Re: Mediators of the association between pre-eclampsia and cerebral palsy: population based cohort study
In a large population based cohort study in Norway, Strand et al found that offspring of mothers with preeclampsia had an increased risk of cerebral palsy (unadjusted odds ratio (OR): 2.5; 95% confidence interval (CI): 2.0 to 3.2 ) . Upon adjustment for small-for-gestational age status, the OR slightly decreased. Additional adjustment for gestational age reversed the apparent direction of the association, with preeclampsia being associated with a reduced risk of preeclampsia (OR: 0.73; 95% CI: 0.56 to 0.96). Further, in subgroup analyses, non-small-for-gestational age babies born very preterm (<32 weeks) and exposed to preeclampsia had reduced risk of cerebral palsy compared with unexposed babies of the same age.
Should we conclude that preeclampsia protects preterm babies from cerebral palsy?
Before accepting this counterintuitive finding, the authors should have considered the possibility that adjustment for a mediator such as gestational age could induce bias in the estimate of the association between preeclampsia and cerebral palsy [2,3]. Indeed, if there was an unmeasured common cause of both the mediator gestational age and the outcome cerebral palsy (i.e., a confounder of the association between gestational age and cerebral palsy), adjustment for gestational age could create a spurious association between preeclampsia and cerebral palsy. Intrauterine infection could be such a common cause .
This type of bias is called collider stratification bias and is a form of selection bias . There are several examples in perinatal epidemiology where adjustment for or conditioning on a mediator can be misleading [2,5]. One example is the observation of a protective effect of maternal smoking on mortality in low birth weight new born, referred to as the “birth weight paradox” . This protective effect is an artifact caused by the conditioning on birth weight - a mediator between smoking and infant mortality - without adjustment for common causes of birth weight and infant mortality.
To be sure that such a bias is not at work, Strand et al should have assessed common causes of gestational age and cerebral palsy, and should have adjusted for these factors in the analyses. Another strategy would have been to conduct sensitivity analyses to assess the potential effect of these factors on the estimate of the association between preeclampsia and cerebral palsy .
1. Strand KM, Heimstad R, Iversen AC, et al. Mediators of the association between pre-eclampsia and cerebral palsy: population based cohort study. BMJ 2013;347:f4089.
2. Wilcox AJ, Weinberg CR, Basso O. On the pitfalls of adjusting for gestational age at birth. Am J Epidemiol 2011;174:1062-8.
3. VanderWeele TJ, Hernández-Diaz S. Is there a direct effect of pre-eclampsia on cerebral palsy not through preterm birth? Paediatr Perinat Epidemiol 2011;25:111-5.
4. Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology 2004;15:615-25.
5. Hernández-Díaz S, Schisterman EF, Hernán MA. The birth weight "paradox" uncovered? Am J Epidemiol 2006;164:1115-20.
Competing interests: No competing interests