Paul L. Mosquin

Reconstructing exposures from small samples using physiologically based pharmacokinetic models and multiple biomarkers.

This study examines the use of physiologically based pharmacokinetic (PBPK) models for inferring exposure when the number of biomarker observations per individual is limited, as commonly occurs in population exposure surveys. The trade-off between sampling multiple biomarkers at a specific time versus fewer biomarkers at multiple time points was investigated, using a simulation-based approach based on a revised and updated chlorpyrifos PBPK model originally published. Two routes of exposure, oral and dermal, were studied as were varying levels of analytic measurement error. It is found that adding an additional biomarker at a given time point adds substantial additional information to the analysis, although not as much as the addition of another sampling time. Furthermore, the precision of the estimates of exposed dose scaled approximately with the analytic precision of the biomarker measurement. For acute exposure scenarios such as those considered here, the results of this study suggest that the number of biomarkers can be balanced against the number of sampling times to obtain the most efficient estimator after consideration of cost, intrusiveness, and other relevant factors.

Estimation of upper centile concentrations using historical atrazine monitoring data from community water systems.

A survey sampling approach is presented for estimating upper centiles of aggregate distributions of surface water pesticide measurements obtained from datasets with large sample sizes but variable sampling frequency. It is applied to three atrazine monitoring programs of Community Water Systems (CWS) that used surface water as their drinking water source: the nationwide Safe Drinking Water Act (SDWA) data, the Syngenta Voluntary Monitoring Program (VMP), and the Atrazine Monitoring Program (AMP).The VMP/AMP CWS were selected on the basis of atrazine monitoring history (CWS having at least one annual average concentration from SDWA ≥ 1.6 ppb atrazine since 1997 in the AMP). Estimates of the raw water 95th, 99th, and 99.9th centile atrazine concentrations for the VMP/AMP CWS are 4.82, 11.85, and 34.00 ppb, respectively. The corresponding estimates are lower for the finished drinking water samples, with estimates of 2.75, 7.94, and 22.66 ppb, respectively. Finished water centile estimates for the VMP/AMP CWS using only the SDWA data for these sites are consistent with the results. Estimates are provided for the April through July period and for CWS based on surface water source type (static, flowing, or mixed). Requisite sample sizes are determined using statistical tolerance limits, relative SE, and the Woodruff interval sample size criterion. These analyses provide 99.9% confidence that the existing data include the 99.9th centile atrazine concentration for CWS raw and finished water in the Midwest atrazine high-use areas and in the nationwide SDWA dataset. The general validity of this approach is established by a simulation that shows estimates to be close to target quantities for weights based on sampling probabilities or time intervals between samples. Recommendations are given for suitable effective sample sizes to reliably determine interval estimates.

Sparse-data bias accompanying overly fine stratification in an analysis of beryllium exposure and lung cancer risk

PURPOSE Berylliums classification as a carcinogen is based on limited human data that show inconsistent associations with lung cancer. Therefore, a thorough examination of those data is warranted. We reanalyzed data from the largest study of occupational beryllium exposure, conducted by the National Institute of Occupational Safety and Health (NIOSH). METHODS Data had been analyzed using stratification and standardization. We reviewed the strata in the original analysis, and reanalyzed using fewer strata. We also fit a Poisson regression, and analyzed simulated datasets that generated lung cancer cases randomly without regard to exposure. RESULTS The strongest association reported in the NIOSH study, a standardized rate ratio for death from lung cancer of 3.68 for the highest versus lowest category of time since first employment, is affected by sparse-data bias, stemming from stratifying 545 lung cancer cases and their associated person-time into 1792 categories. For time since first employment, the measure of beryllium exposure with the strongest reported association with lung cancer, there were no strata without zeroes in at least one of the two contrasting exposure categories. Reanalysis using fewer strata or with regression models gave substantially smaller effect estimates. Simulations confirmed that the original stratified analysis was upwardly biased. Other metrics used in the NIOSH study found weaker associations and were less affected by sparse-data bias. CONCLUSIONS The strongest association reported in the NIOSH study seems to be biased as a result of non-overlap of data across the numerous strata. Simulation results indicate that most of the effect reported in the NIOSH paper for time since first employment is attributable to sparse-data bias.

Confounding after Risk-Set Sampling in the Beryllium Study of Sanderson et al.

