239006 Assessing performance of Q-statistics for identifying cancer clusters in residential histories using simulated clusters in US and Danish case-control studies

Tuesday, November 1, 2011: 1:20 PM

Chantel Sloan, PhD , School of Medicine, Vanderbilt University, Nashville, TN
Geoffrey Jacquez, PhD , BioMedware, Inc., Ann Arbor, MI
Carolyn M. Gallagher, MPA, MPH , PhD Program in Population Health and Clinical Outcomes Research, Department of Preventive Medicine, SUNY Stony Brook, Stony Brook, NY
Mary H. Ward, PhD , Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD
Rikke Baastrup Nordsborg, MSc , Institute of Cancer Epidemiology, Danish Cancer Society, Copenhagen, Denmark
Ole Raaschou-Nielsen, PhD , Institute of Cancer Epidemiology, Danish Cancer Society, Copenhagen, Denmark
Jaymie R. Meliker, PhD , Graduate Program in Public Health, Stony Brook University Medical Center, Stony Brook, NY
Few cancer clustering investigations have evaluated residential mobility even though exposure to environmental carcinogens may occur decades before a cancer diagnosis. Recently developed Q-statistics can be used to investigate disease clusters based on mobility histories by quantifying space- and time-dependent nearest neighbor relationships. Qik identifies which individuals are centers of spatial clusters through time; and Qikt shows where and when local clustering takes place. These statistics are calculated repeatedly through space and time and therefore face the problem of multiple testing in determining statistical significance. Using case-control residential histories from 2378 participants in a US study and 6594 participants from a Danish study, we created a series of simulated clusters to examine Q-statistic performance and statistical significance. Results suggest the intersection of cases with significant clustering over their life course, Qik, with cases who are constituents of significant local clusters at given times, Qikt, allows us to identify simulated clusters. Using this intersection approach, Q-statistics identified large true positive cluster regions with few false positives; true smaller cluster regions were difficult to differentiate from false positives. Specifically, if three or more significant (Qik, p=0.001, Qikt, p=0.05) cases are detected in the same cluster region then it may be considered a true cluster. Setting k-nearest neighbors equal to 10 or 15 consistently showed strong performance. This approach has potential for identifying clustering in mobile populations but given differences in mobility patterns, future work is required to investigate generalizability of this rule set to other case-control datasets.

Learning Areas:
Chronic disease management and prevention
Public health or related research
Social and behavioral sciences

Learning Objectives:
Explain why residential histories may be important for studying cancer clusters. Assess the value of Q-statistics for studying clustering in residential histories of case-control studies. Discuss sensitivity of Q-statistics to key parameters, including cluster size, cluster density, population mobility, and choice of k-nearest neighbors.

Keywords: Geographic Information Systems, Epidemiology

Presenting author's disclosure statement:

Qualified on the content I am responsible for because: I am PI on two NIH-funded grants investigating new methods for finding cancer clusters in mobility histories; this work is a product of these two projects.
Any relevant financial relationships? No

I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.

See more of: Cancer Epidemiology
See more of: Epidemiology