COMPARISON OF EXPOSURE ASSESSMENT METHODS FOR RISK ESTIMATION OF DISINFECTION BY-PRODUCTS

Abstract

Listen

Translate article

ISEE-243 Abstract: Summary statistics of quarterly water utility monitoring data in conjunction with data collected on water consumption and use are widely employed to calculate exposure in epidemiologic studies concerning disinfection by-products such as trihalomethanes (THM). However, this use of monitoring data can result in exposure misclassification due to spatial variation in THM concentrations within water distribution systems. Two methods, each designed to minimize misclassification error due to spatial variation, have been reported. The “weighting” metric assigns a weighting factor between 0.0 and 1.0 to frequency counts based on the degree of spatial variation (high to low) in each set of samples (sampling event) in a quarterly monitoring report. The “threshold” metric applies a suite of statistical tests to the same data in order to eliminate sampling events with high spatial variation, and thus restricts the study population used in the risk estimate. We compared these two methods in terms of classifying spatial variability in monitoring data, classifying exposure based on THM concentration, and calculating epidemiological risk estimates. We used utility (460 sampling events reported by 105 utilities over the period 1998–2002) and case/control data (200 participants matched to the THM data at date of conception) from Texas subjects in the National Birth Defects Prevention Study. We found good agreement between the two metrics in terms of categorizing and predicting sampling events as high or low in spatial variation. Across utilities with high and low THM concentrations, the weighting metric consistently assigned a weighting factor of < 0.89 when the threshold metric predicted low spatial variability. The proportions of cases and controls for the analyses of the threshold metric and the non-transformed data in the ’20 and <80 mg/L exposure categories were the same (56% and 44% respectively). These proportions were slightly different in the weighting metric analysis (<80 = 62%, ≤20 = 38%). Associated risk using the threshold metric (n=33) was higher (mean OR (CI) = 3.0(0.5,17.9)) than the risk calculated by the weighting metric (n=31, mean OR (CI)=1.9(0.3,10.8)) or by using non-transformed monitoring data (n=41, mean OR (CI)=1.8(0.4,2.1)). Our results indicated that using exposure metrics that adjust for spatial variation appear to improve detection of risk of birth defects in association with THM in water supplies. However, further validation by verifying predicted classification of degree of spatial variation for both metrics through field monitoring of DBP concentrations in utility water distribution systems is needed.