Project 2

Analyzing Patterns in Epidemiologic and Toxicologic Data

Developing improved methods for mapping epidemiologic and chemical mixture data on reproductive and developmental outcomes while adjusting for known risk factors.

Project Leadersheat map proj 2

Jon Levy, Project Leader
Boston University School of Public Health

Susan Korrick, Co-Leader
Brigham and Women’s Hospital, Harvard T.H. Chan School of Public Health

 Project Description

The mapping of routinely collected health data is now common, frequently leading to public demands for investigation of perceived “hot spots.” However, localized increases seen in these maps can vary over time and be caused by confounding of unmeasured risk factors. Epidemiologists have had, until recently, few methods that appropriately account for where and when people are exposed.  Project 2 will evaluate a valuable and novel statistical approach, tensor product smooths, which are not well established in environmental epidemiology. Tensor product smooths within a generalized additive model (GAM) framework can account for interactions of different units and scales, making it an effective approach for space-time analyses. We will also adapt the GAM to allow for oversampling from regions of sparse data by incorporating a sampling fraction weight. This is a potentially powerful approach to modeling areas of low population density which can otherwise be a substantial limitation in spatio-temporal analyses. We will apply tensor product smooths to geocoded individual level data to investigate birth defects in parts of Massachusetts and Rhode Island (formal collaboration with SRP Project 1), and Attention Deficit Hyperactivity Disorder (ADHD)-related behaviors in children born while their mothers were living near the New Bedford Harbor Superfund Site.

This spatial analysis method also offers a new approach to a different question: interaction in chemical mixtures. Toxicology studies are usually performed one chemical at a time, but real-world exposures such as those at Superfund sites usually include many chemicals that may interact in various ways. We will extend methods from space-time interactions in epidemiologic data to the analysis of chemical mixtures (in collaboration with Projects 3 and 5). We will then adapt the concepts and methods for toxicology and apply them to the same epidemiological datasets analyzed for space-time interactions. This will include mixed exposures to (1) tetrachloroethylene and trihalomethanes and (2) PCBs and metals. In collaboration with the Research Translation Core, we will make our open source code freely available and develop a user-friendly software tool in the R statistical package that runs the statistical models and maps the results to allow barrier-free access to these new tools by the research and practice community. Project 2 will address important methodological issues in analyzing mixtures and exposures to mixtures in epidemiological studies, a critical need in real world risk assessment of hazardous waste exposures.

Research and Findings from Previous Cycle: Geographic Information Systems now allow the use of analytic techniques in spatial epidemiology previously not feasible. As a result the mapping of routinely collected health data is now common and often provokes concern when patterns of disease rates appear to have “hot spots,” although it is well understood by epidemiologists that the results may be biased by failure to collect and control for many known risk factors that are unevenly distributed over the area of the map.

Project 2 participates in our interdisciplinary work with the New Bedford Harbor and surrounding communities. Learn more about the history of the harbor and our work there.

Summer 2016 State of the Science Update

Project 2 focuses on examining and visualizing patterns in epidemiologic data using Generalized Additive Models (GAMs). GAMs are flexible statistical models that allow us to describe relations and interactions between risk factors. We have applied GAMs in a variety of contexts, including for different types of health outcomes, as a tool to map spatial variations in exposures or health outcomes, and to evaluate the effects of chemical mixtures (interactions between exposures). We are currently using GAMs to investigate the relation between chemical and non-chemical stressor mixtures and health outcomes in the New Bedford Cohort, a study which was started by Dr. Susan Korrick as part of the Harvard Superfund Research Program. Outcomes of interest include ADHD-related behaviors in childhood and other maladaptive behaviors in adolescents. In recent work, we applied GAMs to assess spatial patterns of ADHD-related behaviors near New Bedford and determine whether chemical exposures or non-chemical factors explained these patterns. This work emphasized the value of spatial analysis in identifying high-risk populations and evaluating the role of risk factors. In addition, we have applied GAMs for epidemiologic analyses within the New Bedford Cohort. In recent preliminary analyses considering prenatal exposures to mixtures of neurotoxicants, we found that exposure-associated risk of maladaptive behavior in adolescents varied depending on maternal characteristics. The complex effects of mixtures of chemical and non-chemical stressors in determining maladaptive behavior would likely not have been discovered using more traditional epidemiologic models.

Resources and Packages


BUSRP Publications

This page is licensed under a Creative Commons attribution/share-alike license.