We used cancer surveillance data from 17 states/regions that are a part of the SEER Program, including the states of California, Connecticut, Georgia, Hawaii, Iowa, Kentucky, Louisiana, New Jersey, New Mexico, and Utah; and, the metropolitan areas of Detroit and Seattle. The cancer cohort included individuals who were diagnosed from January 1, 2000 to December 31, 2009 with follow-up through December 31, 2010. We excluded cancer patients whose initial date of diagnosis was derived solely from death certificates or at autopsy and patients who were alive but without sufficient follow-up time for analysis (eg, survival time of 0 days), and restricted analyses to the only cancer (sequence 00) or the first of multiple primary diagnoses (sequence 01).

Male and female cancer patients of all races and ages were included if they were diagnosed with a primary malignant cancer as defined by the International Classification of Diseases for Oncology Second Edition, Third Edition (ICD-O-3) (18) for site and histology and SEER Behavior Recode for Analysis for malignancy (http://seer.cancer.gov/behavrecode/). SEER Behavior Recode for Analysis was used to account for changes in behavior coding, starting with 2001 diagnoses. We chose SEER Behavior Recode for Analysis of 3 (malignant), which meant that cancers were classified as malignant in both ICD-O-3 and its predecessor ICD-O-2. SEER Site Recode, which is based on ICD-O-3 site and histology definitions, were used to identify the four specific cancer sites (http://seer.cancer.gov/siterecode/icdo3_dwhoheme/index.html).

We examine the underlying characteristics of each of the 12 states and 2 regions (Detroit and Seattle) with SEER registries, identifying potential root causes of in variations survival estimates, by race/ethnicity, age, cancer staging and mortality, and tobacco use using several sources. Median age by sex and race distributions were collected from the 2000 Census (19) for each state and region. Cancer death rates and the percent of cancer cases diagnosed at localized stage for each region/state were obtained using SEER*Stat (20,21). Mortality data included deaths from any malignant neoplasm from 2000 to 2009 for both sexes and all races. Rates per 100 000 were age-adjusted to the 2000 US Standard Population. The percent of malignant cancers diagnosed at localized stage was identified using SEER Summary Stage 2000, and are defined as cancers that are confined to the original organ site (22). Finally, we used 2000 data from the Centers for Disease Control and Prevention’s (CDC) Behavioral Risk Factor Surveillance System (BRFSS) to describe the prevalence of tobacco use (23), which is associated with higher morbidity and mortality rates in the underlying population (9). BRFSS surveys are a cross-sectional, state-based cluster sample of individuals identified via random-digit dialing. Prevalence of tobacco use was defined as the percent of respondents who smoked at least 100 cigarettes in their lifetime and were current smokers at the time of the survey.

Cancer registries use an algorithm to assign a single, disease-specific, underlying cause of death using information from death certificates. In some cases, attribution of a single cause of death may be difficult and misattribution may occur (eg, a death may be erroneously coded to the site of the metastasis). To capture deaths related to the specific cancer but not coded as such, SEER created an improved algorithm to calculate cause of death, the SEER cause-specific death classification variable (http://seer.cancer.gov/causespecific/) that takes into account causes of deaths in conjunction with tumor sequence (ie, only one tumor or the first of subsequent tumors), site of the original cancer diagnosis, comorbidities [eg, AIDS and/or site-related diseases (3)].

In this study, we used two sets of life tables: US-based life tables (USLT) and state-specific life tables (SLT). Both sets of life tables were constructed using decennial life tables from the NCHS and an interpolation method to calculate life tables for the individual years between the census years. The USLT have been the default life tables used in SEER*Stat to calculate relative survival for the SEER registries. The method used to derive them is described in detail at http://seer.cancer.gov/expsurvival/. Recently NCHS published 1999–2001 decennial SLT for all races, whites, and, for most states, blacks (17). Life tables for blacks in Hawaii and Utah were not available because of extremely small black populations.

Relative survival was calculated by actuarial methods as the ratio of all-cause survival and expected survival. Although relative survival could be adjusted when relative cumulative survival increased from a prior interval or was over 100%, we report unadjusted estimates in order to compare rates. Differences based on adjusted estimates were similar to the unadjusted. Expected survival is estimated using both the USLT and SLT by matching the survival cohort on age, sex, and race and year. Life tables for all races combined were matched to individuals with unknown or unspecified races. Individuals were matched by the state of registration to the specific SLT.

Cancer-specific survival (CS) was calculated using the SEER cause-specific death classification variable. The events of interest were defined as deaths attributed to the cancer using the SEER cause-specific death classification variable (http://seer.cancer.gov/causespecific/), and deaths due to other causes were censored. Significance was determined at P values less than .05 and 95% confidence intervals (CIs) are presented where appropriate. Analyses were conducted using SEER*Stat software.

Five-year cause-specific survival and relative survival estimates using SLT and USLT for males diagnosed from 2000 to 2009 are described in Table 2 for all cancers combined, colorectal, lung and bronchus, and prostate cancers. Overall, differences in relative survival were small. For all SEER 17 regions combined, relative survival based on SLT was approximately 1 (%) point lower than survival based on USLT for all cancers (−0.9), colorectal (−1.1), and prostate (−1.2). Utah and Connecticut had the most pronounced differences at −2.3 and −2.2 points, respectively. In the two states with the highest cancer death rates, Kentucky and Louisiana, SLT relative survival was statistically significantly higher than USLT relative survival: +1.1 and +1.2 points, respectively. The differences between SLT and USLT relative survival followed the same pattern for prostate cancer but the differences were more pronounced in Utah (−3.5), Connecticut (−3.3), and Louisiana (+2.8). There were no significant differences between SLT and USLT relative survival for lung cancer and few significant differences for colorectal cancer.

We saw similar patterns for women, although differences were generally smaller (Table 3). SLT-derived and USLT-derived survival differences were most pronounced in Connecticut for all cancers combined (−1.7), colorectal cancer (−3.5), and breast cancer (−2.0) when compared to other SEER regions. There were more significant differences between SLT and USLT relative survival estimates for colorectal cancers among women as compared to men, ranging from −2.4 in Los Angeles County to −3.5 in Connecticut.

For men diagnosed at younger ages (<75) differences between relative survival estimates using USLT as compared to SLT were small (Figures 1A–D). However, in the older age groups (aged 85+), there were substantial differences in survival. Among white males, SLT-based relative survival for all sites combined, colorectal, and prostate cancer were 10, 12, and 17 points lower than USLT-based relative survival, respectively. For black males, there were also marked differences but in the opposite direction: 10, 9, and 17 points higher for all sites combined, colorectal, and prostate cancer, respectively. Furthermore, and counter to known prostate cancer racial disparities between blacks and whites (25), the five-year relative survival generated using SLT for whites aged 85 and older was lower (63%) than blacks (81%).

Five-year relative survival estimates based on SLT were consistently lower than those based on USLT across all cancer sites for black and white women aged 85 and older (Figure 2A–D). Marked differences were found for colorectal (−11) and breast (−17) cancer among white women, whereas the differences for blacks were only −6 and −10 survival points, respectively. Unlike the pattern we saw for black and white men, white women had consistently higher survival rates than black women.