Using Residential Histories to Investigate the Role of Segregation in Racial Cancer Inequities

Background: Segregation and neighborhood socioeconomic disadvantage are associated with racial disparities in cancer prognosis and survival. However, limited availability of historical data means that few epidemiologists have been able to study the health effects of experiencing either segregation or neighborhood disadvantage in the past. Objective: Using available sources of residential history data in people with cancer, estimate past exposure to neighborhood concentrated disadvantage and evaluate its association with stage at diagnosis and survival. Methods: Two complementary sources of residential history data were used. In the Panel Study of Income Dynamics, residential histories were obtained from geocoded addresses obtained over the course of study participation. Participants who reported an adult history of cancer were included, and their vital status was obtained from study records. Past addresses were linked to a database of Census-tract level concentrated disadvantage scores. Cox proportional hazards models were used to evaluate the association between neighborhood disadvantage before diagnosis and length of survival after diagnosis. In the UI Hospital and Health System tumor registry, all analytic cases of colorectal cancer diagnosed between 1995-2004 were selected. Vital status and cause of death were ascertained through linkage to the National Death Index. Multinomial logistic models were used to evaluate the association between past neighborhood disadvantage and cancer stage. Cox proportional hazards models were used to evaluate the association between past neighborhood disadvantage and cancer-specific survival. Results: Once each dataset was prepared, there were no racial inequities in stage at diagnosis, all-cause mortality, or cancer-specific mortality among the included cases. This is consistent with prior research indicating that residential histories are less complete and available in people who are not white, who have lower incomes, or who have died. Past neighborhood disadvantage did not predict all-cause survival in the survey sample or cancer stage in the tumor registry sample. Both past and at-diagnosis disadvantage predicted survival in the tumor registry sample, but residential history data did not improve prediction. Conclusion: Residential history data are challenging to collect and interpret, and may introduce bias into observational studies when they are less complete and available in some cases. Researchers should carefully assess the risk of data loss and bias before conditioning study inclusion on the availability of these data. However, the Panel Study of Income Dynamics is a rich source of information about social determinants of health. Similar approaches could be more successful in the study of health conditions or behaviors that are more common than cancer.