Sources of Selection and Information Biases When Using Commercial Database-Derived Residential Histories for Cancer Research.

Abstract Purpose This study examined potential sources of selection and information biases when using residence history information from a commercial database to construct residential histories for cancer research. Methods We searched the LexisNexis® database for residence data on 3473 adults diagnosed with cancer between 2005 and 2016 using name, address at diagnosis, and birth date. Residential histories were generated from the results using open-source statistical programs from the NCI. Multivariable regression models analyzed the associations of the search results with demographic characteristics and all-cause mortality. Results Racial/ethnic minorities were less likely to match to vendor residence data compared to non-Hispanic whites (N-HW) (odd ratios (ORs) [95% confidence intervals (CIs)] for non-Hispanic blacks, Hispanics, and Asian/Pacific Islander were 1.66 [1.30, 2.12], 2.92 [2.18, 3.90] and 4.53 [2.72, 7.55], respectively). Being non-Hispanic black was negatively associated with years of residential history (vs. N-HW, s coefficient [95% CI] = -2.57 [-3.40, -1.73]). Not matching to residence data was associated with an increased 5-year odds of death from any cause (vs. matched subjects, OR [95% CI] = 5.92 [4.29, 8.50]). Conclusions Differential ascertainment of residence history by race/ethnicity and association of ascertainment with prognosis are potential sources of selection and information biases when using residence data from a commercial database.