New Study Designs for Interpretation of Chemical Exposure Data

Enrique F. Schisterman. Epidemiology Branch (EB), Division of Epidemiology, Statistics & Prevention Research (DESPR), Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD).

Either through lack of power or systematic design flaw, data collected using standard designs are often ill suited to investigate complex exposures and outcomes adequately. Methods have been developed to analyze complex chemical mixtures and attempt to account for different sources of bias and unnecessary variability from data using traditional designs. However, the effectiveness of these techniques is limited by the data collection process, where new study designs would lead to data better suited to address the challenges posed by complex chemical mixtures. More suitable designs might be able to limit more complex statistical methods in lieu of more straight-forward techniques. These types of advancements in the study design phase will improve our ability to quantify relations between mixtures of chemical exposures and outcomes under the constraints of cost and complexity.

We have developed state of the art study designs to evaluate the impact of various types of chemical exposures on human health effects. These study designs include case-time-control, case-crossover, two-stage, and the hybrid (pooled-unpooled) designs. Our aim is to explore and maximize already stated benefits of each while broadening their use for the investigation of biochemical exposures. Our emphasis is on limiting difficulties specific to biochemical exposures, such as high cost of sample procurement and measurement leading to reduced power, intermittent exposures leading to risk of acute events, and accounting for the effects of multiple types of measurement error, specifically limits of detection and random measurement error. Scientific manuscripts have been written and submitted for peer review for publication in a special issue of the journal Statistics in Medicine as well as others. These papers document the epidemiological and statistical issues, offer statistical approaches for obtaining valid parameter estimates along with confidence intervals, and empirically demonstrate the utility of the proposed methodology.

Implications

This study continues to offer DESPR investigators and external consultants an opportunity to develop new and innovative design approaches for epidemiological studies of chemical exposures that account for the difficulties of high cost of sample procurement and measurement, measurement error, and correlated exposures. New designs incorporate biomonitoring when possible to improve patient safety while conducting informative and efficient investigation of exposures and potential human health risks. These new study designs provide researchers with tools to investigate chemical exposures and associated outcomes with increased power, efficiency, and reliability, while remaining fiscally responsible. Employing these innovative designs for complicated chemical exposures data will enable improved decision making by researchers and policy-makers.

Keywords

chemical exposures, pooling, efficient study design, measurement error, case-crossover

Project Start and End Dates

2009 – 2010

Project ID

MTH0806

Peer-reviewed Publication(s)

Schisterman, E. F. and Albert, P. Editorial. Statistics in Medicine Special Issue. (Submitted).

Liu, A., Liu, C., Malinovsky, Y., Zhang, B., and Zhang, Z. Confidence intervals of a small probability derived from group testing with misclassification. Statistics in Medicine Special Issue. (Submitted).
Lyles, R. H., Lin, J., Tang, L., Zhang, Z., and Mukherjee, B. Likelihood-based methods for regression analysis of binary outcomes with binary exposure status assessed by pooling. Statistics in Medicine Special Issue. (Submitted).

Pfeiffer, R. M., Forzani, L., and Bura, E. Sufficient dimension reduction for longitudinal data. Statistics in Medicine Special Issue. (Submitted).

Roy, A., Danaher, M., Mumford, S., and Chen, Z. Bayesian order restricted inference for hormonal dynamics. Statistics in Medicine Special Issue. (Submitted).

Schildcrout, J. S., Mumford, S., Chen, Z., Heagerty, P. J., and Rathouz, P. A semi-parametric estimation strategy for outcome and time dependent sampling of longitudinal binary data. Statistics in Medicine Special Issue. (Submitted).

Vexler, A., and Tsai, W. Estimation and testing based on data subject to measurement errors: From parametric to non-parametric likelihood methods. Statistics in Medicine Special Issue. (Submitted).

Whitcomb, B. W., Perkins, N. J., Zhang, Z., and Ye, A. Regression analysis for hybrid pooled-unpooled case-control studies with gamma distributed exposure. Statistics in Medicine Special Issue. (Submitted).

Other publication(s)

Malinovsky, Y., Albert, P. S., and Schisterman, E. F. Pooling strategies for outcome under a gaussian random effects model. Biometrics. (In revision).

Abstract revision date

May 2011

*This abstract was prepared by the principal investigator for the project.

News

News & Resources

View our resource center to find press releases, testimonies, infographics and more.

Jobs
TSCA