needs to develop detailed plans for reporting results to participants and for taking account of any consequent changes in behavior that reduce exposures.


Our discussion has identified a range of issues for the NCS program staff to consider. The most important area to address concerns the potential of biological samples for assessment of chemical exposures.

Recommendation 3-9: The NCS should consider the use of personal air sampling methods for a subsample of participating women and their children for measuring exposure to air pollutants.

Recommendation 3-10: The NCS should incorporate methodology to measure paternal exposure to environmental chemicals (both persistent and nonpersistent). More generally, the NCS should consider collecting for fathers, not only chemical exposures, but also biological samples and interview data on paternal characteristics that may affect children’s health and development to the same degree as it collects such information for mothers.

The collection of biological samples from even a subsample of fathers would enrich the NCS study aims and provide the opportunity to assess paternal contributions to children’s health.

Physical Exposure Measures

The NCS research plan recognizes the influence of the physical environment, in the form of housing and neighborhood characteristics and conditions, on child risks and outcomes as children age. Early measurement properly focuses heavily on the home environment, given that preschool-age children spend the majority of their time in the home. Key housing dimensions include the physical configuration and condition of the residential structure, which would be measured with a combination of respondent questions and interviewer observation. Other measures to be collected include household appliances, such as a vacuum cleaner, clothes dryer, and vaporizer (NCS Research Plan, Vol. 2, App. H, p. H-1). Community measures include visual assessments of the neighborhood in terms of type and condition of housing and presence of industries, businesses, recreational areas, traffic, and graffiti; geographic information systems data on traffic, nearest hospitals, recreational facilities, and Superfund and brownfield sites; aerial photography of topographical features; and census population density.

Methodological Concerns

Currently, the NCS research plan fails to acknowledge any critical or sensitive periods during which children’s health and development may be most vulnerable to exposure to specific housing and community factors. Hypotheses regarding such periods would provide opportunities to test for their presence and effects. Given that data collection burden is appropriately emphasized as a concern, the specific measures being proposed should be linked to the developmental stage of the child rather than collecting all measures for all children of a specified age. Greater precision in the data-gathering would allow more in-depth measures to be collected at developmentally appropriate assessment points and may eliminate the need for other measures.

The indicators and the frequency at which they are measured appear not to capture the dynamic nature of some of the home or community exposures. The research plan appropriately notes that many structural features do not change frequently (provided families do not move), yet other exposures may vary daily or weekly, and capturing information at several points over a short period of time versus at a single point in time is state of the art but is not mentioned in the plan. A stronger rationale for some of the listed factors (for example, location of the garage, outdoor additions) should be provided, as their relevance is questionable.

Given the wide variety of housing and community conditions to be encountered, it is not clear whether the existing measures being proposed for use will be valid across types of settings and housing. No mention is made of using the Vanguard Centers to further validate or refine these measures.

It is not clear how physical exposures are going to be modeled in the statistical analyses. For example, how will data from aerial photography on rivers and mountains be used to create indicators of exposure? The links between sources of data on physical exposures, creation of indicators of exposure, and the hypotheses for child outcomes need considerable strengthening.

While turning to readily accessible information as a source of data on community characteristics is a cost-effective strategy, careful attention should be given to how such data are used to construct indicators of neighborhood features. Issues to be resolved include using single factors versus composite indices, creating concurrent versus dynamic historical profiles of the neighborhood, and choosing the appropriate unit of analysis at the community level (for example, census tracts, administrative neighborhoods, etc.). Moreover, a good definition of neighborhood should be provided, given the interest in various types of exposures at the community level. Each of these issues should be considered in light of reformulated and refined hypotheses about the role of community factors in shaping risks and adverse outcomes.


Recommendation 3-11: The NCS should provide a clearer rationale for some of the housing and neighborhood conditions it proposes to measure and revisit its data collection plans to ensure that needed measures are obtained at developmental stages when children may be more vulnerable to risk factors. The goal should be a set of measures and data collection plans that are optimal with regard to analytic utility and response burden.

