Hum. Reprod. Advance Access originally published online on October 16, 2007
Human Reproduction 2007 22(12):3267; doi:10.1093/humrep/dem332
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Letters to the Editor |
Gestational age and gestational age-at-delivery: cause, effect, or time-scale?
Division of Epidemiology, Statistics and Prevention Research, National Institute of Child Health and Human Development, 6100 Executive Blvd, Rm 7B03A, Rockville, MD 20852, USA
1 Correspondence address. E-mail: whitcomb{at}mail.nih.gov
We recently read with great interest ‘Should we adjust for gestational age when analyzing birth weights? The use of z-scores revisited’ by Delbaere et al. (2007)
. Specifically, Delbaere and colleagues consider the question whether birth weight (BWT) differs between children born from single-embryo transfer (SET) and singletons born from double-embryo transfer (DET) and describe the impact of adjustment for gestational age (GA)-at-delivery. We support the application of causal thinking to model specification, and we commend the authors for use of directed acyclic graphs (DAGs) in their analysis of BWT data.
Investigators benefit from tools like DAGs for addressing complex substantive questions by formalizing causal assumptions about GA-at-delivery. Delbaere and colleagues discuss approaches to address this variable and, in the context of assessing the effect of SET/DET on BWT, contend that GA-at-delivery is on the causal pathway; because GA-at-delivery and BWT also may share common unmeasured causes, adjustment for GA-at-delivery induces collider-stratification bias. This proposal warrants further consideration.
GA is an indirect representation of the true status of the highly timed and interrelated processes underlying embryonic and fetal development. Biologic development occurs in a series of events, rather than units of time; however, GA is a convenient proxy for actual biologic development, albeit one that is prone to measurement error. Except in rare circumstances such as when assisted reproductive technologies are utilized, dating is based on estimation. Similarly, GA-at-delivery is the value for the GA variable as recorded at delivery. It, too, is an indirect measure of factors that affect the duration of gestation. As such, GA-at-delivery cannot be a causal factor, per se; but rather is a time-stamp reflecting a combination of factors including biologic maturity and clinical practice. Notably, if neither BWT nor GA-at-delivery is causal of the other, any meaningful correlation between them must be due to some shared causal mechanism.
The distinction between GA-at-delivery and the factors it represents is not simply a matter of semantics. Since age cannot cause prematurity, some biologic factor impacting the length of gestation or fetal development rate is a more likely candidate. Consequently, GA-at-delivery cannot mediate the effect of conception/implantation factors and outcomes like BWT. As Delbaere and colleagues have concluded, adjustment for GA-at-delivery is ill-advised; however, it will not result in collider-stratification bias. Instead, it will be a source of over-adjustment if BWT and GA-at-delivery share any common cause, measured or unmeasured, resulting in estimates that are biased toward the null.
The matter is further complicated by informative censoring from indicated obstetrical interventions. When the BWT ranges for a given GA-at-delivery are not representative of all BWT at that GA, analysis is prone to bias (Joseph, 2004, 2007). We have previously shown bias resulting when causal assumptions as reflected by DAGs and analytic approaches are mismatched (Schisterman et al., 2005); however, if true causal relations are mis-specified then the analysis appropriate to that incorrect causal system will be wrong as well. Thus, while good judgment in analysis is important for minimizing bias, statistical modeling using DAGs is still highly susceptible to the choices of investigators in assignment of ‘cause’ and ‘effect’.
References
Delbaere I, Vansteelandt S, De Bacquer D, Verstraelen H, Gerris J, De Sutter P, Temmerman M. Should we adjust for gestational age when analysing birth weights? The use of z-scores revisited. Hum Reprod (2007) 22:2080–2083.
Joseph KS. Theory of obstetrics: the fetuses-at-risk approach as a causal paradigm. J Obstet Gynaecol Can (2004) 26:953–960.[Medline]
Joseph KS. Theory of obstetrics: an epidemiologic framework for justifying medically indicated early delivery. BMC Pregnancy Childbirth (2007) 7:4.[CrossRef][Medline]
Schisterman EF, Whitcomb BW, Louis GM, Louis TA. Lipid adjustment in the analysis of environmental contaminants and human health risks. Environ Health Perspect (2005) 113:853–857.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||