The Impact of State-Specific Life Tables on Relative Survival

There is little debate that measures of survival are a valuable tool available to clinicians, epidemiologists, and public health professionals (1). First proposed by Ederer et al. (2) relative survival compares the survival probabilities of a diseased population (ie, cancer patients) to the survival probabilities of the general population. The resulting value, known as relative survival, is the ratio of observed survival to expected survival which represents the excess mortality associated with a cancer diagnosis. Relative survival is often used as the primary measure in population-based descriptive studies as in general it provides a more reliable estimate of the net survival from cancer than cause-specific survival because it does not rely on death certificate information, which is generally prone to errors in the coding of cause of death and has been shown to have high degrees of misclassification for cancer causes of death (3–5). Unlike crude probabilities of death, net survival represents cancer survival in the absence of competing risks (6). In this issue, Howlader et al.(7) and Mariotto et al. (8) further discuss differences between net and crude survival measures. A key challenge of relative survival is in choosing the population with which to compare to the cancer cohort or in choosing the best life tables to represent the background mortality of the study population. There is a growing body of research questioning the accuracy of relative survival estimates that use life tables from populations that are either not comparable to the cancer cohort [eg, background mortality risk is substantially higher or lower than the cancer cohort, misclassification or unmeasured factors such as race/ethnicity that contribute to differential death rates in the comparison population (9–11)] or when the reference population has a high rate of deaths due to cancer (12). In the United States, researchers using cancer surveillance data from the Surveillance Epidemiology and End Results (SEER) Program often use the US population life tables as the reference group for relative survival estimates, matched by age, sex, race, and calendar year to the cancer cohort (13). The use of US population life tables in and of itself is not necessarily a problem if the cancer cohort has comparable background death rates and characteristics. However, we know that background mortality varies by sex, age, race, and geographic areas as well as socioeconomic status (SES) (14–16). The use of national life tables to calculate state-specific or regional survival has overestimated relative survival in states with lower background death rates and underestimated relative survival in states with higher background death rates (9). Baili et al. (9) reported that even after matching on age, sex, year, and race, relative survival for several US states and regional registry populations were systematically higher when using US-matched life tables compared to the relative survival estimates using SLT [aka: the CONCORD approach (9)]. The only exception was found in Louisiana, which had higher US-matched relative survival estimates as compared to its state-matched relative survival estimates. The authors attributed these differences to the lower background death rates in all states/regions in their study with the exception of Louisiana, which had higher background death rates. Moreover, for early stage prostate and breast cancer, relative survival has been shown to be higher than 100%, indicating that the life tables may not be appropriate for representing survival from other causes when examining cancer survival for these populations (9). With the recent release of the 2000 state decennial life tables by the National Center for Health Statistics (NCHS) (17), this study expands on the work by Baili et al. (9) by comparing five-year relative survival using state- or regionally-matched life tables to US-matched five-year relative survival, focusing on a more contemporary cohort of cancer patients diagnosed from 2000 to 2009. We also examine the underlying characteristics of each state/region to identify potential root causes of variations in survival estimates and assess the variations in survival estimates by age, race, and cancer site for lung and bronchus cancer, colorectal cancer, prostate cancer, and female breast cancer.