Quantifying Differences in Remaining Life Expectancy after Cancer Diagnosis, Aboriginal and Torres Strait Islanders, and Other Australians, 2005–2016

This article is featured in Highlights of This Issue, p. 1137

Aboriginal and Torres Strait Islanders, the First Nations peoples of Australia continue to experience poorer overall cancer survival than other Australians (1–6). These differences likely reflect logistic, social, environmental, and health system factors occurring across the cancer continuum (3, 4, 6, 7).

Commonly used population-based relative survival estimates can however be misunderstood and, by ignoring noncancer mortality, are less informative for the communication of cancer prognosis (8). Estimating the average number of years a person can expect to live following cancer the “remaining life expectancy” after a cancer diagnosis, provides a potentially more easily interpretable survival-based measure with real-world meaning, by quantifying a cancer's impact over a lifetime horizon, rather than for example, the first 5 years after diagnosis (8, 9). Previous studies have used this measure to quantify the impact of a cancer diagnosis (9–16), including calculating the proportion of the socioeconomic disparity in RLE due to the cancer itself (15), and to quantify differences across population groups (such as socioeconomic disadvantage) by estimating how many life years could be gained if cancer survival disparities between the groups were removed (10, 13–16).

Although Australia's overall life expectancy is high compared with other countries (17), the average life expectancy of Aboriginal and Torres Strait Islander Australians is around 8 years shorter than other Australians (18). To the best of our knowledge, no study to date has used RLE to measure how the impact of a cancer diagnosis among Aboriginal and Torres Strait Islanders differs to that of other Australians. By using a large, population-based cohort extracted from cancer registries with accurate Aboriginal and Torres Strait Islander identification, we quantify the disparities in RLE for Aboriginal and Torres Strait Islanders diagnosed with cancer and estimate what proportion of these disparities are due to cancer-specific mortality and other cause (noncancer) mortality.

This study was conducted in accordance with the Declaration of Helsinki. Ethics approval was obtained from the Aboriginal Health and Medical Research Council Ethics Committee (1256/17), New South Wales (NSW) Population and Health Services Research Ethics Committee (2017/HRE0204), and the Northern Territory (NT) Department of Health and Menzies School of Health Human Research Ethics Committee (2016–2689). The need for informed consent from participants was waived as the study involved secondary analysis of routinely collected de-identified data.

Analysis was restricted to the cancer registries of Queensland, Western Australia, NT, and NSW as they are considered to have high quality Aboriginal and Torres Strait Islander identification (19). Latest (2016) census-based estimates indicated that around 84% of Australia's Aboriginal and Torres Strait Islander population lived in these four jurisdictions combined (20). All four registries, to which notification of all invasive cancers (excluding keratinocyte cancers) is a statutory requirement, approved release of de-identified data from the Australian Cancer Database (21).

The initial cohort comprised 843,905 individuals diagnosed with cancer during 2005 to 2016 (up to 2015 for NT). Vital status was determined through routine annual linkage of Australian National Death Index, with follow-up to 31 December 2016. We restricted our cohort to those aged 15 to 89 years at diagnosis (n = 767,718) due to different classification of childhood cancers(22) and the maximum age limit for Aboriginal and Torres Strait Islander life tables (23). The exclusion of cases identified at death (4,451, 0.5%), with negative survival times (1,771, 0.2%) or with missing Aboriginal and Torres Strait Islander status (n = 52,257, 6.2%) gave the final cohort of 709,239 persons. For each cancer type, only the first primary diagnosis during the study period was considered.

Analyses are presented for 16 leading individual cancer types (see Supplementary Table S1 for International Statistical Classification of Diseases and Related Health Problems, 10th revision ICD-10) codes) which collectively comprised 87% to 91% of all cancers diagnosed within study period, all other cancers and all cancers combined.

We used published year-specific life tables for the Aboriginal and Torres Strait Islander population, stratified by age, sex, and state (23) for all calendar years except 2008, 2009, 2013, and 2014. Linear interpolation was used to estimate the life tables for these missing years. Corresponding life tables for all Australians (24) were used for other Australians who comprise around 97% of the total population (20).

Survival was measured in days from the date of diagnosis to either date of death or the study endpoint (31 December 2016), whichever came first. Cases alive at the study endpoint were censored.

Survival analyses were carried out in the framework of relative survival, which is the ratio of the observed all-cause survival for the cancer cohort to the expected all-cause survival in a cancer-free population (8).

