Abstract
Epidemiologists have used trihalomethanes (THMs) as a surrogate for overall disinfection byproduct (DBP) exposure based on the assumption that THM concentrations are proportional to concentrations of other DBP classes. Toxicological evidence indicates THMs are less potent toxins than unregulated classes like haloacetonitriles (HANs). If THMs are not proportional to the DBPs driving toxicity, the use of THMs to measure exposure may introduce non-trivial exposure misclassification bias in epidemiologic studies. This study developed statistical models to evaluate the covariance and proportionality of HAN and THM concentrations in a dataset featuring over 9500 measurements from 248 public water systems. THMs only explain ∼30% of the variance in HANs, whether the data is pooled in a classic linear regression or hierarchically grouped by water system in a multilevel linear regression. The 95% prediction interval on HANs for the median THM concentration exceeds the interquartile range of HANs. Mean HAN:THM ratios range from ∼2.4% to ∼80% across water systems, and varied with source water category, season, disinfectant sequence and distribution system location. The HAN:THM ratio was 265% higher in groundwater systems than in surface water systems and declined by ∼40% between finished effluent and maximum residence times in surface water systems with chlorine-chlorine disinfection. A maximum likelihood approach was used to estimate the misclassification bias that may result from using THMs to construct risk-ratios, assuming that HANs represent the "true" DBP exposure risk. The results indicate an odds ratio of ∼2 estimated with THM concentrations could correspond to a true odds ratio of 4-5. These findings demonstrate the need for epidemiologic studies to evaluate exposure by measuring DBPs that are likely to drive toxicity.