The National Children's Study Research Plan: A Review (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

2 NCS Goals, Conceptual Framework, and Core Hypotheses T he National Children’s Study (NCS) was designed to respond to the requirements of the Children’s Health Act of 2000. Design choices were made in accordance with the overarching goals and conceptual framework developed for the study and to reflect the data needs for analy- sis of 28 core or meta-hypotheses (comprising over 100 specific hypoth- eses) that link environmental exposures to child health and developmental outcomes. This chapter provides an overview of the NCS design to orient the reader and convey a sense of the large scale and scope of the effort; it then presents and comments in broad terms on the NCS goals, conceptual framework, and selection of core hypotheses as outlined in the research plan (NCS Research Plan, Vol. 1, Précis, Chs. 1, 2, 4). The chapter ends with a recommendation for lengthening the time between activities of the Vanguard Centers and the corresponding activities of the other study cen- ters. The purpose of this recommendation is to provide adequate time, not only for the NCS to learn from the experience of the Vanguard Centers with regard to study operations, but also to adequately develop key constructs and hypotheses. As a prefatory note to our critique of various aspects of the NCS in this and subsequent chapters, we acknowledge and appreciate the enormous ef- fort that has gone into its development. Its designers, in an environment of uncertain funding and with the need to involve many different agencies and research communities, have exhibited professionalism and strength of pur- pose throughout. They have often confronted difficult choices, given that no study, no matter how complex, could cover every possible exposure and 23 

24 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN outcome of interest in the same depth. Funding would never be sufficient to do that, nor would the burden on study participants be acceptable. We detail instances in which we think the particular choices made for the NCS merit reconsideration or refinement—indeed, earlier work by NCS working groups and advisers often included options that we think should be on the table but did not appear in the final research plan. Nonetheless, we remain cognizant throughout of the resource limitations for the study, as well as the design constraints imposed by some of the decisions needed to bring the NCS to its current state. At the same time, we note—as does the NCS research plan itself—the opportunity afforded by the study’s length to alter course in one manner or another as appropriate, not only to take account of our comments, but also to take advantage of relevant new measurement techniques and research insights. DESIGN IN BRIEF The NCS is designed to be a long-running, observational panel study of a nationally representative probability sample of 100,000 births to be followed from before birth to age 21. Data will be collected on multiple exposures and multiple outcomes with repeated measures over time. Sample Design The sample design (see NCS Research Plan, Vol. 1, Secs. 6.1-6.4) is multistage and based primarily on households (discussed further in Chapter 4). Omitting some details, the first stage (already completed) was to draw a sample of 110 primary sampling units (PSUs) in 105 locations; the PSUs are counties or groups of counties (or parts of counties in Cook County, Illinois; Harris County, Texas; and Los Angeles County, California), each of which is expected to experience a minimum of 2,000 births during a 4-year enrollment period for the study. The second stage (implemented to date in the PSUs covered by the seven Vanguard Centers) will be to draw a sample of segments in each sampled PSU; the segments are census blocks, groups of blocks or similar subdivisions (some PSUs may be other subunits, such as school catchment areas). A third stage, implemented only in very large segments of sampled PSUs, will be to draw a clustered sample of house- hold addresses. These stages of sampling use stratification on geographic and other variables to ensure representation of population groups (such as minorities) and areas; there is no planned oversampling or undersampling by population groups or areas. Finally, all of the household addresses in the sampled segments (or clusters) will be visited to identify women who are eligible to participate; such women include those ages 18-44 who are not pregnant as well as all 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 25 women in the first trimester of pregnancy, regardless of age. Women who are not pregnant will be recontacted several times to determine whether they have become pregnant and hence are eligible to continue in the study, with the intensity of follow-up varying depending on whether a woman is deemed to have a high or a low-to-moderate probability of becoming pregnant. It is hoped that at least 25 percent of all births in the study will be to women who were not pregnant at the time of enrollment, in which case preconception measures would be available, and that 90 percent of all births in the study will be to women who were enrolled before the end of the first trimester of pregnancy (NCS Research Plan, Vol. 1, Sec. 6.1). The final-stage sampling, enrollment of participants, and initiation of data collection will be spread over several waves, beginning in mid-2008 for PSUs covered by the Vanguard Centers, followed by another set of PSUs in mid-2009, a third set in mid-2011, and the final set in mid-2012. During the preparation and initial sampling phases, the 35 to 40 study centers that will ultimately be part of the NCS operation will engage in outreach activi- ties to inform local communities of the study and encourage participation (see NCS Research Plan, Vol. 1, Sec. 6.5, and Chapter 5 of this report). To identify eligible women who moved into the selected segments after the initial screening, providers of prenatal care, birthing centers, and hospitals will be visited in those segments. In each of the 110 PSUs, the enrollment period will extend over 4 years, with the goal of enrolling about 1,000 newly born children by the end of the period. Data Collection The data collection plan (see NCS Research Plan, Vol. 1, Sec. 6.6) is complex and uses multiple collection modes. Details of items to be col- lected, from whom, when, and by which method have been specified in any detail only for ages prior to birth through 24 months. A provisional schedule of contacts with study participants has been outlined for ages 3- 21, but this schedule is subject to change, as are the specific items and data collection methods. The research plan argues that maximum flexibility is needed to enable the NCS to take advantage of new data collection and measurement technologies. Table 2-1 provides summaries of the planned data collection from pre- birth through age 24 months. Locations and modes of collection include personal home visits, telephone calls to homes, and clinic and hospital visits. The mother (or primary caregiver if the biological mother leaves the child after birth) is the main respondent; the biological father or partner may also participate, although relatively little data will be obtained for that person; and, presumably, the child will be interviewed later in the study be- ginning at an appropriate age. Types of data collected at one or more points 

26 TABLE 2-1  Planned NCS Data Collection from Prebirth Through 24 Months After Birth Type and Time of Biologic Clinical/Developmental Environmental Contact Questionnaire and Diary Samples Examination Samples Prepregnancy Maternal contact only (unless otherwise noted) Home visit–interview Demographics Blood Anthropometrics Indoor air Household composition Urine Blood pressure House dust Medication use Saliva Health behaviors Vaginal swabs Housing characteristics Hair Chemical exposures Product use Occupational exposures Diet Phone follow-up Diet ---- ---- ---- Chemical exposures 

