The Problem of Confounding in Studies of the Effect of Maternal Drug Use on Pregnancy Outcome

Bengt Källén

doi:10.1155/2012/148616

Abstract

In most epidemilogical studies, the problem of confounding adds to the uncertainty in conclusions drawn. This is also true for studies on the effect of maternal drug use on birth defect risks. This paper describes various types of such confounders and discusses methods to identify and adjust for them. Such confounders can be found in maternal characteristics like age, parity, smoking, use of alcohol, and body mass index, subfertility, and previous pregnancies including previous birth of a malformed child, socioeconomy, race/ethnicity, or country of birth. Confounding by concomitant maternal drug use may occur. A geographical or seasonal confounding can exist. In rare instances, infant sex and multiple birth can appear as confounders. The most difficult problem to solve is often confounding by indication. The problem of confounding is less important for congenital malformations than for many other pregnancy outcomes.

Highlights

The golden standard in medical clinical science is the randomized double-blind study
If the study is a case-control study, controls to cases are selected with, for instance, the same maternal age and other characteristics one wants to adjust for. If it is a cohort study, the unexposed subjects are selected with the same characteristics as the exposed subjects
Either matching is made by selection of pairs or triplets of case and control(s) with similar characteristics, or a group of controls is chosen with a composition similar to the whole group of cases (“frequency matching”)

Summary

Introduction

The golden standard in medical clinical science is the randomized double-blind study. If a factor directly affects both exposure and outcome, a confounding will exist and adjustment for it is needed The opposite effect is obtained if the exposure is a drug, the use of which increases with maternal age, and Down syndrome is the outcome This will result in an increased crude OR. If adjustment is made for maternal age, the effect may disappear These two examples show that confounding can result in a too low risk estimate or a too high estimate, depending on whether the effects of the confounder are in the opposite or the same direction on exposure and outcome. The problem of confounding will be relevant for all outcomes

Material and Methods

Results and Discussion

Confounding and Stratification

Confounding by Maternal Characteristics

Confounding by Infant Characteristics

Concluding Remarks