Use of Negative Control Exposure Analysis to Evaluate Confounding: An Example of Acetaminophen Exposure and Attention-Deficit/Hyperactivity Disorder in Nurses' Health Study II

Abstract

Frequent maternal use of acetaminophen in pregnancy has been linked to attention-deficit/hyperactivity disorder (ADHD) in children, but concerns regarding uncontrolled confounding remain. In this article, we illustrate use of the negative control exposure (NCE) approach to evaluate uncontrolled confounding bias in observational studies on pregnancy drug safety and explain the causal assumptions behind the method. We conducted an NCE analysis and evaluated the associations between maternal acetaminophen use during different exposure periods and ADHD among 8,856 children born in 1993-2005 to women enrolled in the Nurses' Health Study II cohort. Information on regular maternal acetaminophen use was collected prospectively in biennial questionnaires. A total of 721 children (8.1%) in the cohort had been diagnosed with ADHD as reported by the mothers. Our NCE analysis suggested that only acetaminophen use at the time of pregnancy was associated with childhood ADHD (odds ratio = 1.34, 95% confidence interval: 1.05, 1.72), and the effect estimates for the 2 NCE periods (about 4 years before and 4 years after the pregnancy) were null. Our findings corroborate those of prior reports suggesting that prenatal acetaminophen exposure may influence neurodevelopment. The lack of an association between acetaminophen use in the pre- and postpregnancy exposure periods and ADHD provides assurance that uncontrolled time-invariant factors do not explain this association.

Keywords: acetaminophen; attention-deficit/hyperactivity disorder; negative control exposure analysis; neurological development; pregnancy; prenatal exposure delayed effects; uncontrolled confounding.

Figure 1.

Directed acyclic graph illustrating the basic principles of negative control exposure analysis. E and D are the exposure and outcome of interest, respectively. The directed acyclic graph is drawn under the assumption of a true causal association between E and D, although there may not be one (this is the question being studied). C represents a set of measured and controlled confounding variables, and U represents a set of unknown or unmeasured confounding variables. The brackets around C suggest that the variable(s) is/are controlled. N is the “negative control” variable.