During pregnancy First trimester home visit Demographics (also P)* Blood (also P) Anthropometrics (also P) Indoor air Household composition (also P)* Urine (also P) Blood pressure (also P) House dust Medication use* Saliva Fetal ultrasound (from medical Drinking water Health behaviors* Vaginal swabs report or clinic visit) Soil Housing characteristics* Hair (also P) Chemical exposures* Product use* Occupational exposures* Diet* Medical history (also P) Stress and social support Depression Tobacco use (P) Cognition (P) Second trimester phone Major life events ---- ---- ---- follow-up Mental health update Medical update Chemical exposures update Housing update continued 27 

TABLE 2-1  Continued 28 Type and Time of Biologic Clinical/Developmental Environmental Contact Questionnaire and Diary Samples Examination Samples Third trimester clinic Updates on: Blood Anthropometrics Indoor air visit    Demographics Urine Blood pressure House dust    Household composition Saliva Fetal ultrasound (self-collected    Medication use Vaginal swabs and mailed in)    Health behaviors Hair    Housing characteristics    Chemical exposures    Product use    Occupational exposures    Diet    Medical history    Stress and social support    Prenatal life events    Depression From birth on Maternal (M)/child (C)/ paternal (P) contact Birth: At delivery, Health behaviors (M) Blood (M) Anthropometrics (C) hospital Diet (M) Urine (M) Dysmorphology and neurologic ---- Chemical exposures (M) Cord blood exam (C) Plans for infant feeding, sleeping, Placenta and cord Digital photographs of face and etc. samples anomalies (C) Heel stick (C) Chart abstraction (M, C) 

3-month phone call Child care Breast milk (mailed in at Medical update (C) 4-6 weeks) ---- ---- 6-month home visit Stress and social support Urine (C) Anthropometrics (C) Indoor air Family process and parenting Hair (C) Dysmorphology exam and photos House dust practices (also P) Saliva (M, P) (C) Drinking water Health behaviors (M) Breast milk Dermatologic exam (C) Soil Depression and Social development observation Visual assessment cognition (M) (M, C) of house and Diet (C) neighborhood Medical update (C) Medication use (C) Media exposure (C) Child care Chemical exposures Temperament (C) Tobacco use (P) Cognition(P) 9-month phone call Child care Medical update (C) ---- ---- ---- Housing update Chemical and occupational exposures (M, C) continued 29 

TABLE 2-1  Continued 30 Type and Time of Biologic Clinical/Developmental Environmental Contact Questionnaire and Diary Samples Examination Samples 12-month home visit Household composition update Blood (C) Anthropometrics Indoor air Family process and parenting Urine (C) Blood pressure House dust practices (also P) Hair (C) Dermatologic exam Drinking water Health behaviors (M) Saliva (C) Cognitive exam Soil Diet (C) Breast milk Motor and language assessments Visual assessment Medical update (C) Social development of house and Medication use (C) observation (child and neighborhood Media exposure (C) father, if available) Noise survey Child care Housing update Chemical and occupational exposures (M, C) Language acquisition and social interaction (C) Tobacco use Cognition (if not assessed at first trimester meeting) 18-month phone call Child care Medical update (C) ---- ---- ---- Diet (C) Housing update Chemical and occupational exposures (M, C) 

24-month phone call Child care Indoor air Medical update (C) ---- ---- House dust Housing update (self-collected Chemical and occupational and mailed in) exposures (M, C) Life events (M) Note: P = paternal, M = maternal, C = child. *Updates if in prepregnancy cohort. Source: NCS Research Plan, Vol. 1, Tables 6-2 through 6-4. 31 

32 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN in time include biological specimens (blood, urine, hair, saliva, breast milk), results of physical examinations (for example, anthropometrics, blood pres- sure), residential environmental samples (air, dust, drinking water, soil), medical records, responses to personal and telephone interview questions, mailed-back questionnaires, diaries or logs of several types, and interviewer observations. Topics covered in interviews at one or more points in time in- clude household composition; housing characteristics; maternal (or primary caregiver) characteristics (demographics, medication use, health behaviors, chemical exposures, product use, diet, medical history, stress and social support, depression and cognition, occupational exposures, child care); paternal (or partner) characteristics (demographics, tobacco use, medical history, cognition, mental health, parenting practice); and child character- istics (medical history, diet, medication use, media exposure, temperament, chemical and occupational exposures). There is also the potential to ob- tain measures of neighborhood characteristics from auxiliary data sources linked to a participant’s census block group. Following data collection at age 24 months by telephone call to the mother, a clinic visit is planned for age 3 years. Another in-home or clinic visit is planned for age 5 years and every few years subsequently—provi- sionally, for ages 7, 9, 12, 16, and 20 (with telephone calls presumably at more frequent intervals), although the location and timing of such visits beginning with the visit for age 5 years are not yet decided (NCS Research Plan, Vol. 1, Table 6-1). Substudies Substudies, outside additions to the core research, and adjunct studies (see NCS Research Plan, Vol. 1, Ch. 16) are included in the research plan to enable the NCS to serve as a platform for additional scientific research. Substudies are defined to include research that is planned in a similar man- ner as the planning for the core research and is supported with NCS funds but applies to only a part of the study cohort. Examples are not given but could presumably include studies of particular population groups with some added data collection. The NCS will also entertain proposals from other agencies, to be funded by the proposing body, for additions to the NCS core research plan. Both substudies and outside additions to the core research must go through the appropriate reviewing process, and approved studies will be included in the workload of all NCS study centers. Adjunct studies are defined as studies proposed by other agencies, with outside funding, to be conducted at only one or a subset of study centers on a portion of the cohort. Adjunct studies will largely be supported with outside funding, though to a limited extent NCS funds may support some 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 33 adjunct studies. Adjunct studies are implemented only with the concurrence of the involved study centers, so they are optional for the centers. STUDY GOALS Description The overarching goal of the NCS, as stated in the NCS research plan (Vol. 1, Précis, pp. xv-xvi), is: to provide information that will ultimately lead to improvements in the health, development, and well-being of children. The primary aim of the NCS is to investigate the separate and combined effects of environmental exposures (chemical, biological, physical, and psychosocial), as well as gene-environment interactions, on pregnancy outcomes, child health and development, and precursors of adult disease. In addition to this broad purpose, the Study has several more specific goals: (1)  Determine the presence or absence of effects, both harmful and help- ful, related to the timing, frequency, magnitude, and duration of specific chemical, physical, biological, and psychosocial exposures in children’s environments from preconception to adulthood. (2)  Determine possible environmental contributions to, or causes of, spe- cific diseases and conditions of children, including, but not limited to, prematurity and other outcomes of pregnancy, neurological and devel- opmental disorders, psychiatric and behavioral disorders, altered physi- cal development and sexual maturation, obesity and insulin resistance, asthma, and injuries. (3)  Determine how genotypic variation and mechanisms, and the interac- tion of genes with environmental factors, influence disease risk and devel- opmental trajectories in children. (4)  Serve as a national resource for future studies of child health and de- velopment by providing a rich database and repository of environmental and biological samples and information that can be used to address future questions and hypotheses. Overall Assessment We draw one overarching conclusion about the responsiveness of the NCS goals and design, in broad terms, to the requirements of the Children’s Health Act of 2000. Conclusion 2-1: The stated overall and specific goals for the NCS—and the design of the NCS to achieve those goals—largely reflect the stipu- 

