American Journal of Epidemiology Advance Access originally published online on July 17, 2006
American Journal of Epidemiology 2006 164(4):315-316; doi:10.1093/aje/kwj239
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Response to Invited Commentary |
Basso et al. Respond to "Simple Models for a Complicated Reality"
1 Epidemiology Branch, National Institute of Environmental Health Sciences, National Institutes of Health, US Department of Health and Human Services, Research Triangle Park, NC
2 Biostatistics Branch, National Institute of Environmental Health Sciences, National Institutes of Health, US Department of Health and Human Services, Research Triangle Park, NC
Correspondence to Dr. Olga Basso, Epidemiology Branch, MD A3-05, NIEHS, NIH, HHS, P.O. Box 12233, 111 TW Alexander Drive, Research Triangle Park, NC 27709 (e-mail: bassoo2{at}niehs.nih.gov).
Received for publication March 24, 2006. Accepted for publication April 25, 2006.
The commentary by Schisterman and Hernández-Díaz (1
) on our paper (2) underscores the challenge facing anyone who wishes to analyze birth weight and understand its implications for public health. There is a spectrum of possible explanations for the characteristic U shape of the weight-specific mortality curve (figure 1). At one extreme, the mortality curve can be regarded as the direct effect of birth weight on mortality (figure 1, left panel). In this view, the relation of birth weight (or, more specifically, fetal growth) with mortality is undistorted by confounding factors. This assumption underlies every birth-weight study or intervention that assumes that a change in birth weight will have a predictable effect on mortality.
|
The problem with this conceptualization is that the available data argue otherwise. Changes in birth weight do not have predictable effects. Factors such as high altitude can produce relatively lower birth weights without increasing mortality (3
). Furthermore, true risk factors, such as smoking, that raise mortality while decreasing birth weight produce the so-called birth-weight paradox, in which small babies born to mothers who smoke have relatively lower mortality (3). However, if what is important for mortality is not small size but the reason behind small size, and more than one factor affects both size and mortality, there may be no paradox at all.
We believe that the presence of unmeasured confounders may be an essential clue to understanding the role of birth weight. In the extreme case, birth weight itself would have no causal association with mortality, and confounding would completely explain the relation (figure 1, right panel). It is this extreme case that we modeled in our paper (2). By exploring the properties that would be required of unknown confounders in order to create the empirical curve for babies at term, we produced an excellent fit to the curve with a few simple assumptions. We found that the necessary confounders would affect only a tiny fraction of babies, producing high mortality in this group as well as substantial changes in fetal growth.
How likely is this extreme case? In one sense, it is unlikely. The biologic factors that might produce such strong confounding are not obvious, although there are some possible candidates (2). This model does have one advantage, however. Unlike the model at the opposite extreme, which produces paradoxes and consternation, our model is completely consistent with the observed patterns of weight-specific mortality.
We believe that unmeasured confounding may be key to understanding the relation of birth weight to mortality. In demonstrating the possibility of complete confounding (at least in principle), we hope to shift attention toward exploration of those unmeasured confounding factors. Our extreme example may not literally describe reality, but we would not be surprised if the truth lies closer to this end of the spectrum than to the other. If so, this model may turn out to be one of the useful ones.
 |
ACKNOWLEDGMENTS |
|---|
This research was supported by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences.
Conflict of interest: none declared.
|
References |
|---|
 TOP References |
|---|
- Schisterman EF, Hernández-Díaz S. Invited commentary: simple models for a complicated reality. Am J Epidemiol 2006;164:312–14.
[Free Full Text] - Basso O, Wilcox AJ, Weinberg CR. Birth weight and mortality: causality or confounding? Am J Epidemiol 2006;164:303–11.
[Abstract/Free Full Text] - Wilcox AJ. On the importance—and the unimportance—of birthweight. Int J Epidemiol 2001;30:1233–41.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||