Risk factors for cancer from agricultural exposures in the ambient environment are studied to estimate the contribution of these environmental exposures to cancer in the rural population. Rural populations in agricultural areas have exposures to pesticides and nitrates in drinking water that are intermediate between occupationally-exposed groups and the general population. Case-control studies have been conducted on non-Hodgkin's lymphoma and cancers of the brain, bladder, colon, rectum, stomach, esophagus, pancreas, and kidney. Databases of water contaminants, gathered for routine monitoring purposes, are being used to estimate past exposures via public water supplies to individuals in these case-control studies. Nitrate levels in public water supplies in Iowa were not associated with risk of bladder and pancreas cancer. However, higher intakes of dietary nitrite were associated with an increased risk of pancreas cancer. Overall, drinking water nitrate levels were not associated with risk of colon cancer; however, specific subgroups with increased nitrosation ability were at increased risk. Elevated nitrate levels in public drinking water supplies were associated with an increased risk of non-Hodgkin's lymphoma in Nebraska. An analysis of nitrate in drinking water in Minnesota and Sardinia, Italy where exposure levels were lower found no association. Elevated nitrate in public water supplies in Nebraska was associated with an increased risk of glioma in women but not men. This hypothesis is being investigated further in a recently completed case-control investigation of non-Hodgkin's lymphoma in four centers: Iowa, Seattle, Los Angeles and the Detroit area. The issue of nitrate as a cancer risk factor is being pursued further in an ecologic study of census-block groups in the Platte River Valley of Nebraska and Colorado, where elevated nitrate levels are among the highest in the country. Nitrate levels in private wells are being estimated using data on land cover, livestock feedlot locations, soil type, and other information. We are also developing new approaches, and improving existing methods, of exposure assessment in studies of general environmental exposures, with a primary focus on pesticides and nitrate. These better methods are required to better estimate risk and to detect the relatively small increases in risk often encountered in such studies. Geographic information systems (GIS) are being utilized for developing new approaches to estimating indirect exposure to pesticides and for estimating exposure to nitrate in private wells. Remote sensing data (satellite imagery) and historical records were used to estimate the population in Nebraska and Iowa potentially exposed to agricultural pesticides through the proximity of their homes to agricultural land treated with pesticides. First, a feasibility study in Nebraska demonstrated that using available data and a GIS accurate historical crop maps could be produced and that these could be linked to pesticide use data to estimate probabilities of indirect exposure to agricultural pesticides. A total of 24% of the study population in an agricultural area were determined to live within 500 meters of crop fields likely to have been treated with pesticides. This was the first study to estimate the prevalence of potential indirect exposure to agricultural pesticides in the general population. A similar approach was used for Iowa participants in the four center case control study of NHL. It was determined that 58% of the study population lived within 500 meters of corn or soybeans fields, which account for >90% of crop acreage. To validate this method, we compared agricultural pesticides measured in house dust samples to the proximity of homes from crop fields and found a significantly increased risk of herbicide detections in homes within 500 meters of corn or soybean fields. Further work in this area by our collaborators included the development of a method to automate crop mapping from satellite imagery. The method was found to be successful at classifying crops into two categories: corn and other crop types, over a 15 county area of Nebraska. We are continuing to refine the GIS-based approach to estimating agricultural pesticide exposure by incorporating information from a pesticide drift model. Additional analyses are underway in Iowa to compare agricultural pesticide levels in carpet dust samples with several GIS-based exposure metrics including one that incorporates information on wind direction and pesticide drift. With collaborators in California, we are planning to evaluate agricultural pesticide use near residences as a risk factor for childhood leukemia in an ongoing study. Carpet dust samples will be analyzed for pesticides, and information about the current and historical location of crops near residences will be determined. In a pilot study in the Central Valley of California, we compared a metric based solely on California's pesticide use reporting (PUR) database with a metric that incorporates information on the location of crops grown. A comparison of pesticide use within 500 meters of homes based on these two metrics indicates that the PUR metric classifies a substantially greater number of homes as """"""""exposed"""""""" compared with a crop-based metric. Using a GIS, residential proximity to specific industries, hazardous waste sites, and specific chemical releases as reported by the Environmental Protection Agency's Toxic Release Inventory will be evaluated in an ongoing case control study of bladder cancer in New England. As a first step, we evaluated the feasibility of geocoding past residential histories from a pilot study in New Hampshire to determine the historical locations of specific industries associated with bladder cancer risk. The location of residences and industries is usually determined by geocoding (estimating the latitude and longitude) of addresses, but the accuracy of this method has not been systematically evaluated. We evaluated the accuracy of geocoding methods for determining residential location in rural and urban areas of Iowa by comparing geocoded residential addresses to the global positioning system measurements of latitude and longitude taken at Iowa homes in the four-center NHL study. Overall, the median error of geocoded residential locations was approximately 50 meters; however, the median error increased to about 200 meters for rural addresses.
