This analysis is based on data from the Surveillance, Epidemiology, and End Results (SEER)–Medicare file. These data include Medicare claims for persons with cancer residing in SEER program areas (states of Connecticut, Hawaii, Iowa, New Mexico, and Utah and the metropolitan areas of Detroit, San Francisco, Atlanta, Seattle, Los Angeles, and San Jose). At the time of this analysis, data were available for individuals diagnosed with cancer from 1992 to 1999 with Medicare claims data available through 2001. The SEER program sites collect information on all incident cancer cases. These data have been linked by the National Cancer Institute to Medicare claims. A restricted access version of these data was obtained so that the characteristics of each individual’s census tract of residence, from the 1990 U.S. Census, could be appended. This study was reviewed and approved by the Institutional Review Board of Brigham and Women’s Hospital.
We included individuals who had died from cancer between 1992 and 2000, who were at least 66 years of age at the time of death, and who had been diagnosed with breast, colorectal, lung, or prostate cancer before death. We excluded individuals less than 66 years of age to ensure access to at least a full year of Medicare claims. Because our analysis was focused on the role of residential characteristics, we excluded 1,016 individuals with a missing census tract identifier.
The sample was limited to individuals whose race/ethnicity was reported as white, African-American, or Hispanic. Although the racial designation of whites and blacks in these data is reasonably sensitive and specific,10
the designation of Hispanic ethnicity is less so. However, prior analyses have used Medicare data to identify individuals of Hispanic ethnicity.11,12
Our final dataset included 70,669 individuals.
The principal outcome was whether an individual had used hospice care during the 12 months before death. Because the Medicare benefit requires that hospice patients forgo curative treatments, individuals may disenroll or reenroll in hospice care. We therefore defined our outcome as any hospice claim, either at home or in an institutional setting, during this period.
Our principal independent variable was the racial composition of the area where each individual resided at the time of diagnosis, measured by the percentage of African-American and Hispanic residents within the census tract. We categorized areas as: (1) high African-American and high Hispanic, where each group was at least 30% of the population; (2) high African-American only, where the percentage of African-American residents was at least 30% and the percentage of Hispanics was less than 30%; (3) high Hispanic only, where the percentage of Hispanic residents was at least 30% and the percentage of African Americans was less than 30%; and (4) low African-American and low Hispanic, where the percentage of both groups was less than 30%.13,14
To test the robustness of our findings, we also used cut points of 25% and 35%.
Other census tract-level covariates included quintiles for median household income and the percentage of individuals who had graduated from college.
Individual-level independent variables included age (categorized as 66–74, 75–84, or at least 85 years), sex, race/ethnicity (white, African-American, or Hispanic), marital status (married or not married), site of cancer (breast, colorectal, prostate, or lung), stage at diagnosis (unstaged, local or regional, or distant), number of comorbid conditions (none, 1, 2, 3, or more), whether an individual was of “low income” (based on eligibility for state assistance with Medicare premiums and copayments), year of death, and indicators of whether someone was enrolled in a Medicare-managed care plan within the 13 months before diagnosis, at any time after diagnosis, or without Medicare coverage within 13 months before diagnosis (these indicators were used to adjust for individuals who may have less complete information about comorbid conditions). Information about hospice use is recorded for individuals with a Medicare-managed care plan as these services are paid directly by Medicare.