Social disadvantage and infant mortality: the birth weight paradox revisited via latent classes and sensitivity analyses
De Stavola, B., Daniel, R., Silverwood, R., Stuchbury, R. & Grundy, E. (2014) UK Causal Inference Meeting Centre for Mathematical Sciences, University of Cambridge, UK. 28 - 29 April 2014 [ONS LS]
The observation that the relative risk of infant mortality in socio-economically disadvantaged groups is reversed at the lower end of the birth weight distribution (the so- called low birth weight paradox) has received much attention in the literature [1-5]. Several alternative explanations have been offered, more recently focused on the likely impact of unmeasured confounding of the birth weight-infant mortality relationship, leading to collider bias when the effect of social disadvantage is stratified by birth weight. An additional complication however arises from the fact that birth weight is an indicator of multiple processes, including prematurity and in utero growth retardation, and these may have different influences on infant mortality . A more complex causal formalization of the mediating/moderating effect of birth weight is therefore required.
We investigate this paradox using data from the Office for National Statistics Longitudinal Study. This is a record linkage study which comprises linked census and birth/mortality data for 1% of the population of England and Wales. It includes birth weight information as recorded at birth registration and socio-economic indicators collected at the various censuses. We use data from 1981 to 2011 comprising nearly 200,000 births and over 1,200 deaths.
Using such a large and long-term cohort study adds complexities to the analyses. In particular data do not include information on gestational age/prematurity and require handling temporal changes in birth weight distribution, as well as mortality rate. We address these challenges by allowing for a mixture of two birth weight distributions: the first comprises babies who are mostly born at term and at low risk of mortality, and the second babies who are generally small/preterm and at high risk of mortality [2,7]. We allow for these distributions to vary by sex, ethnicity, calendar time and region, and use the predicted probability of belonging to the compromised latent group as the mediator for the effect of socio-economic disadvantage on infant mortality. We then examine the impact of unmeasured mediator-outcome confounding via sensitivity analyses adapted from VanderWeele  and Imai et al .
 Yerushalmy, J. Relationship of parents cigarette smoking to outcome of pregnancy. Implications as to problem of infering causation from observed associations. American Journal of Epidemiology. 1971;93(6):443-456.
 Wilcox AJ. Birth-weight and perinatal mortality. The effect of maternal smoking. American Journal of Epidemiology. 1993 May;137(10):1098-104.
 Wilcox AJ. Commentary: On the paradoxes of birth weight. International Journal of Epidemiology. 2003 Aug;32 (4):632-3.
 Hernandez-Diaz S, Schisterman EF, Hernan MA. The birth weight "paradox" uncovered? American Journal of Epidemiology. 2006 Dec;164(11):1115-2.
 Vanderweele T, Munford SL, Schisterman EF. Conditioning on Intermediates in Perinatal Epidemiology. Epidemiology . 2012; 23 (1):1-9.
 Basso O, Wilcox AJ, Weiberrg CR. Birth Weight and Mortality: Causality or Confounding? Am J Epidemiol 2006;164:303–311.
 Wilcox AJ. On the importance - and the unimportance - of birth weight. International Journal of Epidemiology. 2001 Dec;30(6):1233-41.
 Imai K, Keele L, Yamamoto T. Identification, inference, and sensitivity analysis for causal mediation effects. Stat Sci. 2010;25(1):51-71.
Available online: Link
Output from project: 0301445