34 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN lations of the Children’s Health Act of 2000. In the broadest terms, the NCS goals and design are responsive to the call in the act for a “national longitudinal study of environmental influences (including physical, chemical, biological, and psychosocial) on children’s health and development.” The NCS represents an impressive marshalling of federal agency re- sources and research community input to develop and move toward imple- mentation of the most complex and far-reaching panel study of children ever mounted in the United States. Of necessity, the NCS design is observational rather than experimental; it observes children rather than assigning them randomly to treatments, which could not ethically be done for environmen- tal exposures—for example, children could not be randomly assigned to drink water with or without a known environmental contaminant. A con- sequence is that, while the NCS will often be able to establish associations between exposures and outcomes and in many cases the strength of such associations, it will not generally be able to establish causality. An exception is instances in which the NCS can take advantage of natural experiments (for example, the addition to or deletion of some highly potent substance from the food supply mandated by the U.S. Food and Drug Administration, changes in assistance program eligibility requirements adopted by some states or counties—see Chapter 3). Associational results can be very useful to the extent that they are developed using a well-specified and theoretically grounded model that includes potential confounding factors; estimated with a high degree of precision and robustness to sensitivity tests; based on measures for all vari- ables in the analysis that are conceptually valid and obtained with a high degree of accuracy and precision; and confirmed by the results of other studies using different methods (for example, case-control studies) or other data sources (for example, other national panel surveys using comparable measures). Subsequent chapters discuss the strengths and weaknesses of the NCS observational, prospective cohort study design in detail. Here we draw a conclusion about two major strengths of the NCS design, which are the large, nationally representative, equal probability sample design and the large numbers of variables to be collected on both outcomes and exposures over a long period of time. We draw another conclusion about four overall weaknesses of the NCS design, which are an inadequate plan for under- standing health disparities, inadequate conceptualization of health and development and an overemphasis on disease and impairment, constraints on the study’s ability to measure chronic and intermittent exposures, and challenges to data quality. 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 35 Conclusion 2-2: The large, nationally representative, equal probability sample design, together with the inclusion of a large number of out- come and exposure measures over a long time span, are major strengths of the NCS. In particular, the sample design is an appropriate platform for the study, considering resource constraints, the need to represent all population groups and geographic areas, and the difficulty of devising an alternative disproportionate sampling scheme that would not unduly disadvantage some groups and areas that turn out to be of analytical interest. Conclusion 2-3: In four overarching areas, the NCS design, as repre- sented in the research plan, is not, or may not be, optimal for achieving the goals of the Children’s Health Act. These areas are: • insufficient attention to understanding disparities in child health and development among population groups of children defined by race, ethnicity, language, socioeconomic status, and geo- graphic area, which the act explicitly mandates; • inadequate conceptualization of important constructs, including health and development, and an overemphasis on disease and impairment relative to health and functionality and on risk fac- tors relative to protective health-promoting factors; • impaired data collection schedules and types of measures to sup- port evaluation of some of the effects of chronic and intermittent exposures on child health and development; and • underappreciation of the challenges to obtaining the highest possible quality of data from an observational design, which include the decentralized data collection structure of the study and limitations on the frequency of home and clinic visits and on the collection of medical and other administrative records for study participants. Overall Strengths Sample Design A linchpin of the NCS is the nationally representative sample design, which is important for estimating the incidence and prevalence of specific health and development conditions of concern as the study cohort matures. This design choice ensures that all population groups and geographic ar- eas are represented in the sample, responding to the requirement in the Children’s Health Act to cover “diverse populations of children.” The 

