Cancer survival measures are an important tool for cancer patients, clinicians, researchers, and policy makers to understand the burden of cancer at both the individual and population levels.(1) Observed (crude) survival measures, which indicate probability of death from cancer or other causes, are informative for cancer patients and their physicians as an indicator of prognosis. Net survival measures, which indicate cancer survival in the absence of other (non-cancer) causes of death, are more commonly used by researchers and others to indicate progress towards cancer control through examination of trends over time, and comparisons between populations.(2) Net survival metrics can also be used to indicate cancer disparities, and evaluate progress towards elimination of such disparities.(3,4)

Net survival can be considered in two settings: relative, and cause-specific. Relative survival compares survival among a population of cancer patients to that which would be expected in a comparable cancer-free population. Because this framework is not dependent on cause of death information, it has been considered the superior approach, and has been widely used in the cancer surveillance community.(5–7) It may be particularly useful for international comparisons where specific cause of death might not be known for different countries or populations, and is also preferable when background mortality is suspected to differ between populations being compared.(8) However, one key limitation of relative survival is that calculation of this measure requires life-tables that well represent the experience of the population.(9,10) Age and sex-adjusted life-tables are not always accurate for certain population subgroups, biasing relative survival estimates.(11)

Conversely, cause-specific analyses consider only death due to the cancer as being an “event”; individuals dying from any other cause are censored at the date of death. A critical assumption of this method is accuracy of cause of death information;(1) misclassification in cause of death can occur because cancer patients may have complicated medical histories which include many comorbidities that could have contributed to their reason for dying. Furthermore, as time since cancer diagnosis increases, it can become difficult to distinguish and classify death due to that specific cancer versus other causes. Cause-specific survival is considered the superior methodologic approach when cause of death information is known to be accurate, and in patients diagnosed with localized disease, or who have been heavily exposed to specific risk factors including infectious agents and smoking.(12) Cause-specific analysis has also been used among populations for whom the general age- and sex-adjusted lifetables were thought not to reflect that population’s life expectancy experience.(13)

Choice of survival metric may affect our ability to understand health disparities affecting minority populations in the U.S. We are interested in cancer disparities that affect American Indian and Alaska Native (AIAN) people.(13–16) Previous work examining survival among AIAN people has used a cause-specific framework to indicate improvements in survival from lung and colorectal cancers among Alaska Native people,(13) as well as to examine racial and ethnic disparities in survival in New Mexico.(17) In 2018, the National Cancer Institute developed and released lifetables accounting for differences in mortality by geography (county-level), socioeconomic status (SES), and race/ethnicity, including among AIAN people.(11) These more specific lifetables have been shown to more accurately estimate relative survival among minority groups.(11) In the present study, we were interested in exploring how choice of survival metric influences assessment of cancer survival disparities specifically among AIAN people for all cancer sites combined, as well as specific cancer types; thus, we calculated survival estimates using different available approaches (observed, net: relative, cause-specific), and compared estimates between AIAN people and non-Hispanic whites (NHW). As a secondary aim, we compared survival estimates among AIAN people stratified by age, sex, cancer stage, and IHS region (all cancer sites combined, and leading cancers).

Case counts utilized for observed and relative survival analyses are given in Table 1. There were 75,485 AIAN cancer cases (all sites) included in this study. The most common tumor was female breast, followed by cancers of the lung and bronchus and colorectal cancers. Females comprised a little over half of the AIAN study population (52%). Over half (58%) of the AIAN study population was aged 20–64 years at the time of diagnosis, compared to 45% among NHW. Most AIAN patients (58%) were diagnosed with a cancer at regional, distant or unknown stage, compared to 51% among NHW. Reflective of the overall AIAN population by region, the largest proportion of cases among AIAN people were seen in the Southern Plains region, followed by the Southwest and the Pacific Coast.

Understanding and addressing cancer disparities among minority populations is of interest to the National Institutes of Health, professional associations including the American Cancer Society, clinicians, and the communities themselves. Although changes in cancer survival are most accurately evaluated in the presence of incidence and mortality,(27) cancer survival can illuminate differential effects of post-diagnosis factors such as access to treatment, treatment effectiveness, and pre-diagnosis factors such as access to screening. Different survival metrics (observed, net: cause-specific and relative) reflect different underlying methodologies, and as such provide different information suitable for different purposes. The purpose of this study was to evaluate how the use of different survival metrics might influence investigations into cancer survival disparities affecting AIAN people.