Psychosocial Exposure Measures

The NCS research plan proposes to collect information on an exceedingly diverse set of psychosocial exposure measures, including family process measures, such as the home environment and domestic violence; maternal mental health measures, such as depression; psychosocial stress measures, such as perceived chronic stress; social support; health behaviors, such as smoking; and features of child care. At the same time the research plan reflects a relative lack of attention to fathers, family structures and transitions among structures, and other significant support figures. The latter may be particularly important in enhancing resilience in the face of adverse family characteristics. Each of these domains has a robust research literature providing justification for literally hundreds of possible measures of psychosocial influences on child health and development. The NCS cannot possibly cover even a fraction of these possible subdomains, and rarely can it afford to allocate enough interviewing or observation time to collect the highest quality measures in any of them.

Overall, however, the research plan does not provide a clear justification for its implicit set of decisions across the daunting set of psychosocial measures to collect. It is clearly not appropriate for reviewers of the research plan to merely suggest that the NCS expand to collect measures of neglected subdomains, or to devote considerably more interviewing time to obtain gold standard versions of the included domains. The existing plan for data collection already imposes heavy burdens on respondent time and attention. Rather we call for a careful elucidation of the choices made with specific reference to goals, hypotheses, and relevant outcome measures.

Relating Exposure Measures to Outcome Measures

The panel thinks that the best way to make decisions about what psychosocial exposure methods to collect rests on a refinement and focusing of the domains of interest in terms of outcomes. For example, if conduct disorders were to become a major focus, one would want to assess disciplinary interactions at key points through the life span (see, e.g., Patterson and Forgatch, 1995). In addition, one would want to tackle the tricky problem of obtaining reliable information on family criminal histories and violence in the home and, beginning at least in middle school, detailed information on peer associates. Decisions about where to invest respondent time and study dollars need to be made in light of decisions about what kinds of variation and what problem outcomes the study wishes to focus on.

Obtaining High-Quality Measures

A general observation about the psychosocial measures concerns the frequency and type of planned data collection. It is striking that there are no plans for an in-home interview between age 1 and at least age 5. This gap precludes accurate measurement of such key elements of the psychosocial environment as the home learning environment during a key phase of development. To obtain high-quality assessments, the NCS needs to conduct more frequent home-based interviews, performed by trained interviewers, at least for a subsample (see Chapter 2).

Given its time constraints, we support the choice of the relatively abbreviated Infant/Toddler Home Observation for Measurement of the Environment (HOME) scales for assessing the physical, emotional, and learning environment. We question, however, that the study plans to invest in a videotaped assessment of mother-infant interaction at 6 months, rather than a high-quality assessment of attachment at 12 or 18 months, such as the Strange Situation, which is feasible to administer (Ainsworth and Bell, 1970).

Finally, we note that, as part of its contribution to the development of state-of-the-art methods, the NCS could engage in the development of high-quality but quick-to-administer measures of psychosocial exposures. The psychological, social, and behavioral sciences have developed reliable, valid, and theoretically informed measures for assessing psychosocial exposures and outcomes. However, at the present time, many of these assessments are lengthy and costly to administer. They demand coding of videotaped interaction, use of laboratory-based equipment, or the like. Advances in knowledge frequently depend on addressing these limitations. For example, a MacArthur Network developed a parental assessment of language development that has led to an explosion of knowledge (compared with transcription of audiotape), and the Early Childhood Longitudinal Study developed a short form of the Wechsler Preschool and Primary Scale of Intelligence (WPPSI). Another successful example is the HOME scales for assessing the physical, emotional, and learning environment, which the research plan includes. The NCS, with its long frame of observation (over 21 years) and large sample, is well suited for this type of methodological research conducted through substudies and adjunct studies. The NCS, in turn, can benefit from the development of high-quality, easily administered assessments that can be used to replicate its findings in other studies.


Recommendation 3-12: The NCS should reconsider its psychosocial measures to ensure that they will provide high-quality data for outcomes of interest for child health and development. In the face of resource and respondent burden constraints, the NCS should emphasize the quality and analytic utility of information, even if some measures must be dropped in order to substitute other assessments more desirable on various grounds.

Recommendation 3-13: The NCS should dedicate a portion of funds to support research and development of reliable and valid instruments of key psychosocial measures that are practical and economical to administer.