36 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN sample design also uses equal probabilities of selection, which, by defini- tion, means that estimates for larger population groups and geographic areas will be more precise than the corresponding estimates for smaller population groups and areas. The drawback of equal probability sampling is the inability to oversample—and thereby improve the precision of estimates for—small population groups or geographic areas that may be of particular interest in terms of their exposures to certain kinds of environmental risk factors. Given the large number of outcomes and exposure measures of interest for the NCS, however, a design that oversampled some groups and areas might well mean that some undersampled groups and areas, which sub- sequently turned out to be of analytical interest, would have less precise estimates than needed for analysis. (Given a budget constraint that dictates the total number of cases, oversampling some groups must be offset by undersampling other groups, which lowers the precision of estimates for the entire sample as well as the undersampled groups.) Moreover, the very large sample size of 100,000 births planned for the NCS means that many, if not all, small population groups of interest, such as minorities, will have large numbers of sample cases. Thus, African American and Hispanic chil- dren will be adequately represented for most purposes, although smaller minority groups, such as specific groups of Asian children, may not have sufficient sample size for many analyses. On balance, we support the sample design adopted by the NCS given its resource constraints (see Chapter 4), although—as noted below and elsewhere in the report—we encourage the consideration of adjunct studies that provide additional sample for scientifi- cally interesting groups when resources and respondent burdens permit. Breadth of Data Collected If carried out fully as planned, the NCS will provide a massive data- gathering platform, with the advantage of an observational frame that is both long (21 years) and wide (containing many hundreds of variables mea- sured on 100,000 children and their caregivers). This massive frame will be a major asset in identifying relationships among exposures, mediating factors, and outcomes that would otherwise be invisible or obscured and in permitting a wide range of analyses that, over time, can take advantage of new theoretical constructs and analytic methods. The prospective longitudinal design, in which children are followed over the entire course of pregnancy, childhood, and adolescence, should enable a large number of life-cycle relationships between exposures and outcomes to be assessed, many of which could not be investigated in depth before. Some of these relationships are posited in the study’s hypotheses, but others will emerge in ways that cannot now be anticipated. A related strength of the 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 37 study is the effort to recruit births prior to conception and during the first trimester of pregnancy, when some kinds of environmental exposures may prove to be critically important for some kinds of outcomes. The very richness of the data presents analytical challenges, including the risk of asserting “false positive” relationships that appear to be statisti- cally significant but that have in fact been generated by chance. This may be particularly problematic for studies of the association of specific genes with specific outcomes—an area in which reviews of the genetic association literature have documented large numbers of findings that could not be rep- licated in other studies— but also for studies of environmental exposures in which the association may be at the low level (e.g., an odds ratio less than 1.5) (see Chapter 3). Overall Weaknesses Health Disparities The Children’s Health Act called for the proposed longitudinal study to “consider health disparities among children,” and we understand that an early NCS working group considered issues of measuring health and developmental disparities among racial and ethnic groups, although any products from this group were not available on the NCS website. A study such as the NCS has the potential power to shed light on very important issues regarding the mechanisms at work that might explain the associa- tions between low socioeconomic status and poor health and development outcomes. The rationale for including disparities is strong given the large gaps on many health measures among population groups and geographic areas. One example is the much higher rate of low birth weight among chil- dren born to African American mothers compared with other population groups (Federal Interagency Forum on Child and Family Statistics, 2005: 32). Yet neither the NCS statement of goals nor the design, as presented in the research plan, explicitly reflects this mandate. With regard to major design elements: • As noted, the sample is large and the sample design uses equal probability sampling because of the difficulty of specifying an al- ternate design that would be optimal in view of the wide range of outcomes, exposures, population groups, and geographic areas of interest. This choice means, however, that some racial, ethnic, and language minorities may have insufficient sample sizes for some analyses. Moreover, the NCS proposes to accept lower response rates in areas, such as inner cities, that are traditionally hard to 

38 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN survey, which will reduce effective sample sizes for disadvantaged groups relative to other groups (see Chapter 4). • Only a handful of the more than 100 specific hypotheses in the NCS research plan directly addresses disparities, even though many outcomes of interest, such as premature birth, birth defects, obesity, and premature puberty, could benefit from explicit consideration of disparities. One hypothesis considers disparities in asthma pro- pensities by race, ethnicity, and socioeconomic status, and another hypothesis considers disparities in schizophrenia propensities by socioeconomic status. It is not clear, however, that the NCS can adequately assess schizophrenia, given that it does not follow par- ticipants after age 21. Moreover, several hypotheses that relate socioeconomic status, ethnicity, and type of geographic area to broad developmental outcomes are so general as to lack analytical rigor and meaning (see section below on “Core Hypotheses” and Chapter 3). • The NCS data collection plan skimps on information on the avail- ability, use, and quality of medical, educational, and other services for study participants, which may be important mediators in pro- ducing disparate health and development outcomes (see Chapter 3). This is a consequential omission because of known differences in access to and use of services and in attitudes toward service provid- ers among racial, ethnic, language, and socioeconomic groups. For example, disadvantaged minorities, immigrants, and poor families compared with other groups disproportionately live in areas that lack medical and other services and disproportionately distrust medical and other service providers (Institute of Medicine, 2003). Remedying the lack of attention in the NCS to understanding the reasons for the large observed disparities in child health outcomes among population groups, beginning prenatally and extending to adulthood and beyond, should be a priority for the NCS designers. The social and struc- tural barriers to health care, including health insurance and access to health and education services, are particularly important to address in order to understand the causes and consequences of health disparities. Many of the exposures and outcomes of concern to the NCS occur disproportionately among poor and low-income children. There are ways to accomplish this within the current NCS design—for example, by encouraging appropriate adjunct studies that would develop hypotheses and carry out analyses of racial and ethnic disparities in child outcomes. Such studies could add sample for specific racial and ethnic groups and perhaps add data items, such as more extensive medical histo- ries from provider records than currently planned (see further discussion in 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 39 Chapter 3). In addition, information on the spatial distribution of clinics and other service providers could be obtained from the Internet and added to the NCS to shed light on study participants’ access to services, although there is no ready substitute for parental reports on use of and attitudes toward service providers. Recommendation 2-1: The NCS should give priority attention to seeking ways to bolster the ability of the study to contribute to un- derstanding of health disparities among children in different racial, ethnic, and other population groups, including the reestablishment of a working group to oversee this area and the encouragement of appropriate adjunct studies. Concepts and Emphasis The NCS research plan is much stronger on disease, impairment, and the identification of risk factors for abnormal development than it is on health, functionality, and the identification of protective and positive factors for normal development. It does not, in fact, define the key constructs of health and development. Moreover, hypotheses in the area of child health and development are very general in nature, including an additional meta- hypothesis in the September 17, 2007, supplement to the research plan on “influences on healthy development.” Overall, the research plan appears deficient in its conceptualization and proposed analyses of the factors that promote healthy growth and development and that forestall or moderate the adverse effects of environmental stressors and toxins (see section below on “Conceptual Framework”). Chronic and Intermittent Exposures In a troubling number of instances, the ability of the NCS to “evaluate the effects of both chronic and intermittent exposures on child health and development,” as required in the Children’s Health Act, seems impaired by the proposed schedule of data collection (see Chapter 3). Four examples that apply to prebirth and the first 24 months of age (the only period for which data collection has been fully specified to date) include: • Glucose metabolism, which, if altered in pregnancy, is hypothesized to be a risk factor for birth defects, will be measured for pregnant women on only two or three occasions, which is insufficient to capture variations that may be important for analysis. Moreover, glucose metabolism will probably not be measured in the first few weeks of pregnancy, when much of organogenesis occurs. 