The key outcome measure was the average RLE (years) per individual following a cancer diagnosis, stratified by Aboriginal and Torres Strait Islander status. Other derived measures included the (absolute) disparity in RLE between Aboriginal and Torres Strait Islanders and other Australians and the impact of differences in cancer survival or expected survival on this disparity. We also calculated the total remaining life years for the cohort of Aboriginal and Torres Strait Islanders diagnosed with cancer in 2015 (because 2016 NT data was unavailable), and the potential gain in life years if disparities in cancer mortality were removed.

Theoretically, the RLE following cancer is represented by the area under the observed (all-cause) survival curve for the cohort (11). To estimate the long-term observed survival, we used the approach of Andersson and colleagues (25) that involves extrapolation of both the relative and expected (or population) survival curves. Extrapolation of relative survival was carried out using flexible parametric relative survival models (26), with the model parameters chosen to be consistent with previous studies (11, 25). Similar to previous studies (11, 15), future expected mortality rates were assumed to be constant, and equal to that in 2016 (the last year of available data in the population life tables).

Life expectancy measures were standardized to the covariate (age, sex, state/territory) distribution of Aboriginal and Torres Strait Islanders, to allow comparisons that solely focused on the survival differences by Aboriginal and Torres Strait Islander status as both groups had the same distribution of other covariates.

For each cancer type, flexible parametric relative survival models were fitted, using age at diagnosis, sex, state/territory, and Aboriginal and Torres Strait Islander status as covariates. Models for all cancers combined, all other cancers and head and neck cancers also included an additional covariate for broad cancer type and interactions between age or sex and broad cancer type to account for difference in case-mix over diagnosis year and population sub-groups (26). Age was modelled using restricted cubic splines. Interactions between Aboriginal and Torres Strait Islander status and other covariates (age, sex, state/territory) were also tested, but not included in final model based on likelihood ratio tests. Further details of the methods are in Supplementary Methods and Materials (Section A).

Similar to a previous study (12), estimates of RLE for all cancers combined were generated by calculating a weighted average of the cancer-specific estimates by Aboriginal and Torres Strait Islander status. The weights reflected the corresponding distribution of these cancers in the cohort. No confidence intervals (CI) are presented for these weighted averages.

Expected mortality was incorporated into these models using published life tables stratified by Aboriginal and Torres Strait Islander status, age, sex, year and state/territory from 2005 to 2016 (23, 24). Five-year cancer-specific relative survival estimates stratified by Aboriginal and Torres Strait Islander status were also calculated.

Results are presented as the averaged measure over all individuals in the cohort over all ages.

The observed disparity in the RLE was the difference between the standardized RLE for other Australian patients with cancer (RLE_OTHER, standardized to the covariate distribution of Aboriginal and Torres Strait Islanders) and the standardized RLE for Aboriginal and Torres Strait Islanders diagnosed with cancer (RLE_FN):

Disparities in the RLE experienced by Aboriginal and Torres Strait Islander patients with cancer compared with other Australian patients with cancer reflects differences in both the impact of the cancer itself as measured by relative survival and noncancer mortality (expected survival).

The separate impact of cancer and noncancer mortality on these disparities was quantified by adapting the method of Rutherford and colleagues (15). Briefly, we calculated the standardized “cancer-adjusted remaining life expectancy” (RLE_{FN_CANCER}) by assuming that Aboriginal and Torres Strait Islanders had the same cancer mortality (relative survival) as other Australians but retained their own underlying population mortality (expected survival). This was used to estimate the disparity observed for Aboriginal and Torres Strait Islanders compared with other Australian patients with cancer, that was due to differences in noncancer mortality by:

The disparity that was due to differences in cancer mortality, which constituted the rest of the observed disparity comparing Aboriginal and Torres Strait Islander peoples with other Australian patients with cancer, could then be obtained by:

Both measures were also expressed as a percentage of the RLE for Aboriginal and Torres Strait Islanders.

The total number of remaining life years for the cohort of Aboriginal and Torres Strait Islander peoples diagnosed with cancer in 2015 was estimated by multiplying the cancer-specific number of cases by the corresponding RLE measure. Please see Supplementary Methods and Materials (Section B) for further details.

All analyses were performed with Stata/SE version 16 (RRID:SCR_012763). Models were fitted with the stpm2 package (26). Five-year relative survival, RLE and the differences in these estimates by Aboriginal and Torres Strait Islander status were estimated using standsurv (Supplementary Material, Section A) (27). Associated 95% CI were estimated using the delta method. Example Stata syntax are available in Supplementary Methods and Materials, Section C.