PURPOSE Beryllium is classified as carcinogenic on the basis largely of limited human data showing a modest increase in lung cancer from occupational exposure. With occupational exposure now curtailed, earlier results merit more scrutiny. We simulated data to understand the design implications of a landmark case-control study. METHODS We generated datasets from the original occupational cohort by randomly assigning lung cancer events to workers independently of their exposure. We analyzed the simulated data on the basis of different modes of risk-set sampling, with risk sets defined by calendar time, age, or both, to assess how much bias existed using several exposure metrics. We controlled for several time related variables to assess confounding. Finally, we re-analyzed the data from the original study, controlling for time-related covariates. RESULTS No bias occurred using any type of risk-set sampling with unlagged exposures. When exposure was lagged 10 or 20 years, however, there was considerable confounding by year of birth and year of hire, which remained uncontrolled in the original study. CONCLUSIONS Simulations and reanalysis show that much of the reported association with lagged exposure is attributable to confounding by year of birth and year of hire. Lagging changes the exposure variable and can thus lead to changes in the amount of confounding.

Population coverage and nonresponse bias in a large-scale human exposure study

We used estimates derived from screener variables of the National Human Exposure Assessment Survey (NHEXAS) Phase I field study in EPA Region V (one of three NHEXAS Phase I field studies) to examine biases resulting from survey nonresponse and/or incomplete population coverage inherent in the study design. For variables with population values obtainable from Census projections, the combined effect of nonresponse and coverage bias was tested for after each stage of nonresponse using design-based weights. For variables where population values were not available as Census projections, nonresponse bias was tested for after the screener stage of nonresponse using weights adjusted for screener nonresponse. Additional tests for bias were performed using final survey weights to evaluate the performance of survey weight adjustments in reducing observed bias. Comparison of biases estimated using both design-based and adjusted weights was used to identify potentially important weight adjustment variables for future exposure studies, identify possible weaknesses in survey design strategies, and support the use of nonresponse and poststratification weight adjustments to reduce bias in future survey studies.

Flow-covariate prediction of stream pesticide concentrations

Potential peak functions (e.g., maximum rolling averages over a given duration) of annual pesticide concentrations in the aquatic environment are important exposure parameters (or target quantities) for ecological risk assessments. These target quantities require accurate concentration estimates on nonsampled days in a monitoring program. We examined stream flow as a covariate via universal kriging to improve predictions of maximum m-day (m = 1, 7, 14, 30, 60) rolling averages and the 95th percentiles of atrazine concentration in streams where data were collected every 7 or 14 d. The universal kriging predictions were evaluated against the target quantities calculated directly from the daily (or near daily) measured atrazine concentration at 32 sites (89 site-yr) as part of the Atrazine Ecological Monitoring Program in the US corn belt region (2008-2013) and 4 sites (62 site-yr) in Ohio by the National Center for Water Quality Research (1993-2008). Because stream flow data are strongly skewed to the right, 3 transformations of the flow covariate were considered: log transformation, short-term flow anomaly, and normalized Box-Cox transformation. The normalized Box-Cox transformation resulted in predictions of the target quantities that were comparable to those obtained from log-linear interpolation (i.e., linear interpolation on the log scale) for 7-d sampling. However, the predictions appeared to be negatively affected by variability in regression coefficient estimates across different sample realizations of the concentration time series. Therefore, revised models incorporating seasonal covariates and partially or fully constrained regression parameters were investigated, and they were found to provide much improved predictions in comparison with those from log-linear interpolation for all rolling average measures. Environ Toxicol Chem 2018;37:260-273.

Evaluation of the use of bias factors with water monitoring data

Aquatic exposure assessments using surface water quality monitoring data are often challenged by missing extreme concentrations if sampling frequency is less than daily. A bias factor method has been previously proposed to address this concern for peak concentrations, where a bias factor is a multiplicative quantity to upwardly adjust estimates so that the true value is exceeded 95% of the time. In other words, bias factors are statistically protective adjustments. We evaluate this method using a research data set of 69 near-daily sampled site-years from the Atrazine Ecological Monitoring Program, dividing the data set into 23 reference and 46 validation site-years. Bias factors calculated from the reference data set are used to evaluate the method using the validation set for 1) point estimation, 2) interval estimation, and 3) decision-making. Sampling designs are every 7, 14, 28, and 90 d; and target quantities of assessment interest are the 90th and 95th percentiles and maximum m-day rolling averages (m = 1, 7, 21, 60, 90). We find that bias factors are poor point estimators in comparison with alternative methods. For interval estimation, average coverage is less than nominal, with coverage at individual site-years sometimes very low. Positive correlation of bias factors and target quantities, where present, adversely affects method performance. For decision rules or screening, the method typically shows very low false-negative rates but at the cost of extremely high false-positive rates. Environ Toxicol Chem 2018;37:1864-1876.