The impact of different causal models on estimated effects of disinfection by-products on preterm birth

Background: Previous epidemiologic studies of preterm birth and drinking water disinfection by-products (DBP) reported inconsistent results especially for third trimester exposures. These inconsistencies may have been due to differences in the underlying causal model assumed and methodological issues, including the method of analysis (cumulative vs. density-sampling of controls and matching on gestational age) and appropriate control of confounding. Methods: We use data from previously published research to illustrate how different causal models, methods of analysis, and the choice of covariates to control impact results. Results: Exposure at high measured TTHM levels ( >= 60 mu g/l) during the last trimester - with cumulatively sampled controls - corresponded to negative effect estimates when comparing preterm to term births and averaging exposure over different length periods. In contrast, density-sampling of controls with an exposure truncated at 36 weeks gestation and adjustment for possible confounding by exposures experienced in prior trimesters led to moderate changes in risk at the highest level of exposure averaged over the four weeks prior to birth. Conclusions: We recommend that future research on an exposure to DBPs and risk of preterm birth explore the sensitivity of their findings to different model specifications, specifically: (1) cumulative vs. density-sampling of controls when evaluating third trimester or whole pregnancy exposures, taking into account exposure-averaging length; (2) short-term peak exposures vs. long-term exposures; and (3) adjustment for exposure during prior pregnancy periods when evaluating later trimester exposures to account for possible 'priming' effects of early exposures. (C) 2010 Elsevier Inc. All rights reserved.