40 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN • Blood samples will be collected for enrolled children only at birth and at 12 months, when more frequent collection would be desir- able, not only to measure certain chronic conditions, but also to pick up episodic conditions. • Maternal depression, known to be an important stressor for chil- dren, will be measured during pregnancy and at 6 months post- partum, but not subsequently, even though depressive episodes that are consequential for child well-being may extend beyond or occur later than the pregnancy and postpartum periods. Moreover, chronic depression is more likely to interfere with child rearing than temporary depression. • Social development of children will be observed only at 6 and 12 months, and motor and language development will be assessed only at 12 months, which is not frequent enough to identify and categorize children by important developmental lags (or advances) or to relate lags and advances to risk and protective factors. Budget constraints and response burden issues necessarily limit the ability of the NCS to obtain all of the detail that might be desirable at the desired frequency for all domains of interest. To ensure that sufficiently rich data are collected for important domains, we urge the NCS to do two things: first, to develop a richer conceptual framework than is presented in the research plan to guide the choice of hypotheses and specification of measures (see section below on “Conceptual Framework”) and, second, to revisit each hypothesis to determine whether it is sufficiently compelling and clearly specified to retain in the study and, if so, whether adequate data are proposed for collection to permit appropriate analysis. In some instances, there will clearly be a need to add measures, frequency of collection, or both, to the schedule; in other instances, it may make sense to drop or limit the scope of a hypothesis, so that, on balance, the NCS data collection schedule is optimized subject to resource and respondent burden limits. Data Quality The NCS prospective cohort design requires very high quality of the data over the life of the study for its findings to be useful. It is important that the quality of the data be maintained at a uniformly high level across all study locations and all years of the study. We are concerned about the ability of the NCS to provide data of comparable quality across all study locations across time given its complex, multisite organizational structure in which some 35-40 study centers have responsibility for enrollment and data collection (see the discussion in Chapter 4). We are also quite concerned that some aspects of the proposed data 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 41 collection, which appear to be largely driven by budget constraints, will provide data of lesser quality than could and should be obtained. In addi- tion to specific issues in particular domains (see Chapter 3) and the issue raised above about the timing of exposure measures, we generally question the relative infrequency of planned personal visits, particularly in the home, and the decision to limit the collection and coding of medical and other administrative records. Beyond the first year of life, children will not be followed up annually in homes or clinics. Mothers (or primary caregivers) will be asked retrospectively over the telephone to fill in information about important intervening events and conditions, so that many data items will be subject to recall bias and selective attrition, as well as misreporting (for example, of a medical condition or a learning problem in school). The gaps will not be filled with service-based administrative records, which, for the most part, have been judged too expensive to acquire and abstract. Thus, medical record abstraction is planned for delivery and neonatal examination for all 100,000 study participants. Following birth, however, medical and clinical event data will be collected by a personal health record, with the parent as the primary respondent, using a diary or log format developed for the NCS. In a limited number of cases, medical records may be abstracted to confirm information regarding certain condi- tions. The NCS will explore the use of electronic medical records, but the research plan notes that a national, uniform system of electronic records does not now exist and may not be developed within the life of the study. No collection is planned of other types of potentially useful administra- tive records, such as school records or social service agency records—for example, regarding foster care or reports of domestic abuse—although procedures are being developed to collect environmental samples from day care centers and schools attended by study participants. Service-based administrative records are not necessarily as complete or accurate as might be supposed. For example, medical records will not cover children’s health problems in cases in which treatment is not sought, and the recorded diagnosis or diagnoses may not accurately reflect the underlying pathologies for insurance coverage or other reasons. Nonethe- less, relying almost exclusively on retrospective parental reports on medical conditions and treatments and on such sensitive topics as domestic violence or alcoholism and failing to collect records from service providers would seem to markedly impair the quality of the data for understanding child health and development. In sum, as a result of the decision to limit personal visits following the first year of life and to limit the collection of medical and other administra- tive records, data quality on health conditions and other exposure and out- come measures is likely to be lower, and it will be more difficult to identify critical or sensitive periods at which time a specific exposure might cause 

