Superfund Research Program

Analyzing Patterns in Epidemiologic and Toxicologic Data

Project Leader: Veronica M. Vieira (University of California-Irvine)
Grant Number: P42ES007381
Funding Period: 1995-2017

Project-Specific Links

Final Progress Reports

Year:   2016  2004  1999 

Routinely collected disease data are often mapped by town or county. Disease registries typically record residence at diagnosis, but this can obscure important environmental clues because some diseases take years to develop. Since registries have information on very few risk factors, an area of elevated risk may be due to the presence of many people with an unrecorded risk factor, e.g., smoking. Attempts to relate town disease rates to average exposures in the town can produce very misleading results. Maps based on suitably conducted studies of individuals can solve these problems.

Dr. Ozonoff and his research team have developed statistical methods to map individual level data while accounting for known risk factors. After testing the method using synthetic data, they investigated the association between residential history and colorectal, lung, and breast cancer on Upper Cape Cod, Massachusetts. Rather than causing “hot spots” to disappear, adjusting for known risk factors in these data sometimes revealed them. Maps of colorectal cancer were relatively flat. Assuming 15 years of latency, lung cancer was significantly elevated northeast of the Massachusetts Military Reservation (MMR), a Superfund site. Breast cancer hot spots tended to increase in magnitude with increased latency (number of years from first exposure). The basic pattern for breast cancer remained even after taking into account multiple residences. Significant hot spots were located near two pollution plumes and the MMR. Since the plumes likely did not have the same positions during the exposure period (assuming latency) and subjects variously used private wells or public water, this concordance does not establish exposure. This hypothesis will be tested by constructing a detailed individual-level exposure model using information from the case-control study.

In the last year project investigators further refined their previous analyses. They tested for possible screening bias by analyzing whether controls had undergone mammography, adjusting for age and first degree relative with breast cancer. The resulting mammography map was flat, suggesting no spatial screening bias for breast cancer. They also analyzed the cancer data using two cluster-detection procedures: SatScan (which locates the single most likely circular cluster or “hole”) and the M statistic (a global test for clustering). These two statistics cannot adjust for covariates as they are currently implemented for point data; the investigators therefore made comparisons to their results using unadjusted data. SatScan and their mapping method typically located hot or cold spots in the same location. However, SatScan performs less well when there is more than one such spot or if it is not circular. Since their mapping method smoothes data at a fixed scale for each map, it may miss or dilute effects at very small scales.

Data from Project 1’s current cohort study of reproductive/developmental outcomes on Cape Cod were recently geocoded. The researchers have so far analyzed birth weight, both as a continuous outcome and as low birth weight (<2500 g, the standard cutoff). Neither map was flat as judged by statistical testing. Adjustment for standard individual-level risk factors—such as maternal smoking—had little effect. This result indicates that there is unexplained geographical variation in the risk of low birth weight in the study area.