NCI’s Surveillance, Epidemiology and End-Results (SEER) program includes a set of population-based registries that collect data on all cancer cases occurring in defined geographic areas. From 1992 through 1999, SEER data covered 14% of the US population; coverage increased to 26% in 2000.8
SEER collects detailed information on tumor characteristics, demographics, and treatment and maintains follow-up of all registered cases. The data are primarily abstracted from hospital records, surgical centers and radiation facilities. Because adjuvant therapy is most often provided in an outpatient setting, NCI annually selects certain cancer sites and supplements the routine data abstraction with patterns of care studies. In addition to re-abstracting the data, each patient’s physician was asked to verify the treatment provided. These physicians also were asked whether other physicians might have provided care and if so, the newly identified physicians were then contacted. To assure consistency of abstracting and coding, the supervising abstractor from each registry attended a central training.
The current data include white patients who were age 20 or older and diagnosed with local or regional melanoma in 1995, 1996, or 2001. Patients were ineligible if they had previously been diagnosed with cancer other than non-melanoma skin, were diagnosed on the death certificate or at autopsy, or were diagnosed simultaneously with a second cancer. African-American patients were not sampled because of their low incidence rate of melanoma. The nodal biopsy technique performed on each patient was independently abstracted from their medical record and did not rely on information provided by cancer registry data. Patients were identified as having a sentinel lymph node biopsy if the medical record provided any indication of the procedure and was recorded regardless of any subsequent lymph node dissection (e.g. complete nodal dissection) performed. In 1995, 1996, and 2001 a total of 262, 751, and 1200 patients, respectively, were identified from participating registries. The registries participating in the patterns of care studies changed over time. Because this study focused on trends over time, we limited the registries to those participating in at least two of the three years (the metropolitan areas of San Francisco, San Jose/Monterey, Atlanta, Detroit, Seattle, Los Angeles County, and the states of Connecticut, Iowa, New Mexico, and Utah). This excluded an additional 479 cases.
Bivariate analyses were performed to determine the association between certain tumor characteristics and clinical, non-clinical, treatment variables, as well as the year of diagnosis. All estimates were weighted to reflect the population from which the sample was drawn. The sample weights, calculated as the inverse of the sampling proportion for each sampling stratum (defined by SEER registry, white Hispanic / white non-Hispanic ethnicity and sex), were used to obtain estimates that are representative of all eligible Caucasians melanoma patients in the study areas. We used the statistical software SAS and SUDAAN (Research Triangle Institute, Research Triangle Park, NC). The SUDAAN software allows for use of sample weights and adjusts the standard errors appropriately.
Follow-up was available through December 31, 2007. In a multivariate Cox proportional hazard model we examined melanoma-specific mortality by sex adjusting for age group, Charlson Score, tumor thickness, margins, and year of diagnosis. All tests were two-sided.