42 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN specific alterations of function. In this regard, the NCS is currently best de- signed to study relationships involving factors operative during pregnancy and the perinatal period because of the intensity and frequency of observa- tion during this phase of child development (see Table 2-1). Although we recognize the budgetary costs of both more frequent in- home visits and the abstraction and coding of medical and other administra- tive records, we urge the NCS to reconsider both decisions. Children grow and change very rapidly, and it seems imperative to have more frequent personal visits. If this is not feasible for the full study, then more frequent visits should be implemented for a sufficient subsample (ideally randomly selected) to test the effects on data quality and retention of participants and make it possible to use appropriate statistical methods to adjust the data for other participants. With regard to medical records, we urge the NCS, at a minimum, to make arrangements with providers to permit the NCS to obtain relevant records to keep in a repository to be abstracted when bud- getary resources allow. We also urge the NCS to investigate the feasibility of obtaining relevant records from schools and social service agencies. Recommendation 2-2: The NCS should seek resources and develop methods to obtain more frequent in-person measures and medical and other administrative records data on study participants. CONCEPTUAL FRAMEWORK Description Chapter 2 in the NCS research plan (NCS Research Plan, Vol. 1) is labeled “Conceptual Design and Framework”; it states broad design cri- teria, such as assessments of multiple exposures and outcomes, repeated measures, following children through to adulthood, and generalizability to the U.S. population, on which we comment in the previous section. It also provides a very simple conceptual model, in which environmental exposures in various domains are linked to outcomes in various domains with gene expression and health care as mediating factors (NCS Research Plan, Vol. 1, Sec. 2.1, Figure 2-1). We found Chapter 1 in the NCS research plan, which is labeled “Back- ground,” to be more informative about the conceptual thinking that un- derlies the study design. It provides a succinct statement of the principles underlying the study. Vol. 1, p. 1-1 states: The National Children’s Study design rests on the principle that both health and susceptibility to disease are determined by dynamic processes that occur throughout life. Perturbations (“insults”) that impact health 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 43 states may occur any time from preconception through adult life. These insults can affect viability, differentiation of major organ systems, somatic growth, and the development of functional processes including maturation of metabolic systems. A range of determinants acting either in concert or synergistically may impact growth and development. These include the built and natural environments with their chemical and physical factors, the social environment, individual behaviors, and biological factors includ- ing genetics. Of particular importance are the earliest stages of human development, pregnancy and early childhood, when cell division, differen- tiation, and maturation are most rapid. These health determinants may influence development in many ways. For those with high potency when acting at critical periods of develop- ment, such as thalidomide or Accutane, severe birth defects will result in most exposed offspring. Most environmental factors, however, are not so potent. More often, factors operating at critical or sensitive periods of development will interact with other factors over the life course to raise or lower the risk of adverse health outcomes. These factors may be genetic or non-genetic. For example, accelerated weight gain during childhood is associated with increased risks of diabetes and cardiovascular outcomes later in life; this phenomenon is accentuated among children born with restricted fetal growth. . . . Only with this appreciation of the complexity of interactions among genetic and environmental factors will we be able to inform the next generation of caregivers about effective prevention and treatment to lower the burden of common chronic conditions of childhood and later-onset diseases that arise from early developmental results. Assessment We appreciate the acknowledgment in the NCS research plan of the complexity of child development and the complex ways in which myriad factors may interact to affect that development. However, we find the con- ceptual framework for the NCS to be less well developed than we would have expected for this stage of the project. While invoking a dynamic, mul- tifactorial, life-course model, the research plan does not specify any particu- lar model or models to guide decisions on which conditions and potential causal factors are chosen for study, the timing of data collection points, the types of data to be collected, and the overall analytical approach. There is also little attempt to capitalize on the interconnections among the condi- tions chosen for study. Relatedly, we find that the research plan fails to define key constructs. While noting that the Children’s Health Act mandated a study on the many environmental influences affecting the health and development of children, the plan does not provide clear definitions of “health” and “development.” 

44 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN There are a number of existing approaches to the conceptualization and modeling of health and development that could be used as a framework against which the data collection and analysis for the NSC could be as- sessed. For example, a life-course approach has been defined (Ben-Shlomo and Kuh, 2002:285) as: the study of long-term effects on chronic disease risk of physical and social exposures during gestation, childhood, adolescence, young adulthood, and later adult life. It includes studies of the biological, behavioral and psycho- social pathways that operate across an individual’s life course, as well as across generations, to influence the development of chronic diseases. While Ben-Shlomo and Kuh (2002) focus their discussion on chronic illness, they argue that such an approach would also apply to wider notions of health and well-being. The Institute of Medicine report (2006), Genes, Behavior, and the Social Environment: Moving Beyond the Nature/Nurture Debate, discusses the kinds of life-course models that are needed to un- derstand the interactions of genetic, behavioral, and social factors as they affect human health. A life-course model would include the identification of critical or sen- sitive periods when a specific exposure might cause specific alterations of function, as well as the identification of risk factors or effect modifiers later in life. These additional risk and protective factors could be further charac- terized as acting independently or in correlated clusters or additive “chains of risk.” In addition, a life-course model could incorporate different levels of influence from the individual to the family and could include neighbor- hood and broader societal influences. The analytic strategy for investigating such a chain of events would be broader than a focus on logistic regression and confounders and would require multilevel modeling as well. Moreover, consideration of multiple influences on outcomes could serve to link some of the outcomes selected. A broader conception of health and a dynamic framework for assess- ing health and development trajectories over time can be found in a recent report by the Board on Children, Youth, and Families (National Research Council and Institute of Medicine, 2004); see Box 2-1. This approach emphasizes optimality and functionality rather than disease and deficit and in that sense is more consistent with the World Health Organization definition of health as “positive, physical, mental and social well-being and not just absence of disease” (World Health Organization, 1946; see also Drotar, 1998; Eiser and Morse, 2001). Such an approach would lead to more of a focus than is evident in the NCS research plan on factors that protect against adverse outcomes or that mitigate the effects of risk factors; on multiple factors affecting outcomes; and on the interplay of chronic ill- ness and development for such specific conditions as asthma (see Hayes, 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 45 BOX 2-1 A New Definition of Children’s Health The report of the Committee on Evaluation of Children’s Health, Children’s Health, The Nation’s Wealth (National Research Council and Institute of Medicine, 2004), defines children’s health as follows (p. 33): Children’s health is the extent to which individual children or groups of children are able or enabled to (a) develop and realize their potential, (b) satisfy their needs, and (c) develop the capacities that allow them to interact successfully with their biological, physical, and social environments. This definition draws upon an explanation from the World Health Organization that health is a state of complete physical, mental, and social well-being, not merely the absence of disease or infirmity; a collaborative effort of the European Union Health Monitoring Programme to develop the Child Health Indicators of Life and Development (CHILD) model; the positive health principles embraced by the Ot- tawa Charter for Health Promotion (1986); and the research literature cited in the report. Key features of the new definition of children’s health include three distinct but related domains (pp. 34-35): • health conditions, “a domain that deals with disorders or illnesses”; •  unctioning, “which focuses on the manifestations of individual health in daily life”; f and •  ealth potential, “which captures the development of health assets that indicate h positive aspects—competence, capacity, and developmental potential.” The report describes these domains in detail and also describes ways to measure not only aspects of each domain, but also the influences on children’s health, which are defined to include (p. 46): • Children’s biology • Children’s behavior • Physical environment o Prenatal exposures o Childhood exposures o Home, school, and work settings o Child injury and the provision of safe environments o The built environment • Social environment o Family o Community o Culture o Discrimination • Services • Policy 

