Given our concerns we decided to make two estimates (Method A and Method B). In Method A, for those sites for which the American Indian rates were "best," we chose the race/ethnic group that was the next lowest in death rate for that cancer site. In Method B, we used American Indian rates when they were the "best." Two sets of tables are provided, but we strongly prefer those that are not based on American Indians as the "best'' or target rate (Method A).
While Method A are preferred estimates, this method does not make any adjustment for known race misclassification. Overall we expect that the estimated number of "excess" deaths from cancer would be around ten percent fewer for women and closer to twenty percent for men. This percentage would vary by cancer site.
Please note that I strongly believe this exercise can be useful. As I mentioned on the phone, I view differences in cancer mortality risk as, first and foremost, reflecting a variety of cancer causes and differential group responses to risk factors. Illustrating the differences between groups provides clues to form hypotheses on cancer causes, risk factors or biology. Secondly, for some cancer sites, we understand the causes and processes of cancer development sufficiently to be able to say how one group can achieve the lower risk shown by another group. Perhaps the best example I know of in the second case is the low rates of cancer among Mormons in Utah and Los Angeles, which reflect life-style choices that could be replicated in other populations.
All of us at NCHS are appreciative to our NCI colleagues who made available their SEER database and helped us make the calculations.
Please contact Mr. Maurer or Drs. Hoyert or Rosenberg (301-436-8884) if you have any additional questions.
Edward J. Sondik, Ph.D.
National Center for Health Statistics