Biological Exposure Measures

The NCS investigators plan to obtain measures of biological exposures at several points in time. The measures fall into six main categories: allergens, markers of infection/inflammation, maternal glucose metabolism, endocrine markers, parental medical history, and other health behaviors. Most are to be obtained through a combination of questionnaire and direct measurement.

The panel of allergens includes those related to cats, dogs, mice, rats, cockroaches, and mites, as well as a panel of molds and pollen to be obtained from regional ambient monitoring. Infections are generally to be obtained by a medical history provided by the parent with ascertainment of a variety of inflammatory markers. Maternal glucose metabolism during pregnancy is to be obtained from studies done as part of clinical care and obtaining a hemoglobin A1c, a measure of the average glucose for several weeks. The endocrine markers of interest include maternal thyroid function and cortisol as a measure of stress. Cortisol measures will be obtained from the mother and child on several occasions, and exposure to stress will also be ascertained by questionnaire. Parental history will include attention to the presence of chronic illness, diet, physical activity, use of tobacco and alcohol, use of illicit drugs, and use of medications and supplements. Other health habits to be ascertained include dental health (by questionnaire), maternal sleep patterns, and maternal douching.

This list of potential markers of biological exposure is comprehensive. Several, such as diet, allergens, and maternal physical activity, are to be obtained by well-standardized methods. The use of standard and well-tested approaches to obtaining data on these biological exposures is a strength of the study.

Methodological Concerns

Less clear is whether plans to obtain other exposure measures are as state of the art. Information on child infections is to be obtained primarily by maternal history. The recall period for this information (6 months in the first year) is greater than that generally considered desirable to produce accurate data on relatively brief illnesses. Moreover, it is unclear that mothers will be able to report accurately about the type of infecting organism, information that could prove vital to analysis of subsequent growth and development. The timing of the measurement of inflammation relative to the infection that may have caused it is also not clear, raising questions about what types of infection/inflammation may be missed.

Interest in maternal glucose metabolism in pregnancy is driven primarily by its potential effect on birth defects and child obesity. However, the planned measure of glucose metabolism during pregnancy will be an average measure over some weeks that is not obtained close to the onset of pregnancy, when organogenesis is occurring. The research plan indicates that the study did not think more active measures of glucose metabolism could be obtained given that the first data collection point during pregnancy is a home visit, but it does not speculate on the effect that decision might have on the testing of core hypotheses.

Likewise, the research plan notes that the only endocrine measures of interest are maternal thyroid function during pregnancy and cortisol levels in the mother and infant at various times. With regard to the former, it should be noted that child hypothyroidism may be a mechanism for the linkage between hormonally active agents and reproductive outcomes, so it is not clear why only maternal thyroid function is of interest. It would have been helpful in the discussion of cortisol levels to have some sense of the time over which cortisol will represent an accurate marker of stress in order to assess whether the frequency of data collection will be adequate.

The parental medical history seems quite comprehensive. In conjunction with measures of maternal physical activity, there are plans to estimate infant physical activity at 6 and 12 months through developmental observations and questions about usual activities. It is not clear what significance infant physical activity at this age has, whether developmental observations can actually characterize infant activity, or whether questions about usual activities provide information that is comparable to concerns about parental activity.

Other health behaviors are to be elicited by questionnaire. Ascertaining intensity and frequency of intake of alcohol, and even more so of illegal drugs, from interviews with pregnant women is potentially problematic. In particular, exposures are likely to be underestimated. It is troubling to see little if any discussion of how this potential threat will be addressed. Simply promising anonymity may not suffice. One plan that seems particularly problematic is ascertaining dental health through questionnaire and not examination. With the poor coverage of dental health care in the United States, it is likely that many dental problems will go unrecognized by the parent. To the extent that dental disease, such as peridontitis, is a risk factor for prematurity and other child outcomes, inaccurate data will make it more difficult to establish the link.

A serious concern is that there does not appear to be a conceptual framework guiding the selection of the various biological markers of exposure. In particular, it is not clear that all aspects of parental health that might be relevant to people’s ability to be effective parents will be ascertained.


