Screening geographic areas for unusual survival experience or stage at diagnosis with application to breast and colon cancer

An approach was devised for the problem of detecting geographic areas with poor survival for cancer patients. Methods were proposed to deal with the two major sources of variability in summary survival measures for geographic areas: the distribution of prognostic factors for individuals within an area and random variation. In addition, small geographic areas often had to be combined so that a sufficient number of cases could be obtained in each area for calculation of survival statistics, and a procedure for doing so was developed. Then stage-specific and overall 5-year survival estimates for each geographic area were converted to standardized normal deviates so that outlying observations could be detected using the theory of normal-order statistics. A similar approach was used to determine which geographic areas had unusually high proportions of advanced disease at diagnosis. Our outlier detection procedure was designed for screening available data for geographic areas possibly deserving further study rather than for concluding that survival or staging was substandard in those areas. Methods were applied to data from the Surveillance, Epidemiology, and End Results (SEER) Program for breast and colon cancer. For each cancer site, some geographic areas were identified with unusual survival or stage distributions.