Cancer data in Australia

The number of deaths from cancer is based on the number of deaths for which cancer is determined to be the underlying cause of death. The underlying cause of death is defined by the World Health Organization as ‘the disease or injury which initiated the train of events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury’.

Cause of death coding in the National Mortality Database

The National Mortality Database (NMD) obtains cause of death information from death registrations; these must be certified by a doctor using the Medical Certificate of Cause of Death or by a coroner. The doctor is asked to provide their best medical opinion on the cause of death and outlines all diseases and conditions that caused and contributed to death.

The cause of death information on the death registration is sent to the Australian Bureau of Statistics (ABS) for coding and analysis. No further information or additional detail is available for doctor-certified deaths. For any coroner-referred cancer-related deaths, the ABS accesses information from the National Coronial Information System which includes the police report, toxicology, forensic pathology and coronial finding. The ABS codes these certificates according to WHO guidelines applying the International Classification of Diseases 10^th Revision (ICD-10). There are numerous complex rules related to cancer that must be applied. Data is disseminated for statistical purposes and used to create unit record datasets, such as is used in the NMD.

Cause of death coding in the Australian Cancer Database

Cause of death information is also compiled by the various state and territory cancer registries for people diagnosed with cancer. Medical coders refer to rules and guidelines described in the ICD-10 to assist in determining the cause of death. Cancer registries determine cause of death by considering what is stated on the death certificate as well as any additional information the registry has available to them, potentially including sources such as hospital admissions and pathology information. This additional information can result in a different cause of death being assigned than can be ascertained from the death certificate information alone (which is the only information available to code cause of death information in the NMD). State and territory cancer registries provide this coded cause of death information to the AIHW to include in the Australian Cancer Database (ACD). It should be noted that state and territory registry processes and information available can vary across jurisdictions to some degree.

Previously, the National Mortality Database (NMD) has been used by the AIHW as the sole basis for reporting deaths from cancer. However, as AIHW has started analysing and reporting mortality for more detailed cancer types, it has become apparent that the NMD may not be as suitable for reporting on certain cancer types. The AIHW has developed a new cancer mortality series using the ACD, which makes it possible to assess the suitability of  NMD data for reporting on different cancers.

At the time of releasing the 2022 update of Cancer data in Australia, these cancer mortality investigations remain a work in progress. More information about cancer mortality data is expected to be released prior to the 2023 release of CdiA. The cancer mortality figures published in the initial release of the CdiA 2022 report are unlikely to be revised as a result of the investigations, with planned work more likely to provide greater depth of understanding of cancer mortality data.

With often finer level cancer information available, the cancer registry-derived cause of death is more likely to be able to be coded to finer level causes of death for cancer. However, as the ACD cause of death is often reliant on probabilistic data linkage, some deaths may not be recorded within the ACD because a link could not be successfully made.

Determining cause of death can be complex and may depend on the information available. While the ACD is likely to be more precise than the NMD (for the reasons mentioned previously), some issues may still exist. For example, there are instances where a person may be diagnosed with a cancer in different states and territories at different points in time. In this case both cancer registries will register the cancer incidence that occurred within their jurisdiction and when deriving a cause of death may have access to different information and therefore could potentially arrive at a different cause of death. For these records, when AIHW receives the data, it uses an algorithm to derive a consistent cause of death for the person in the ACD.

It is expected that the combination of ACD and NMD mortality data will lead to more informed cancer mortality reporting. This paper looks to provide some understanding of how and why the data sources can differ. The 2022 release of the Cancer mortality by age data visualisations illustrates how the two sources compare for many different types of cancer.

When compared with the ACD, the NMD routinely reports more deaths from all cancers combined than the ACD (Table 1). Please note that these comparisons exclude non-melanoma skin cancer (NMSC) deaths because, for NMSC, the ACD only collects mortality information on the rare types of these cancers while the NMD also includes deaths from common NMSCs - basal and squamous cell carcinomas. This difference in scope influences comparisons between ACD and NMD deaths from cancer so deaths from this cancer are accordingly excluded in Table 1.

A death from cancer in the NMD but not the ACD will either be because a death is recorded in the NMD but not in the ACD or because the death is recorded in both databases but the ACD records the death as a non-cancer death. It is also possible for a death to be considered as a cancer-related death within the ACD but not within the NMD. It is unlikely for a death to be recorded in the ACD but not in the NMD because the NMD includes all deaths in Australia whereas the ACD only includes a subset of all deaths (for people who have been diagnosed with cancer since 1982).

The AIHW aims to get a better understanding of the differences between the databases at the unit record level during the remainder of this year. Until this work has been completed, it is not possible to provide certainty regarding the nature of differences. It is possible that the larger difference observed between the ACD and NMD in 2017 may be due to the ACD not yet receiving the notification of deaths for some records.

At this preliminary stage of investigation, there appear to be relatively small differences between the count of deaths from all cancers combined according to the NMD and the ACD, suggesting that NMD data is likely to be suitable to use for reporting on deaths from all cancers combined.