Most important among the concerns we have raised about biological exposures is the timing of data collection.

Recommendation 3-14: The NCS should review some of the proposed measures of biological exposures, such as maternal glucose metabolism and child cortisol levels, to ensure that the proposed times for data collection are appropriate for capturing the underlying exposure.

Genetic Measures

Each project outlined in the NCS research plan includes investigation of how genetic variation contributes to variation in risk of the study’s key outcomes, ranging from childhood obesity to neurobehavioral traits. With advances in high-throughput genotyping technologies, it is now possible to directly measure hundreds of differences in particular genes and millions of mutations in the whole genome of large numbers of individuals to elucidate the genetic contributions to a human trait or disease. In general, human variability in any trait arises from a complex interaction among genetic variations and environmental variations. The association studies proposed by the NCS are currently the most efficient strategy to explore the putative contribution of genetic variations or gene-environment interactions to variations in disease risk.

The basic design and analysis principles of genetic association studies have been well established for decades, albeit with continual evolution in such areas as study designs (e.g., family-based association studies, case-only designs); genotyping (e.g., multiplexing assays, array-based genotyping); dealing with underlying genetic confounders, such as variations in population groups studied (Haines and Pericak-Vance, 1998); and reducing the probability of false positive results (e.g., Benjamini and Hochberg, 1995). All of these issues appear to be adequately addressed in the research plan. There are, however, several major weaknesses in the plan proposal that should be addressed before the genetic component of the project is actualized.

Changing Science and Scientific Standards

The field of genetics and genomics is changing rapidly. Just within the past year, dozens of genome-wide association studies have identified new genes and variations that are involved in such complex traits as blood glucose levels, obesity, height, and variation in common disease risk (English and Butte, 2007; Fox et al., 2007; Hayes et al., 2007; Peeters et al., 2007). In many of the NCS proposed analyses, in contrast, the old approach of studying “established” candidate genes (i.e., genes that are thought to be involved in a disease or trait because of the knowledge of underlying biological pathways) is put forward and is a major weakness of the proposed genetic studies.

The NCS research plan does discuss in several places the use of new technologies, namely, gene expression profiling and epigenetic profiling. These new methods may be of some use in unraveling the potential molecular mechanisms underlying genetic associations. However, in the NCS, these biological signatures will be measured on tissues—namely, components of blood—that may not be relevant to the trait or disease being investigated. For example, neurodevelopmental outcomes are not likely to be associated with transcriptomic variation in lymphocytes, which will be the only biological tissue available for study. Great care must be exercised in making inferences from these transcriptomic and epigenomic types of studies because they are, in many cases, studies of convenience rather than studies designed for their scientific rigor. Although it may be convenient to measure gene expression or epigenetic changes in blood samples, there is very little evidence that the gene expression profiles in this tissue are biologically relevant to the neurobehavioral outcomes or other outcomes investigated in the NCS.

The scientific standards for genetic association studies are also quickly changing, as scientists come to grips with the limited success of the past two decades of genetics research, which has failed to identify the key genetic factors with reproducible or replicable effects on common human diseases or traits. The previous lack of high scientific standards for publication has resulted in the dissemination of false information (including false positive results from genome-wide association studies), the waste of millions of taxpayer dollars, and an increase in genetic deterministic thinking among the public. Three systematic studies of the genetic association literature have documented the extent of the problem (Hirschhorn et al., 1999; Ioannidis et al., 2001; Lohmueller et al., 2003), in which the odds of a published genetic association being replicated when 3-4 other studies were conducted were approximately 1 in 30. Moreover, Lohmueller et al. (2003) found that less than half of what are considered established susceptibility genetic markers pass the standards of a rigorous statistical meta-analysis. Cited reasons for the lack of reproducibility include genetic and environmental differences across the populations being studied, low statistical power, and misclassification of disease outcomes (e.g., Cardon and Bell, 2001; Colhoun et al., 2003; Freedman, Reich, Penney, et al., 2004).

