Evaluation of uncontrolled confounding in studies of environmental exposures and neurobehavioral testing in children

Neurobehavioral tests are commonly used in studies of children exposed to low-level environmental concentrations of compounds known to be neurotoxic at higher levels. However, uncontrolled or incomplete control for confounding makes interpretation of results problematic because effects of confounders are often stronger than the effects of primary interest. We examined a priori the potential impact of confounding in a hypothetical study evaluating the association of a potentially neurotoxic environmental exposure with neurobehavioral function in children.

We used 2 outcome measures: the Bayley Scales of Infant Development Mental Development Index and the Stanford-Binet Intelligence Scale Composite Score. We selected 3 potential confounders: maternal intelligence, home environment, and socioeconomic status as measured by years of parental education. We conducted 3 sets of analyses measuring the effect of each of the 3 confounding factors alone, 2 confounders acting simultaneously, and all 3 confounders acting simultaneously.

Relatively small differences (0.5 standard deviations) in confounding variables between "exposed" and "unexposed" groups, if unmeasured and unaccounted for in the analysis, could produce spurious differences in cognitive test scores. The magnitude of this difference (3-10 points) has been suggested to have a meaningful impact in populations. The method of measuring confounders (eg, maternal intelligence) could also substantially affect the results.

It is important to carefully consider the impact of potential confounders during the planning stages of an observational study. Study-to-study differences in neurobehavioral outcomes with similar environmental exposures could be partially explained by differences in the adjustment for confounding variables.