Note: the difference is calculated as the number of deaths from the NMD minus the number of deaths from the ACD and this difference is also presented as a proportion of the ACD.

Source: National Mortality Database and Australian Cancer Database 2018 

Tables 2 and 3 highlight that the comparability of the ACD and NMD deaths from cancer reduces as age increases. This is likely because the underlying cause of death by age may be more complex at older ages and the number of possible underlying causes of death for an individual are more likely to be greater. Table 2 shows that there is a relatively high level of agreement between the ACD and NMD for the age groups from 0 to 60 years old. However, there are greater differences for people aged 80 years and over and a large number of deaths occur in this age group.

Note: the difference is calculated as the number of deaths from the NMD minus the number of deaths from the ACD.

Source: National Mortality Database and Australian Cancer Database 2018 

The comparison of cancer-related deaths in 2016 examines a year where the overall difference between the ACD and NMD is greater than in 2013. Similar to 2013 data, the 2016 mortality data confirms that the overall difference between the ACD and NMD is likely to be most strongly influenced by differences in the oldest age groups.

Note: the difference is calculated as the number of deaths from the NMD minus the number of deaths from the ACD and this difference is also presented as a proportion of the ACD.

Source: National Mortality Database and Australian Cancer Database 2018 

In general, cancer mortality reporting from the NMD is likely to be more similar to the ACD for broader cancer sites and for more common cancer types. As cancer site information reaches a greater level of detail, it becomes more likely that the additional information available in the ACD enables the underlying cause of death to be recorded as more specific cancer sites.

Urinary tract cancer mortality provides a useful example to demonstrate this. In 2016, the NMD recorded 2,335 deaths from cancers of the urinary tract while the ACD recorded 2,240. Overall, in 2016 the different sources recorded a 4% difference in the number of deaths from cancer in this area of the body (Table 4).

When finer levels of reporting are considered, the differences between the mortality data according to the ACD and NMD increase. When 2016 deaths from the NMD are compared with the ACD, the number of deaths from bladder cancer, kidney cancer and cancer of other urinary organs were respectively 9% understated, 14% overstated and 29% overstated (Table 4).

The 2021 release of CdiA included cancer of other urinary organs for the first time. The release of this and other general groups was done to provide more complete cancer reporting information within the public domain. With the release, it was noted that these ‘other’ groups highlighted inconsistencies between ACD and NMD reporting (for example, incidence of ‘other urinary organs’ was relatively stable, survival was decreasing marginally but the mortality data indicated rapid increases – the rapidly increasing mortality was not consistent with these incidence and survival trends so the inconsistency between ACD and NMD coding becomes more apparent).

The 2022 release of CdiA provides more detail in reporting such as separating cancer of other urinary organs into its component parts of renal pelvis cancer, ureteral cancer, urethral cancer and cancer of overlapping and unspecified urinary organs (paraurethral cancer is part of the urinary tract but is excluded from CdiA reporting and the following analysis because there are most commonly zero cases and deaths reported). For the reasons outlined above, when compared with the ACD, the NMD respectively under-counts deaths for these more specific cancer types by 96%, 63% and 71%, and overstates deaths by 1,600% for cancer of overlapping and unspecified urinary organs (noting the 1,600% is of a small number) (Table 4).

The NMD likely over-states deaths in cancer in overlapping and unspecified urinary organs because there is only sufficient information to identify this broader cancer site rather than a more-specific cancer site. This is because, most commonly, only the term “transitional cell carcinoma” is provided on the death certificates.

Note: the difference is calculated as the number of deaths from the NMD minus the number of deaths from the ACD and this difference is also presented as a proportion of the ACD.

Source: National Mortality Database and Australian Cancer Database 2018

In regards to the cancer of overlapping and unspecified sites in the urinary tract, the difference between the ACD and NMD has become greater in more recent years. It should be noted that not all ‘other and unspecified’ cancer sites within the NMD are overstated when compared with the ACD. However, these groups may be more prone to being over-stated in the NMD.

The reporting of colorectal cancer deaths through the NMD has historically been challenging. Previously, colorectal cancer reporting was under-reported because many deaths were recorded as cancer of the intestinal tract, part unspecified (ICD-10 code of C26.0). To address this, the Australian Bureau of Statistics recommended that colorectal cancer include C26.0 when reporting deaths from colorectal cancer (ABS advice discussing the issue). When C26.0 is included to arrive at an NMD count for colorectal cancer deaths, it is much closer to the count according to ACD data (see Table 5).

When compared with the ACD, the number of colon cancer (C18) and rectal cancer (C20) deaths are lower in the NMD. Cancer of the rectosigmoid junction (C19) is currently not reported on separately within the CdiA but, when compared with the ACD, is considerably greater within the NMD. Within CdiA reporting, C19 and C20 are both included within rectal cancer and when combined, rectal cancer deaths are greater in the NMD compared to the ACD.