In the current NCS research plan, the strategy for investigating genetic associations appears to lack an appreciation of the more rigorous standards now being imposed by the scientific community. There are multiple reasons for adopting a high standard in the NCS by which genetic association studies must be internally and, optimally, externally validated before any type of publication or media release. First, the field of genetics research is finally imposing its own higher standards. Second, given the powerful implications of genetic information (for example, stigmatization and discrimination) for children if a genetic marker of a trait, say ADHD, is identified in the NCS, there must be a mechanism in place for validation at both the population and molecular levels to avoid the reporting of false results, many of which have already flooded the literature and mass media from other sources.

Gene-Environment Interactions

A major strength of the NCS research plan is its emphasis on gene-environment interactions. However, the implications of this emphasis for the measurement of exposures are not fully appreciated. The lack of adequate measures of psychosocial and behavioral variables is particularly important, for it severely limits the ability to examine gene-environment interactions that are likely to affect obesity, neurobehavioral phenotypes, asthma, and pregnancy outcomes (Institute of Medicine, 2006).

The genetic measures proposed in most analyses are quite limited. For each of the major outcome areas, genetic associations are proposed based on previous candidate gene association studies, many of which would not pass more modern standards of evidence. Furthermore, the variation within these genes is not well represented or discussed in the research plan. Most of the genetic variations targeted for study are not considered causative, have no known biological function, and are simply markers of the effect of a currently unknown genetic variation. Thus, more variations within each gene need to be considered if the true goal is to understand the genes responsible for modifying the effects of environmental exposures.

Another weakness in the research plan is the lack of detail on how genetic variation will be measured. There are considerable cost considerations, which are not dealt with adequately. For example, it can be as expensive to genotype 20 single nucleotide polymorphisms (one at a time) as it is to genotype thousands (or millions) of them with array-based genotyping platforms. Given the size and scope of the NCS, there is also the need to consider gene-environment correlations in analyses, in addition to gene-environment interactions.

Data Collection Concerns

Because the DNA sequence variation proposed to be investigated in the NCS genetic association studies does not change over time, the one- time collection of a blood sample for genetic analysis (summarized in NCS Research Plan, Vol. 2, App. G) will be adequate for the genetic analyses proposed. However, unlike DNA variation, gene expression and epigenetic variation will dynamically shift over the life course of the mother, father, and child. They are also tissue-specific. Currently, there is no proposed collection of blood samples for studies of gene expression profiling of parents. In addition, only a single cord blood sample from the child is being set aside for transcriptomic and epigenomic studies. This is a major oversight and an inconsistency between the neurodevelopment analysis plan (NCS Research Plan, Vol. 2, App. A-2) and the biospecimen collection plan (App. G). Although this oversight can be easily remedied by collecting and storing additional blood samples for gene expression studies at other time periods, it must be actively addressed or this opportunity will be missed. As we cautioned above, the NCS must be aware that examining gene expression and epigenetic patterns in blood may not be relevant or appropriate to study with such outcomes as the neurodevelopment outcomes. Great care must be taken, since false positives are much more likely than true positives in these domains.

Appropriateness of the NCS for Genetics Research

The greatest potential benefit of the NCS to the field of genetics is its linkage between prospective environmental exposure data and high-quality, high-density genetics data. The field of gene-environment interaction has very few examples in which prospective environmental exposure data are linked to genetic susceptibility data. Virtually all genetic association study designs are either family-based or case-control designs, because DNA sequence is static and can be measured at any time. The most difficult aspect of a genetics study is finding probands and conducting sound phenotyping. Since it is assumed that genotype precedes phenotype, genetic studies nearly always focus on phenotyping first and genotyping second.

