Subject loss in spatial analysis of breast cancer

Possible selection bias from assignment of latitude-longitude coordinates to the place of residence of all Connecticut women diagnosed with breast cancer from 1992 to 1995 (N=11,470) was evaluated. Exact address-matching was accomplished for 8,121 records (70.8%) and an additional 1,722 records (15.0%) were matched using relaxed criteria. We did not address-match 1,627 records (14.1%) due to missing address information or limitations of the Geographic Information Systems street file. The age-adjusted likelihood of address-matching records was significantly greater for women of color: those born within Connecticut, residents of urban locales or census tracts with low median family incomes and those cases diagnosed nearer to 1992. Few differences in address-matching were attributable to tumor characteristics or therapeutic modality. (C) 1999 Elsevier Science Ltd. All rights reserved.