Influence of exposure assessment methods on risk estimates in an epidemiologic study of total trihalomethane exposure and spontaneous abortion

Trihalomethanes are common contaminants of chlorinated drinking water. Studies of their health effects have been hampered by exposure misclassification, due in part to limitations inherent in using utility sampling records. We used two exposure assessment methods, one based on utility-wide sampling averages, and one based on measurements from the utility sampling site closest to the subject's residence, to reestimate total trihalomethane (TTHM) exposure for 4212 participants in a preexisting study of risk factors for spontaneous abortion (SAB). For both approaches we performed unweighted, weighted, and subset analyses. The weighted and subset analyses were intended to reduce exposure misclassification, and were based on within-utility variance in TTHM measurements for the utility-wide average approach, and the distance between the subject's residence and sampling site for the closest-site approach. In general, the utility-wide average methods produced odds ratios equivalent to or slightly higher than the closest-site methods. Odds ratios obtained using the utility-wide average, but not the closest-site, approach also became progressively stronger in the weighted and subset analyses. A dose-response was seen between SAB and an exposure metric incorporating both TTHM concentration (utility-wide average approach) and personal ingestion, with SAB rates ranging from 8.3% to 13.7% (unweighted), 7.9% to 16.6% (variance weighted), and 6.6% to 23.1% (low-variance subset). Utility-wide average TTHM exposure assessment methods together with variance-based weights and subsets are relatively simple exposure assessment techniques, which may increase the epidemiologic usefulness of utility sampling records.