46 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN 1997; Hobbs, Perrin, and Ireys, 1985; Stein et al., 1993; Thompson and Gustafson, 1999). Likewise there are models of psychological development that might guide the NCS. Although the research plan cites the seminal National Research Council publication on child development, From Neurons to Neighborhoods: The Science of Early Childhood Development (National Research Council and Institute of Medicine, 2000), it does not reflect the conceptualization of child development and the factors affecting develop- ment that are discussed in that report. In the area of physical development, the research plan, as presented, does not incorporate the notion of physical development as a normal developmental process. Finally, the research plan does not discuss the interplay of physical growth and aspects of psychologi- cal and behavioral development (French, Story, and Perry, 1995; Molinari et al., 2002). We are not necessarily recommending that the NSC operationalize any of these constructs of health and development completely, but such frame- works could serve as benchmarks against which the study could map what is feasible for the NSC. For example, some constructs might prove too time- consuming or expensive to measure because they require laboratory-based assessments. Others might require more frequent assessments than are pos- sible with the resources of the overall study. In any event, the rationale for what is being done would be clear, and the gaps evident. Such gaps might very well be useful guides to what substudies or adjunct studies individual sites might conduct. Employing explicit frameworks would not necessarily mean going back to the beginning. However, it would involve a review of the relevant theoretical and epidemiologic approaches, selection of the most appropri- ate with an explicit rationale for the choice, a mapping of the currently proposed conditions and measures to be collected against the framework with a description of the potential choices to be made, taking account of feasibility, costs, and response burden, and an assessment of the implica- tions for analysis capitalizing on the power and longitudinality of the NSC. In choosing specific measures to include on the basis of the chosen framework, the NCS should consider—in addition to, or in place of, those already proposed—such indicators of child well-being and mental health as secure attachment relationships, social competence, cognitive development and achievement, optimism, and empathy. One can add to this list assess- ments of positive emotionality, healthy attributional style, ego-resilency, and many more. Some of this work may already have been done and be contained in background white papers, but such thinking is not evident in the research plan. We recognize the pressures on the NCS to get the study under way, but we caution against stinting on conceptual development. We think there is 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 47 time to consider more fully the appropriate conceptual framework to guide the selection of measures and analytical methods for subsequent stages of the project, given that the data collection plan is being developed in 3-year stages, as described above, and only the first stage has been fully specified thus far. Furthermore, if the NCS adopts our recommendation (see last section below) to allow more time than is currently planned between the completion of an operation by the Vanguard Centers and the implemen- tation of that operation by other centers, then there will also be time to consider more fully the appropriate conceptual framework for the study from the earliest stage. We understand that it is not possible—and should not be attempted—to foresee every research question of interest or to develop fully specified, complete models for every possible analysis at the outset of a long study. Indeed, an advantage of such a lengthy data collection period is that the NCS should be able to support important research that has not yet been conceived. Yet without an appropriately rich conceptual framework and well-developed key constructs, there is no assurance that the most ap- propriate measurements will be obtained for even those conditions that are identified for study at the outset. The NCS may want to consider the use of expert consultants to assist in the assessment and specification of appropriate frameworks, constructs, and measures to guide decisions on measures and the timing of assessments to be obtained in the study, with the Vanguard Centers used for pilot testing as recommended below. Recommendation 2-3: The NCS should clearly define the key con- structs of child health and development and more fully develop a conceptual framework for understanding child health and develop- ment over the life course of infancy, childhood, and adolescence. Core Hypotheses Description As indicated in the research plan (NCS Research Plan, Vol. 1, Sec. 4.2), the framing of hypotheses was considered essential to guide study planning and to ensure that important questions would be addressed. In planning the study, a standard was established that a supporting hypothesis must be developed for inclusion of measures or design elements in the NCS. To guide the selection and prioritization of outcome and exposure measures, 28 topic areas for hypotheses, referred to as core or meta-hypotheses, were developed to fulfill the study’s aim of ascertaining the effect of environmen- tal factors on the health and development of children; see Box 2-2. The specific hypotheses in each of the topic areas identify relevant 

48 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN BOX 2-2 Hypotheses Topics (Meta-Hypotheses) of the National Children’s Study • Birth defects from impaired glucose metabolism • Increased risk of preterm birth from intrauterine exposure to mediators of inflammation • Increased risk of fetal growth restriction, preterm birth, birth defects, and developmental disabilities in children born through assisted reproductive technologies • Maternal subclinical hypothyroidism and neurodevelopmental disabilities/ adverse pregnancy outcomes • Nonpersistent pesticides and poor neurobehavioral and cognitive skills • Prenatal infection and neurodevelopmental disabilities • Gene-environment interactions and behavior • Prenatal and perinatal infection and schizophrenia • Family influences on child health and development • Impact of neighborhood and communities on child health • Impact of media exposure on child health and development • Social institutions and child health and development • Influences on healthy development • The role of prenatal maternal stress and genetics in childhood asthma • Exposure to indoor and outdoor air pollution, aeroallergens, and asthma risk • Dietary antioxidants and asthma risk • Social environmental influences on asthma disparities • Early exposure to structural components and products of microorganisms decreases the risk of asthma • Environmental exposures interact with genes to increase the risk of asthma and wheezing in children • Obesity and insulin resistance from impaired maternal glucose metabolism • Obesity and insulin resistance from intrauterine growth restriction • Breastfeeding associated with lower rates of obesity and lower risk of insulin resistance • Fiber, whole grains, high glycemic index, and obesity and insulin resistance • Genetics, environmental exposures, and type 1 diabetes • Repeated mild traumatic brain injury and neurocognitive development • Behavioral exposures, genetics, and childhood or adolescence onset aggression • Antecedents and resiliency to traumatic life events in childhood • Hormonally active environmental agents and reproductive development SOURCE: National Institute of Child Health and Human Development (2007, Vol. 1, Sec. 4.2, Table 4-1; specific hypotheses in each of these areas are provided in Vol. 2, App. A-2). 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 49 environmental exposures, including physical, chemical, biological, and psychosocial factors, that affect the identified priority outcomes, includ- ing pregnancy outcomes, neurodevelopment and behavior, injury, asthma, obesity and growth, child health and development, and reproductive de- velopment. Many hypotheses also take into consideration the impact of gene-environment interactions. The criteria outlined in the research plan for the meta-hypotheses and specific hypotheses are: • they are scientifically compelling, • they have important public health implications, • they are feasible to test, and • they clearly justify the need for a prospective birth cohort study of 100,000. Assessment Although the panel recognizes that the framing of hypotheses is es- sential to guide study planning, we also think that many of the most in- teresting studies that are ultimately carried out with the data are likely to involve concepts and relationships that could not clearly be envisioned at the outset of the study. In our view, the NCS design, as represented in the research plan, allows for the genesis of new research ideas and hypotheses in a number of ways, including the length and breadth of the data-gathering platform, the nationally representative, equal probability sampling design, the provision for substudies and adjunct studies, and the storage of biologi- cal and environmental specimens for later analysis. Nonetheless, the set of initial hypotheses that are developed to guide the study should be carefully selected and well framed to enable the study to collect the most appropriate outcome and exposure measures. The core hypotheses need to be well based in theory and prior research; reflect ap- propriate conceptual models of plausible pathways that relate exposures, mediating factors, and outcomes; and relate clearly to public health policy issues. We are in broad agreement with the NCS’s selection of the seven prior- ity outcome areas—they represent important topics for research to inform public health policy for children. Turning to the specific hypotheses in each area, however, we think that some of them are more compelling and more fully specified than others. Some hypotheses are very broad and lack clear specification and justification, and other hypotheses are very nar- row and limited in usefulness. By outcome area (see Chapter 3 for fuller discussion): 