Some likely reasons for the differences between colorectal cancer mortality data in the NMD and ACD include:

• The term bowel cancer is commonly used in Australia by doctors as an interchangeable term for colon cancer. The term bowel cancer is coded to C26.0 (Cancer of the intestinal tract, part unspecified) and colon cancer to C18.9 (Colon, unspecified). For statistical analysis it is recommended that these two codes are combined.
• C19 – the term colorectal cancer is often used on death certificates and the term is coded to C19 (Cancer of the rectosigmoid junction). 

In future, it is unlikely that AIHW will publish colon or rectal cancer deaths separately using the NMD because of the relatively large differences between the NMD and ACD. As noted above, the information received on death certificates is unlikely to enable finer level reporting such as colon or rectal cancer from the NMD but it is suitable for reporting at the broader level of colorectal cancer. However, rectal and colon cancer mortality from the NMD has continued to be published in this edition as part of the mortality data investigations. Like many of the broader cancer sites and groups, colorectal cancer reporting within the NMD aligns much more closely to the ACD. 

Notes:

1. C26.0 forms part of the ICD-10 codes to generate colorectal cancer when using the National Mortality Database (discussed in above paragraphs).
2. The difference is calculated as the number of deaths from the NMD minus the number of deaths from the ACD and this difference is also presented as a proportion of the ACD.

Source: National Mortality Database and Australian Cancer Database 2018

Future work

At the time of this CdiA release, AIHW had not completed the full range of mortality data investigations. This is a large body of work, which is likely to be complex and time consuming. The future work and reporting discussed in the following paragraphs are likely directions but may be refined as the project continues.

The next stage of investigations is to compare underlying cause of death information in the ACD with the NMD at the unit record level. The results should enable more definitive advice to be provided and improved understandings of cancer mortality information.

By examining cause of death at the unit record level for cancers such as liver cancer, it may be possible to determine whether the cause of death information on liver cancer within the NMD may include some metastatic cancers. For example, this could occur where liver cancer deaths reported in the NMD have no corresponding record of the person being diagnosed with liver cancer (noting the ACD only includes primary cancers, not secondary (metastatic) cancers).

While the NMD may have access to less information from which to identify underlying causes of death for cancer than the ACD does, it has a longer historical time-series and is more up-to-date. At the time of releasing the CdiA, mortality data in the NMD was sufficiently complete for reporting up to 2020, while the ACD incidence data was sufficiently complete for reporting up to 2018 (with some estimation of late registrations) and ACD mortality data was complete up to 2017. It is possible that, like the 2018 incidence data, there may still be some outstanding mortality information that is yet to be provided to the ACD. Accordingly, the 2017 ACD may under-count mortality to some extent. This will need to be investigated more fully but preliminary work suggests this is likely to be occurring to some extent.

The AIHW is liaising with the ABS in regard to cancer mortality data. Through the investigations and liaison with ABS, it is expected that it may be possible to improve cancer mortality understandings and reporting.

Mortality reporting for Cancer data in Australia (2022 release)

At present, AIHW has included actual ACD mortality data from 2007 to 2017 and projections up to 2022. It has also produced NMD mortality reporting from 1971 to 2020 and projections up to 2022. Within the Cancer mortality by age and Cancer risk data visualisations, there is some general information to help identify which source of data is appropriate for the selected cancer. This general information will soon be accompanied by guidelines to better help people who may wish guidance on how to select the most appropriate source of data. The general information and upcoming guidelines are offered for assistance only, these are not intended to be prescriptive and it is acknowledged that users of the data may wish to undertake their own analysis to select the most appropriate data source for their needs. 

The AIHW usually also reports on cancer mortality by state and territory. The preliminary data investigations are focussing on national data. Cancer mortality statistics by state and territory will not be published by AIHW within the CdiA report until further investigations have been completed to a sufficient standard to release mortality statistics for state or sub-state geographic areas.

Mortality data is normally available within the CdiA summary statistics data visualisations. They have not been included this year as the reporting is too complex to include within this general information page. Mortality data is published within the ‘Cancer mortality by age’ and ‘Cancer risk’ data visualisations. 

Future reporting

The timeliness of the NMD ensures that, wherever its reporting for a cancer is of a sufficient consistency with the ACD, it is recommended for mortality reporting. However, where the NMD produces cancer mortality statistics that are considered to be particularly distant from the ACD results, they are not expected to be released in the CdiA in future.

The AIHW aims to investigate whether it is possible to use the ACD to derive estimated pre-2007 deaths from cancers . If successful, this information will be published to provide a longer time-series of cancer mortality statistics from the ACD.

It is possible that ACD mortality and NMD mortality will continue to be released in the future in some form. However, it is an aim of the project to provide users with cancer mortality statistics that are as simple to use as possible and meet users needs. As the release of data from multiple data sources to report on one item would not achieve the desired simplicity, this is hoped to be remedied in future CdiA reports.