US National Children's Study

Abstract

The authors respond: We thank Robert Michael and Colm Muircheartaigh for their letter to the editor, and we agree with their contention that the more fully representative the participants in the National Children's Study are, the more directly the results can be generalized to the US population and applied to policy formulation. We also concur that representativeness is more critical for valid assessment of sociologic and behavioral determinants of health than for narrower biologic ones. However, “representativeness” is not simply a matter of the initial selection algorithm, particularly in a 21-year follow-up study with a substantial retention burden. The question is not whether formal statistical representativeness is desirable—because it surely is—but a question of how that representativeness is best achieved. We believe that recruitment through medical care providers is likely to yield not only higher response rates but also greater retention over the course of the study. The selection of those practices through which recruitment will occur must be done carefully and strategically to ensure geographic representativeness of the resulting cohort. The current protocol, which essentially represents door-to-door “cold calling,” has worked well for cross-sectional surveys, but it is unproven for enrolling participants in an extremely demanding long-term cohort study—and thus far the results are not reassuring. Statistical sampling of births does not guarantee representativeness of the resulting cohort. We contend that the product of a well-designed approach to identification and recruitment in prenatal care clinics would be superior to household recruitment in obtaining a final sample that represents the population of children in the chosen geographic areas. Although a well-defined statistical sample has the appeal of allowing formal calculation of response proportions and estimation of population values, when attrition from nonresponse reaches a high level, statistical generalizability is, in fact, illusory. Population representativeness in the more general sense is the goal—the enrollment of a cohort that approximates and reflects children in the United States. Probability sampling at the individual level is the ideal means of attaining that goal. However, given other features of the study—namely the intensity of data collection and long duration of follow-up—we believe that there are better ways to obtain a geographically and demographically representative cohort of children, at less cost in terms of both dollars and the need to forego other desirable features of the study. David A. Savitz Mount Sinai School of Medicine New York, NY [email protected] Roberta B. Ness The University of Texas School of Public Health Houston, TX