50 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN • The NCS pregnancy outcome hypotheses are well justified, with the exception of subclinical maternal hypothyroidism as a factor in adverse pregnancy outcomes, for which a compelling case is not offered either on public health or scientific grounds. • The neurodevelopment and behavior hypotheses are not well justi- fied in terms of why some outcomes were selected and not others— for example, it is proposed to study schizophrenia, which often does not become manifest until after the period of study (after age 21), while other behaviors that manifest themselves in adolescence, such as bipolar disorder, eating disorders, substance abuse, and others, are overlooked. • The child health and development hypotheses are very broad: they do not articulate pathways by which such factors as the family and community environments affect or mediate developmental trajectories, and they do not clearly define the range of normal development. • The hypotheses regarding environmental risk factors for the inci- dence of asthma are well developed; they represent an important and original contribution of the NCS to the field of child public health. • The obesity hypotheses focus on maternal intrauterine environ- ments and childhood diets as factors in obesity, to the exclusion of such domains as parental support, physical activity, food avail- ability, schools, and TV watching. • The injury hypotheses are problematic in several respects (for ex- ample, it will be very difficult to accurately measure repeated mild traumatic brain injury) and do not address important public health concerns related to the environmental causes of childhood injuries or the effects of clinical treatment or nontreatment. • The reproductive development hypotheses are potentially an im- portant and original contribution of the NCS, although the specific outcomes (for example, polycystic ovary syndrome) need to be more clearly specified. Hypotheses are nearly missing altogether in the area of racial and eth- nic disparities, as we noted above. The study will undoubtedly contribute to a better understanding of the sources of racial and ethnic, language, so- cioeconomic, and geographic differences in child health, but it has formed almost no hypotheses about these differences that could have informed the study design and made it more useful in this regard. Such hypotheses could have included attention to perceptions of discrimination or stigma that may affect parents’ mental health and their inclination to use social services and medical facilities. Based on the research plan, this subject area appears to 

GOALS, CONCEPTUAL FRAMEWORK, AND CORE HYPOTHESES 51 be largely left to the future: “The NCS will be able to address many major health disparities that currently exist in the United States, and to collect sufficient data to address others not yet recognized” (NCS Research Plan, Vol. 1, Sec. 2.2.2). Finally, we note that no matter how well specified a hypothesis may be, considerable thought and imagination will be required to test it adequately on NCS data (see further discussion in Chapter 4). The difficulties of test- ing hypotheses on observational data are well recognized in the social and health sciences. Such data are rife with selection biases that can produce distorted outcomes unless great care is taken in both design and analysis. For example, what appear to be neighborhood effects should often be at- tributed to the characteristics (sometimes unmeasured) of individuals who live in an area. As the NCS moves from data gathering to analysis, more attention will need to be given to methodological issues in the use of ob- servational data than is evident in the research plan. USING THE VANGUARD CENTERS AS PILOTS An important shortcoming of the NCS, as currently planned, is the absence of a pilot phase for thorough testing, evaluation, and refinement of study concepts, measures, instruments, and procedures. The study design is complex in every aspect, from developing appropriate constructs, hypoth- eses, and measures of outcomes and exposures, to identifying and enrolling participants, maintaining a high response rate over the life of the study, administering the very large number of survey, clinical, and environmental sampling instruments, and managing very large databases to which many study centers have contributed. Moreover, in this and other chapters, we raise many questions about concepts, measures, frequencies and types of data collection, and other aspects of the study. For the study to achieve its full potential, we think that more time is needed at the outset for refinement of key constructs and measures and for full testing, evaluation, and modification, as appropriate, of study proce- dures. At present, it appears that little, if any, formal experimentation will be included in the work of the Vanguard Centers. Moreover, while the Vanguard Centers will begin enrollment and data collection a year in ad- vance of other study centers, a year does not seem long enough to allow for thorough analysis of how well procedures worked in the Vanguard Centers and for improvements to be effected prior to when operations begin in the other centers. We urge the NCS to consider lengthening the time between the start of Vanguard Center operations and the start of operations in other study centers. Although we do not think that we can make a specific recom- mendation about the duration of the pilot, we do think that an additional 

52 THE NATIONAL CHILDREN’S STUDY RESEARCH PLAN 6 months to a year between the Vanguard sites data collection and the collection of data by all the sites would be beneficial. The increase in time should be implemented not only for the first phase of enrollment through age 24 months, but also for the introduction of each new set of operations and instruments for subsequent 3-year periods throughout the life of the study. In addition, we encourage the NCS to build in experimentation with instruments and study procedures into the Vanguard operations when the extant literature is not sufficiently clear about the preferred approaches. Recommendation 2-4: We strongly urge the NCS to delay enroll- ment at new sites to make effective use of initial findings from participant enrollment and data collection in the Vanguard Center sites to improve study procedures, as appropriate, and to refine key concepts, hypotheses, and measures of outcomes and exposures. Throughout the life of the study, the NCS should use the Vanguard Centers to pilot test and experiment with data collection methods and instrumentation.