The value of the NCS is that it can solve both problems—that is, some genetic studies can be conducted prospectively in this NCS cohort, and environmental studies on the same disease will have good unbiased, exposure data. This is a unique opportunity for high-quality gene-environment interaction research. After 22 years, there will be sufficient cases for several diseases (autism, ADHD, asthma) to conduct nested case-control GXE interaction studies with prospective environmental data. This is of critical value. The concept of critical developmental windows indicates that the timing of exposure matters a great deal. In either case-control or family-based association designs, sampling is dependent on case designation. With either design, environmental exposure is nearly always either measured cross-sectionally or retrospectively. Exposures during developmental win dows cannot be reconstructed retrospectively. Case-control designs for gene-environment interaction ignore what is known about gene expression changes in development. Genes are not always active throughout each life stage or may be more or less active during different life stages (e.g., hormones and adolescence). Logically, a chemical or other environmental exposure that modifies a genetic variant may only do so when the exposure corresponds with a developmental stage. Unless the exposure dose and the timing of exposure are matched to the genetic variant, critically important gene-environment interactions may be missed. This is the greatest value of the NCS: Not only will it provide measures of environmental exposures of interest, but it will also provide their timing and relationship to genetic variation. This is not well articulated in the research plan, yet it is the study’s greatest strength in this field.


Recommendation 3-15: The NCS should adopt a clear mechanism by which genetic association studies are internally and, optimally, externally validated before any results are published or released to the media. The NCS should also revise its proposed “established” candidate gene approach to take advantage of the new information emanating from the current wave of genome-wide association studies, with appropriate replication.

Recommendation 3-16: The NCS should consider consolidating its genetic studies in order to reduce costs and to coordinate the best science at the least cost to the project. One approach would be to simply collect the biological samples and properly store them for later genetic analysis when a better selection of polymorphisms and cost-effective genotyping across studies are possible.

Missing Exposures

Access to and Quality of Services

Clearly, even such a large study as the NCS is limited in the data that can be collected. A discussion of what exposures were considered and rejected and why would have been useful for the panel in its review. In particular, a notable omission is information on access to services, especially health services, both as potential mediators of outcomes and as factors in the accuracy of information obtained through maternal report.

A substantial amount of information—such as diagnoses of child health problems—is to be ascertained through interviews with the mother, but children and mothers do not have equal access to health care providers, teachers, or day care workers who may be capable of making the correct diagnoses. Furthermore, a mother may not be able to report reasonably accurately what she has been told. Unequal access to services or access to lower quality services will lead to biased ascertainment of outcomes. For example, in the 1980s, Newacheck, Budetti, and McManus (1984; Newacheck, Budetti, and Halfon, 1986) examined the doubling since 1960 of the proportion of children reported to have chronic conditions on the National Health Interview Survey. They concluded that much of this doubling did not reflect major shifts in the population, such as deinstitutionalization. Rather, it reflected changes in attitudes and perhaps diagnostic practices among clinicians, and it may have included better ascertainment resulting from better access to medical care for disenfranchised populations through programs like Medicaid. Similarly, diagnoses of learning difficulties may differ depending on the availability of services in the school, teacher demand, and because of changes in the conditions for which special education services may be received. Furthermore, much of the observed increase in diagnoses of autism reflects changes in the special education categories and more awareness of the spectrum (Croen, Grether, Hoogstrate, and Selvin, 2002).

In addition, access to and the quality of services may actually affect outcomes. Interinstitutional variation in the quality of medical and day care and its effect on outcomes are now well established. For example, the NICHD Study of Child Care and Youth Development documented differences in outcomes with higher or lower quality of day care, while Vohr et al. (2004) documented an almost fivefold difference in the rates of cerebral palsy and mental retardation among infants born at 1,000 grams across 12 neonatal intensive care units. The NCS research plan proposes to pay some attention to child care, schools, and religious institutions in the revised section on child health and development, but it is not clear that attention to such issues will also inform other outcomes (e.g., schools as a source of environmental exposures for asthma and as a factor in diets leading to obesity). Clearly, access to early intervention services might alter developmental trajectories, but such services do not seem to be listed for investigation.

Only for asthma does there appear to be a concern about studying the effect of access to medical care. Yet there are many conditions for which variations in developmental screening and other procedures will lead to differential diagnoses and referral for services. For some conditions, like autism, earlier intervention has clearly led to improved outcomes. The NCS does not propose to investigate such factors, despite the well-documented substantial variation that occurs in the services children receive (Mangione-Smith et al., 2007).


Recommendation 3-17: The NCS should add measures of access to and quality of services, including medical care, education, child care, and services, as potential mediators of health and development outcomes and to improve the assessment of information obtained through maternal reports.