Sensitivity analyses were conducted to investigate the effect of choosing different number of knots in the statistical models and different times (years) for when the extrapolated survival curves were assumed to have become zero. The impact of unknown Aboriginal and Torres Strait Islander status on reported measures was explored by repeating the analyses by assuming that all the unknown cases were: (i) other Australians; (ii) Aboriginal and Torres Strait Islander, or (iii) randomly distributed equally over both these categories.

The data that support the findings of this study are available from Australian Institute of Health and Welfare. Restrictions apply to the availability of these data, which were used under license for this study. Data analyzed for this paper are not able to be shared on any publicly available repository due to legal and confidentiality requirements.

The final study cohort consisted of 709,239 persons of which 12,830 (1.8%) identified as Aboriginal and Torres Strait Islanders (Supplementary Fig. S1). For all cancers combined, all other cancers and each of the 16 individual cancer types except cervical cancer, Aboriginal and Torres Strait Islanders were diagnosed at a significantly younger age (P ≤ 0.001) (Supplementary Table S1).

Aboriginal and Torres Strait Islanders had significantly poorer 5-year relative survival than other Australians for all cancer types (Table 1; Supplementary Fig. S2). However, the magnitude of the difference varied substantially with the greatest difference being for cervical or head and neck cancers and lowest for pancreatic cancer.

For melanoma, female breast, cervical, uterine, and prostate cancer, along with lymphomas, Aboriginal and Torres Strait Islanders had on average at least 15 years of life remaining whereas those diagnosed with esophageal, lung, liver, or pancreatic cancer had, on average, less than 3 years of remaining life (Table 1).

These estimates were consistently lower for Aboriginal and Torres Strait Islanders than other Australians (Table 1; Fig. 1A), when standardizing the covariate distribution (age, sex, state/territory) of other Australians to the same covariate distribution (age, sex, state/territory) as Aboriginal and Torres Strait Islanders. On average, Aboriginal and Torres Strait Islander patients with cancer had 12.0 years of life remaining. The corresponding value for other Australians was 20.0 years, meaning that the disparity was around 8.0 years.

The magnitude of the disparity in RLE ranged from less than two years for lung and pancreatic cancers to more than 10 years for cervical cancers (Supplementary Table S2).

If Aboriginal and Torres Islanders experienced the same cancer mortality as other Australians, their standardized RLE would increase by an average of 2.1 years for all cancers combined (Supplementary Table S2). Values varied by cancer type, ranging from 0.2 years for prostate to 6.6 years for cervical cancer (Fig. 1B). In percentage terms, cancer mortality accounted for 26.2% of the disparity in RLE for all cancers combined, with this percentage ranging from 4.0% (prostate) to 75.8% (liver; Fig. 1C). Noncancer mortality accounted for 73.8% of the disparity in RLE, again with large variations in estimated values by cancer type (Supplementary Table S2, Fig. 1C).

For all cancers, we estimated that the cohort of Aboriginal and Torre Strait Islanders diagnosed with cancer in 2015 (n = 1,342) would gain a total 2,818 life years (2.1 years per cancer) if they experienced the same cancer survival as other Australians (Table 2). This represents an increase of about 17% to what they currently experience.

Differences in these estimates by cancer type reflected variation in both the number of cases diagnosed and the impact that cancer mortality had on the disparities in RLE (Table 2; Supplementary Fig. S3). This meant that the absolute gain in life years was highest for female breast cancer (447 years) and head and neck cancers (473 years), even though the percentage contributions of cancer mortality for breast cancer was relatively small.

Estimates were relatively robust to the number (and position of knots) used for the modelling, with for example the RLE for cervical cancers ranging from 18.7 to 19.2 years for Aboriginal and Torres Strait Islanders depending on the specific selection. Estimates were also not impacted by changing the maximum time period after which the extrapolated survival curves were assumed to be zero from 80 to 100 years with Aboriginal and Torres Strait Islander patients with cancer having around 18.7 remaining life years for both instances.

