BACKGROUND: Prenatal tobacco exposure is a risk factor for the development of externalizing behaviors and is associated with several adverse health outcomes. Because pregnancy smoking is a complex behavior with both daily fluctuations and changes over the course of pregnancy, quantifying tobacco exposure is a significant challenge. To better measure the degree of tobacco exposure, costly biological specimens and repeated self-report measures of smoking typically are collected throughout pregnancy. With such designs, there are multiple, and substantially correlated, indices that can be integrated via new statistical methods to identify patterns of prenatal exposure.

METHOD: A multiple-imputation-based fuzzy clustering technique was designed to characterize topography of prenatal exposure. This method leveraged all repeatedly measured maternal smoking variables in our sample data, including (a) cigarette brand; (b) Fagerstrom nicotine dependence item scores; (c) self-reported smoking; and (d) cotinine level in maternal urine and infant meconium samples. Identified exposure groups then were confirmed using a suite of clustering validation indices based on multiple imputed datasets. The classifications were validated against irritable reactivity in the first month of life and birth weight of 361 neonates (Male(_n)=185; Female(_n)=176; Gestational Age_(Mean)=39weeks).

RESULTS: This proposed approach identified three exposure groups, non-exposed, lighter-tobacco-exposed, and heavier-tobacco-exposed based on high-dimensional attributes. Unlike cut-off score derived groups, these groupings reflect complex smoking behavior and individual variation of nicotine metabolism across pregnancy. The identified groups predicted differences in birth weight and in the pattern of change in neonatal irritable reactivity, as well as resulted in increased predictive power. Multiple-imputation-based fuzzy clustering appears to be a useful method to categorize patterns of exposure and their impact on outcomes.