Confounding by Time-varying covariates in Epidemiological Studies of Waterborne Contaminants

Increased risk of reproductive outcomes has been reported following exposure to drinking water contaminants. Given the small magnitude of reported effects, these studies may be more susceptible to potential biases such as residual confounding. Despite demonstrated seasonality trends in the incidence of adverse reproductive effects and waterborne contaminant occurrence, time-varying confounders (e.g., seasonality and meteorological variables) have not been thoroughly assessed in previous studies, such as disinfection byproducts. To assess the potential for temporal confounding, we examined the effect of total trihalomethanes (TTHMs) on fetal growth from an epidemiological study of 36,000 births during 1997-1998. We examined confounding for several categorical (season, month and year of birth) and continuous variables (precipitation, minimum and maximum ambient daily temperature). Based on preliminary analyses, precipitation and maximum temperature were the strongest confounders. Relative to the lowest TTHM exposure tertile (<34 ug/l), the unadjusted mean birthweight was 22 grams lower (95%CI: -41, -4) for the highest TTHM exposure tertile (>74 ug/l). Following adjustment for precipitation and season, the reduction in mean birthweight for the highest TTHM exposure tertile was nearly twice as large as the unadjusted results (-43 grams; 95%CI: -69, -17). Following adjustment for maximum daily temperature, precipitation and year of birth, the mean birthweight for the highest exposure tertile was 58 grams (95%CI: -88, -28) less than the lowest exposure tertile. Ongoing analyses include an assessment of this confounding relative to other maternal and reproductive risk factors and an examination of other time-varying adjustment procedures (e.g., sinusoidal logistic regression) for dichotomous endpoints.