In addition, sensitivity analysis for unknown Aboriginal and Torres Strait Islander status (Supplementary Tables S3 and S4) showed that although the magnitude of the estimates varied, the overall patterns for the disparities were similar regardless of the true distribution of Aboriginal and Torres Strait Islanders. For example, the RLE for liver cancers ranged from 2.0 to 2.9 years for Aboriginal and Torres Strait Islanders, while the corresponding disparity varied from 2.2 to 3.4 years. The percent contribution that cancer and noncancer mortality made to the disparity were also relatively consistent (Supplementary Table S4).

Using a population-based semi national cohort, this is the first study to highlight how Aboriginal and Torres Strait Islanders diagnosed with cancer consistently faced a lower RLE than other Australian patients with cancer, regardless of the cancer type and their age at diagnosis. On average, about one third of the disparity was caused by the greater mortality due to the cancer itself, with the remaining disparity being due to noncancer deaths.

There was no evidence of marked differences in the patterns for RLE disparities by Aboriginal and Torres Strait Islander status when analyses were stratified by the standardized covariates of age, sex and state/territory, although not all models stratified by state/territory converged due to smaller numbers. Additional models that included relevant interactions also found no evidence that the reported disparities varied by age, sex, or state/territory.

Previous studies have shown that cancer mortality is the biggest contributor to poorer survival among patients with cancer (28, 29). However, these studies have focused on either 5- or 10-year survival outcomes, rather than entire life span of a patient used by life expectancy measures. Moreover, when considering cancer-specific survival, it is typically the first few years in which the disparity faced by Aboriginal and Torres Strait Islanders compared with other Australians is the greatest (3, 30). As follow-up time after a cancer diagnosis increases, and people survive the initial treatment phase, the impact of noncancer causes generally increases (25). This is especially so because for some cancer types there is what is known as “population cure” (31), which is when the mortality among a specific cancer cohort returns to the same level as within the general population. Recent estimates (31) suggest that Australians diagnosed with most cancer types experienced some level of population cure, with the percent cured ranging from 5% for pancreatic cancer to 90% for melanoma.

Multiple factors including biological, genetic, environmental, behavioral, and social determinants of health interact to impact an individual cancer risk and their survival outcomes after a cancer diagnosis (32, 33). Various models have also been developed to explain the complex and multifaceted relationship between key drivers of disparities in cancer outcomes between Indigenous (including Aboriginal and Torres Strait Islander peoples) and nonIndigenous peoples (34, 35). Factors such as structural inequities and institutionalized discrimination translate into differences in determinants of health, exposures, and opportunities (at multiple levels); access to care and quality of care all of which contribute to Indigenous cancer inequities.

Previous Australian studies have shown that the risk of noncancer death was higher among Aboriginal and Torres Strait Islanders than other Australians diagnosed with cancer (36, 37). Suggested contributors to this include social determinants of health (including education, income, employment; ref. 7, 38), along with geographical, behavioral, environmental, and health-system factors (38–40). Another potential contributor is their higher comorbidity burden, although Aboriginal and Torres Strait Islander patients with cancer in Queensland still had a higher risk of other-cause mortality after accounting for comorbidities (36). It is clear that addressing the higher mortality burden among Aboriginal and Torres Strait Islander patients with cancer compared with other Australian patients with cancer needs to consider both cancer-related and noncancer mortality related risk factors.

Aboriginal and Torres Strait Islander patients with cancer face additional logistic, systemic, cultural, and socioeconomic barriers across the cancer care pathway from prevention and diagnosis to optimal treatment and survivorship (4, 6, 41–44). In addition to reducing cancer deaths through improved diagnostic, management and support initiatives, initiatives to reduce the high noncancer mortality burden among this population also need to be considered. This highlights the need to have a multifaceted approach across multiple sectors in partnership with communities to improving cancer survival among Aboriginal and Torres Strait Islanders that considers not just the management of the cancer diagnosis itself but encompasses a range of factors including health literacy, poverty, unemployment, and other social determinants of health. Understanding the reasons behind the avoidable cancer and other-cause deaths among Aboriginal and Torres Strait Islander Australians is crucial to making these interventions effective.

The impact of differences in relative survival or expected survival in explaining the Aboriginal and Torres Strait Islander disparity in RLE also varied by cancer type. For lung and pancreatic cancer most of the difference was due to cancer itself perhaps not unexpected given their overall lethality (1) and low proportion cured (≤10%; ref. 31). By contrast, for prostate cancer most of the difference was due to noncancer mortality, which is consistent with 76% cured proportion (31).