Our findings have direct implications for studies of disparities. We found that differences in survival estimates between AIAN people and NHW varied greatly depending method used, including referent lifetables. In our analyses, differences between methods were different between AIAN people and NHW (Figure 2); this was particularly pronounced for relative survival estimates calculated using different lifetables. Specifically, we observed differences between relative survival calculated with race-, SES-, and geography-specific and general population lifetables for AIAN people, but not NHW. This likely occurs because general population lifetables mainly reflect the mortality experience of the largest group (NHW), thus underestimating referent mortality of AIAN populations in the context of relative survival. We found relative survival ratios calculated using general population and race-, SES-, and geography-specific lifetables to be similar for those cancer sites with the lowest survival (liver, stomach, pancreas, lung and bronchus), but sometimes very different for sites with higher survival (female breast, prostate, colorectal, kidney).(11) One potential explanation for this is that female breast, prostate and colorectal cancers are all screenable cancers. Socioeconomic status is positively correlated with cancer screening, impacting stage at diagnosis and subsequent survival.(28) Race-, geography-, and SES-specific lifetables adjust for these confounding factors at the area-level, generating more accurate estimates of death associated with cancer than general lifetables, as reflected in research by Mariotto et al.(11) This is particularly useful when evaluating changes in survival over time. Alternatively, it may be that for higher survival sites where relative survival is closer to 100% (i.e., the relative survival ratio comparing cancer patient to general population survival is closer to 1), that the background mortality of the population (and the lifetable used to calculate that mortality) has a larger impact on the relative survival estimates. Regardless of mechanism driving these differences, we conclude that choice of lifetable may have drastic effects on estimates, and in turn, interpretation of disparities. Furthermore, we recognize that depending on the study question, the comparisons of survival metrics calculated via different methods and with different life tables may be illuminating.

While race-, geography-, and SES-specific lifetables are useful when evaluating survival disparities between groups, the presentation of survival estimates using general lifetables may also be useful in illuminating disparities between groups with different population-level health behaviors or experiences. Depending on the research question – and in particular in the investigation of health disparities – investigators may find it equally or more revealing to present relative survival estimates based on general and race-, geography-, and SES-specific lifetables, as a means by which to help explain the “crude” but real disparities in survival observed from general lifetables. Further, the relative survival estimates presented herein do not, for example, account for distributions of stage at diagnosis, smoking status, or primary site, all of which impact survival and may be differentially distributed between populations. We used a widely-available and commonly used tool (SEER*Stat) to conduct these analyses; however, we recognize that should one desire to account for these factors in elucidating survival disparities, and/or answer more complex questions that require simultaneous adjustment for multiple factors, a modeling approach may be more appropriate. In particular, flexible parametric models have wide applicability, including in situations where the assumption of proportional hazards may be violated.(29,30)

In our analysis, observed (crude) survival produced the lowest survival estimates, whereas cause-specific methods produced the highest estimates (Figure 1). This latter finding is consistent with those previously reported in the literature.(1,9,10,31,32) and is expected because observed survival uses the broadest definition of an “event” (i.e., death from any cause), whereas cause-specific methods have the narrowest definition (i.e., death from that specific cancer).(1) Furthermore, estimates from the different survival methods were more similar among cancer sites with low survival (lung, pancreas, stomach), relative to those with higher survival (female breast, colorectal, kidney). In situations of low competing risks (i.e., highly fatal cancers), cause-specific and observed survival are expected to be more similar than in situations of high competing risk (i.e., less fatal cancers).

To our knowledge, only one paper by Withrow and colleagues has analyzed how choice of survival method impacts estimates among Indigenous populations;(10) this study showed that, among Canadian First Nations and Inuit populations, cause-specific survival when calculated using a narrow definition of the event produced survival estimates that were much higher than those calculated using cause-specific methods with a broader definition of the event, or relative survival. This finding was particularly pronounced for cancers of continuous organ systems (e.g., digestive and respiratory systems). Our present work adds to this body of literature by examining methods and data pertinent to the U.S. Indigenous context. We anticipate that this information will be of interest to individuals working with U.S. Indigenous communities to reduce their burden of cancer.

