Geocoding Approaches and Breast Cancer Risk: A Comparative Analysis

1. A comparison of two geocoding approaches to estimate the effect of residential ambient pesticide exposure on breast cancer risk Laura Thompson, Loraine Escobedo, Carrie Tayour, Dan Goldberg, Bryan Langholz, Myles Cockburn Supported by: U.S. National Cancer Institute grant CA195218, National Institute of Environmental Health Sciences grants ES019986 and ES018960.

2. Prior research has found elevated risk for certain cancers associated with residential ambient pesticide exposure1-3, but this exposure continues to be difficult to measure. It is difficult to assess personal exposure to pesticides via questionnaires (because people are unlikely to know what they have been exposed to). Because California provides Pesticide Use Reports4, we can use historical address data to determine exposure. Residential exposures are characterized by low-level, long-term exposures, with most relevant exposures likely occurring in the distant past. It is not always possible to collect complete address data and additional spatial reference information. Ambient Pesticide Exposure: Exposure to pesticides that persist in the environment after application.

3. The established method for capturing ambient pesticide exposure consists of a GIS-based approach5-6. • This approach relies on the geolocation of residential and/or occupational address data. • Geocoders typically return a pair of lat/long coordinates representing the centroid of the smallest known area. • Possible area matches include parcels, street segments, ZCTA/USPS ZIP codes, cities, county sub-regions, and counties. • Misclassification of exposure due to geolocation uncertainty can affect risk estimates. • Need to account for geocode certaintyin these spatial temporal studies.

4. Solution: Create a geocoder that can return the smallest area of certainty for each address. Point-Based Area-Based

5. Residential Histories: Breast Cancer in Fresno, Kern, and Tulare Counties, California. • It is not ideal to remove older, lower quality address data. • Area-based methods allow us to account for this uncertainty.

6. Breast Cancer cases point and area-based geocoding outputs. The area-based approach accounts for the locational uncertainty inherent with self-reported address information, and thus it will provide an exposure assessment that is more representative of the locational information we have. We have maintained the appropriate level of uncertainty.

7. Relative Risk: The two geocoding methods resulted in different relative risk estimates. aAdjusted for age, socioeconomic status (quintiles), body mass index (<25, 25–29, or ≥30 kg/m2), age at menarche (<12, 12, or >12 years), age at menopause (<45, 45–54, or ≥55 years), number of births (0, 1, 2, or ≥3), oral contraceptive use (none, 1–5 years, or ≥5 years), menopausal hormone therapy use (none, estrogen only, progesterone only, estrogen plus progesterone, or a mixture of treatments), ever consumed alcohol at least once a week (yes or no), vigorous physical activity (0, 1–6, 7–13, or ≥14 hours per week), and number of years lived in Fresno, Tulare, or Kern counties. cAldrin, chlordane, dicofol, dieldrin, endosulfan, lindane, methoxychlor, and toxaphene. *Exposed at both residences and workplaces. 

8. Relative Risk: The two geocoding methods resulted in different relative risk estimates. aAdjusted for age, socioeconomic status (quintiles), body mass index (<25, 25–29, or ≥30 kg/m2), age at menarche (<12, 12, or >12 years), age at menopause (<45, 45–54, or ≥55 years), number of births (0, 1, 2, or ≥3), oral contraceptive use (none, 1–5 years, or ≥5 years), menopausal hormone therapy use (none, estrogen only, progesterone only, estrogen plus progesterone, or a mixture of treatments), ever consumed alcohol at least once a week (yes or no), vigorous physical activity (0, 1–6, 7–13, or ≥14 hours per week), and number of years lived in Fresno, Tulare, or Kern counties. cAldrin, chlordane, dicofol, dieldrin, endosulfan, lindane, methoxychlor, and toxaphene.  *Exposed at both residences and workplaces. 

9. Why do we observe differences by pesticide? Pesticide-specific changes in ORs may be explained by differences in distribution of different pesticides (i.e., proximity of application to population centers). For example, the widespread and frequent use of Chlorpyrifos may explain why the difference in cases and controls is diminished when using an area-based approach. -12.7% Change 58.1% Change

10. Conclusions and Implications: Next Steps: Accounting for geocoding uncertainty using area-based geocoding results in different relative risk estimates than traditional methods. The degree of this difference varies by pesticide.  Area-based geocoding accounts for the uncertainty in automated geocoding in the determination of exposure and should continue to be explored, especially in circumstances when addresses cannot be manually resolved. Further refine locations within relative large, low geocode certainty areas (i.e., cities and ZCTAs) by using census blocks to determine where individuals live within a polygon and calculating exposure for residential areas (Langholz et al. 2019, under review). Determine an area threshold above which the geocode uncertainty is too high to include an address in exposure assessment.

11. Thank you!Questions? References: Key T, Appleby P, Barnes I, Reeves G, Endogenous Hormones and Breast Cancer Collaborative Group. 2002. Endogenous sex hormones and breast cancer in postmenopausal women: reanalysis of nine prospective studies. J Natl Cancer Inst 94: 606–616, https://doi.org/10.1093/jnci/94.8.606. Key T, Pike MC. 1988. The role of oestrogens and progestagens in the epidemiology and prevention of breast cancer. Eur J Cancer Prev 24: 29–43, https://doi.org/10.1093/jnci/94.8.606. Soto AM, Sonnenschein C. 2010. Environmental causes of cancer: endocrine disruptors as carcinogens. Nat Rev Endocrinol 6:363–370, https://www.nature.com/articles/nrendo.2010.87. CalEPA, California Department of Pesticide Regulation. 2013. Pesticide Use Reporting Pesticide Information Portal (CalPIP). Available: http://www.cdpr.ca.gov/docs/pur/purmain.htm [accessed 11 June 2013]. Rull RP, Ritz B. 2003. Historical pesticide exposure in California using pesticide use reports and land-use surveys: an assessment of misclassification error and bias. Environ Health Perspect 111:1582–1589, https://doi.org/10.1289/ehp.6118. Costello S, Cockburn M, Bronstein J, Zhang X, Ritz B. 2009. Parkinson's disease and residential exposure to maneb and paraquat from agricultural applications in the central valley of California. Am J Epidemiol 169:919–926, https://doi.org/10.1093/aje/kwp006. Langholz B, Escobedo L, Goldberg D, Heck J, Thompson L, Ritz B, Cockburn M. 2019. Analysis of case-control data when there is geolocation uncertainty. Scand J Stat [under review June 2019].