Study strengths include use of high-quality population-based cancer registry and mortality data (21) and a flexible parametric model-based approach (25) combined with recently developed methodology (27) to estimate survival curves that were standardized to a common distribution of confounders (for Aboriginal and Torres Strait Islanders) thus allowing direct comparison of reported measures by Aboriginal and Torres Strait Islander status.

Life expectancy measures are however based on extrapolated information; hence, our reported results may not completely reflect actual observed survival in the future. Nevertheless, flexible parametric models have been shown to allow reliable extrapolation (25, 45). Because the reported life expectancy measures are the overall average across an individual's life span, they cannot provide insight on when cancer or noncancer mortality may have most impact, however, as discussed earlier, cancer mortality tends to have greater impact in the years soon after diagnosis. Presented results are averages across the whole cohort and so may not be applicable to individual patients with cancer depending on their specific characteristics. Expected mortality was based on the best available population life tables for Aboriginal and Torres Strait Islander Australians (23); however, compiling reliable life tables for this population is challenging (1, 23). The lack of detailed clinical information notably stage at diagnosis (12, 46) and treatment also limited our ability to explore reasons for observed patterns. In the absence of information on other prognostic factors such as comorbidities (47), lifestyle factors (48) and socioeconomic status (15, 46), the possibility that differences in these variables across population sub-groups explain at least partially some of the observed disparities by Aboriginal and Torres Strait Islander status cannot be further explored.

Reported estimates may also have been impacted by socioeconomic status and behavioral differences in the ‘screened populations’, such as lower screening uptake among Aboriginal and Torres Strait Islander peoples (1). Given that expected population life tables stratified by demographic variables other than Aboriginal and Torres Strait Islander status, state/territory, sex, age, and calendar year were unavailable, we were unable to account for any differences in expected mortality by factors such as lifestyle and socioeconomic status.

Reported measures were estimated within the relative survival framework, which assumes that cancer deaths are a negligible proportion of all deaths in the general population (8). If there is a sufficiently high proportion of deaths from a specific cancer in the population to inflate the expected number of deaths, excess mortality in the cancer cohort will most likely be underestimated thereby inflating the relative survival. However, in practice, this has not been found to be a major problem for most individual cancer types except possibly for prostate cancer among older people (49).

For all cancers combined, the assumptions behind using the expected mortality for other cause mortality may not be valid, because for all cancers combined, the mortality due to cancer will not be a negligible part of the background mortality. Obtaining a good model for predictive purposes can also be problematic and potentially require a large number of interactions to adequately capture the true survival experience. Reported RLE measures for all cancers combined were thus calculated as weighted average of cancer-specific estimates over 16 individual cancer types with all remaining cancer types collectively categorized as other cancers. This methodology has been previously used to report life expectancy measures for all cancers in Australia (12). The individual cancer types together comprised 87% of all cancers diagnosed among Aboriginal and Torres Strait Islanders in study cohort (91% other Australians).

Because a fundamental assumption of relative survival methods is the independence of cancer and other causes of death and the comparability of cancer cohort and general population (50, 51). They may be less valid for certain cancer types due to shared risk factors like smoking and comorbidities (50–52). An inherent limitation of Australian cancer registry data is the lack of information on covariates such as comorbidities and lifestyle factors which, combined with the lack of life tables stratified by these variables, means we cannot address these issues in current analysis.

The quality of Aboriginal and Torres Strait Islander status identification which is largely based on hospital notifications (1,5,53) varies across cancer registries. This study was limited to four jurisdictions with high quality Aboriginal and Torres Strait Islander identification (1,19) and although these represent over 80% of the Aboriginal and Torres Strait Islander people in Australia, the findings may not be nationally representative. While cases (6% of cohort) with unknown Aboriginal and Torres Strait Islander status were excluded from the main analyses, additional sensitivity analyses showed that the magnitude and general patterns of the disparities in reported measures remained consistent regardless of various assumptions made about the true distribution.

A cancer diagnosis among Aboriginal and Torres Strait Islanders exacerbates the existing disparities in RLE they face compared with other Australians. Effectively addressing the higher mortality burden among this population group will need to consider the management of cancer-related factors and reducing the impact of factors contributing to the disparities in noncancer causes of death.

G. Garvey reports grants from National Health and Medical Research Council outside the submitted work. No disclosures were reported by the other authors.

P. Dasgupta: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. T.M.-L. Andersson: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. G. Garvey: Writing–review and editing. P.D. Baade: Conceptualization, supervision, validation, methodology, project administration, writing–review and editing.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.