Secondary objectives of this study were to describe survival for leading cancer sites among AIAN people by age, sex, IHS region, and stage at diagnosis, and differences in survival estimates between AIAN people and NHW. We observed slight differences in survival between most IHS regions, but Alaska had markedly lower all-site survival and the East region having markedly higher all-site survival. This observation may be explained in part by underlying differences in the primary cancer site distribution by region; for example, incidence of highly fatal lung cancer is lower in the Southwest region than it is in Alaska.(16) Site-specific cancer survival also varied by region: for example, survival from colorectal cancers was 10% lower in the Southern Plains region compared to Alaska, potentially identifying an area where increased screening efforts are warranted. Differences in site-specific survival between regions are attributable to stage at diagnosis, and access to and effectiveness of post-diagnostic treatment services. To our knowledge, there are few recent studies that examine cancer survival by region among AIAN people; however, several recent studies from authors at the Centers for Disease Control and Prevention have shown variations in incidence of and disparity from several leading cancers by IHS region.(15,20,33)

For almost all cancer sites examined, five-year survival was lower among AIAN people than NHW. The magnitude of disparity varied by cancer site, but was highest for colorectal cancer, where net survival measures calculated approximately 10% lower survival among AIAN people. Again, differences were largest for cancer sites that are screenable (e.g., colorectal cancer, female breast cancer), and with high survival (e.g., prostate cancer), and lowest for cancers with high mortality (e.g., lung cancer, pancreatic cancer).

This manuscript has several strengths and limitations. A key strength of this work is the use of the NAACCR CiNA dataset, which includes data from both National Program of Cancer Registries (NPCR) and SEER registries from across the U.S. Therefore, our analyses benefitted from increased sample size and statistical power relative to many survival analyses examining data from AIAN people, which have either used data from a single registry(13) or the SEER system alone.(34,35) A key methodological consideration when interpreting the results of this study are the different methods and resources registries are able to use to ascertain patient follow-up. U.S. registries are supported by, and participate in, the National Cancer Institute’s SEER Program, and/or the Center for Disease Control and Prevention’s (CDC) NPCR. U.S. registries utilize local and national resources to ascertain patient vital status, e.g., state death tape linkage and National Death Index linkage. Registries adhering to SEER Program standards additionally conduct active follow-up on cases to ascertain a date of last follow-up for ≥ 90% of alive patients for the study time period. Registries funded by the SEER Program are able to leverage additional resources, e.g., the Social Security Administration Service to Epidemiological Researchers, to ascertain date of last follow-up among alive patients with efficiency, greatly improving their ability to get more accurate estimates of follow-up time.(36) Registries that are not funded by the SEER Program and do not have resources to ascertain date of last follow-up on all alive patients rely on complete death clearance to estimate survival; this presumed alive model can bias survival estimates upwards in certain populations relative to the observed survival time used for registries with active follow-up for alive patients.(19,37) Because data included in the CiNA Research File for Survival meet stringent data quality standards and because follow-up practices are internally consistent within registry for the time period under study, we anticipate that using “blended” survival follow-up for this study yielded appropriate and internally valid results.

Another limitation of these analyses is that the analytic sample was different between observed/net: relative, and net: cause-specific analyses, due to the requirement for known cause of death in cause-specific analyses; this is not expected to greatly impact results. We describe differences in results produced by four different survival methods; however, the magnitude of these differences will be affected by underlying differences in the populations being compared. For example, if access to and utilization of screening services is similar in the two populations, we might expect to see lower differences between the measures. However, if screening prevalence differs between populations, this may result in observing a larger disparity. Finally, an important limitation of a large majority of studies using population-based cancer registry data to understand the AIAN experience: our analyses were limited to those living in PRCDA counties. This choice is often made because of the potentially higher racial misclassification in counties outside of those areas served by IHS facilities(18,38,39). Yet, it neglects the experience of AIAN people in non-PRCDA counties, which are often urban, and in the East of the U.S. Thus, it should be noted that these results may not reflect the experience of 47.7% AIAN people living outside of PRCDA counties.(40) Future work should continue to explore the cancer burden experience of AIAN people living outside of PRCDA counties.

This manuscript provides two key pieces of information that will be of use to scholars who are interested in cancer disparities among AIAN people. First, we provide information to compare different observed and net measures of survival and conclude that consistency in metric is imperative; studies conducted using different methods should be compared with caution. Additionally, these results – as seen in the differences in relative survival by lifetable – underscore the importance of careful consideration of research objectives in the selection of analytic method. Second, we provide data on survival for leading cancers among AIAN people by sex, age, stage at diagnosis, and IHS region. In general, we observed lower survival among AIAN people than NHW. The magnitude of disparity varied by cancer site, but these data suggest that continued efforts should be made to explain these disparities, and to address them using Tribally-driven approaches.