Policy Environment

The NCS research plan pays little attention to the implications of policy analysis for the study design. A strong analytic design for policy analysis is the use of “natural experiments,” which take advantage of variations in policies across states or communities and across time. An example is the Currie and Gruber (1996a, 1996b) analysis of the impacts of the state expansions of the Medicaid program on use of medical care and birth outcomes. Because states expanded their Medicaid programs in different ways and at different times, the analysis was able to compare outcomes before and after the expansions in each state and control for time effects with the variation in the timing of state implementation. Since interesting policy variation will arise in states, counties, cities, and even neighborhoods, analysts conducting these kinds of policy analyses will need to be able to match detailed information on policies to the geographic location of respondents.

Additional Exposures That Could Be Studied Through Data Linkage

A major potential benefit of the NCS household-based sample design is that a wide range of exposure measures can be obtained by matching existing and future sources of environmental information to respondents’ residential addresses. For example, neighborhood demographic and socioeconomic data drawn from the decennial census and often aggregated to the level of the census tract have long been linked in research studies to individuals and households. With the advent of the American Community Survey, which replaces the decennial census long-form sample, demographic and socioeconomic neighborhood information can be linked to individual data more frequently than once a decade (National Research Council, 2007). Crime data from the FBI’s Uniform Crime Reporting system can be used to characterize community crime. Environmental data from the Environmental Protection Agency monitoring stations can provide a range of useful measures of environmental pollutants. Data on the policy environment for safety net programs can be derived, for example, from state- or county-specific rules for such programs as Temporary Assistance to Needy Families. Data on weather conditions, commercial activity, and access to public transportation may all prove useful in characterizing maternal and child exposures. An enormous advantage of these kinds of measures of neighborhood exposures is that they can be added to the NCS database without increasing respondent burden, provided they can be linked to sample members’ residential addresses and, for some exposures, dates of interview.

While the research plan acknowledges these kinds of linkages, it does not provide a thoughtful discussion of the steps that need to be taken to optimize NCS linkages to other sources of environmental information. A first priority is to geocode (that is, characterize with standardized measurements of location, such as census block, census tract, city, county, state) all of the residential addresses in which sample children reside over the course of the study. Then a wealth of information stored in other geocoded databases will potentially be available to link to the NCS data. How much residentially linked information the study intends to gather itself and under what conditions geocoded data will be available to analysts to perform their own linkages are not clear.

Given the fundamental importance of a full array of exposure data for testing many of the key study hypotheses and for testing new hypotheses that arise over the course of the study, it is vital that researchers—both inside and outside the designated study centers—be able to access information about all respondents’ geographic locations. Such access raises important confidentiality concerns, but these concerns have been met in other national studies, such as the National Longitudinal Survey of Adolescent Health and even Census Bureau data sets, through a variety of mechanisms. One of the most promising for the NCS would appear to be the network of Census Research Data Centers that have been established for researcher use of various census and other governmental sources of information (see Chapter 5).

Turning from the matching tasks to the environmental data themselves, it is important to realize that sources of exposure information are a vital public good for the NCS. The study should encourage researchers, with some combination of internal and adjunct study funding, to compile local exposure data that can be matched to the residential locations of all NCS respondents and to make these data available to all analysts. Westat’s data repository would appear to be the logical place in which these geographic data and their documentation would be stored.

Recommendation 3-18: To facilitate linkages of NCS data with environmental exposures from other databases, such as measures of demographics, crime, government programs, and pollution, the NCS should develop a plan for geocoding the residential addresses from prebirth through adulthood of all participating children to standard census geographic units. In addition, the study should develop arrangements by which researchers, both inside and outside the NCS study centers, can access geocodes for respondent addresses and are encouraged to perform linkages and make their environmental information available to the NCS analysis community. Such arrangements must safeguard the confidentiality of NCS respondents.



We note that the information in Appendix F-1 does not always agree with information in other parts of the research plan.


Omics is a general term for a broad set of disciplines for measuring and analyzing all components interactions from genes to proteins to metabolites.

Copyright © 2008, National Academy of Sciences.
Bookshelf ID: NBK20658


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (784K)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...