Cancer data in Australia

Example

This example is provided to help interpret the findings for the data visualisation where the selection is anal cancer. Help is also available within the visualisation. By hovering the cursor over various figures, written descriptions of results are provided to further help interpret data.

Figure 1 shows there were 329 records where the cause of death in the Australian Cancer Database (ACD) and/or National Mortality Database (NMD) is anal cancer. There were 148 (45%) where the cause of death was anal cancer in both datasets, 146 (44%) where the cause of death was anal cancer only in the ACD and 35 (11%) where the cause of death was anal cancer only in the NMD.

Figure 2 provides the detail of what the NMD recorded the cause of death was for the 146 records where the ACD recorded anal cancer and the NMD did not. The following summarises the results:

• 50 (34%) rectal cancer (excluding rectosigmoid junction)
• 37 (25%) other colorectal cancer (excluding rectal cancer)
• 22 (15%) non-melanoma skin cancer
• 16 (11%) cancer of unknown primary site
• 13 (9%) melanoma of the skin
• 8 (5%) other cause of death

Figure 2a examines the diagnosis information within the ACD and confirms that 145 of the 146 deaths (99%) recorded have a corresponding diagnosis of anal cancer. 

Figure 2b examines the diagnosis information within the ACD and confirms that:

• 6 (4%) of the NMD records have a corresponding diagnosis of the cancer that the NMD noted caused death. 
• 100 (68%) had no record of the cancer that caused death from the NMD of ever being diagnosed.
• 23 (16%) NMD records had an unconfirmed viability of the cause of death.
• 17 (12%) of NMD records had an imprecise cause of death.

Interpreting these results, there is a corresponding diagnosis of anal cancer for almost all of the records where the ACD records anal cancer as the cause of death and the NMD does not. Conversely, 68% of the NMD records had a cancer cause of death but no corresponding diagnosis. Where a notifiable cancer is the cause of death, there should be a corresponding diagnosis in the ACD. Most of the NMD records were of a close by site in the colorectal region. Interpreting these results, it is most likely that anal cancer was the cancer diagnosed but, in many cases where the NMD differed, it recorded a close by site where there was no record of the cancer being diagnosed. The ACD cause of death of anal cancer seems more likely accurate in that it was supported by a diagnosis. 

An additional 12% of records had an imprecise cause of death such as unknown primary site within the NMD. For these records, the ACD cause of death was anal cancer, while the NMD cause of death was based off less information and an imprecise cause of death such as cancer of unknown primary site was recorded. The NMD cause of death in this instance is considered imprecise.

There were 23 records where the NMD cause of death was categorised as 'unconfirmed viability'. These occur where the cause of death in the NMD was not a cancer and therefore the diagnosis information within the ACD was of no value in considering the accuracy of the NMD cause of death. While for the ACD records the diagnosis of anal cancer is supported by a diagnosis, these investigations cannot suggest that this cause of death should be used over the NMD or vice versa. It can only confirm where a cancer that caused death was also diagnosed and therefore a viable cause of death.

It should be noted that not all non-melanoma skin cancer deaths will have a corresponding diagnosis within the ACD. This is because basal and squamous cell carcinomas of the skin are not notifiable cancers and therefore not recorded in the ACD. Because of this, many non-melanoma skin cancer deaths will be categorised as ‘unconfirmed viability of the cause of death’.

Figure 3, 3a and 3b are the same as Figure 2 but instead focusses on where the NMD records anal cancer as the cause of death and the ACD does not. The results of this were similar in that the NMD cause of death was less likely to be supported by a diagnosis.

Differences in the recording of cause of death between the Australian Cancer Database and National Mortality Database: unit record analysis

Background

Prior to 2022, the Australian Institute of Health and Welfare (AIHW) only used the National Mortality Database (NMD) for reporting cancer mortality statistics. In 2022, the Cancer data in Australia report also included cancer mortality counts and rates for selected cancers derived from the Australian Cancer Database (ACD).

It was decided to release the results from the ACD because AIHW internal investigations suggested that the NMD was substantially under- or over-stating the number (and hence rate) of deaths for some cancers. Accompanying the counts and rates were cancer data commentaries that provided insights about why the two datasets may differ. Recommendations of which dataset to use for which cancers was also provided. Please refer to Cancer data commentary 8 and Cancer data commentary 8b for more information on this earlier work.

The Cancer Mortality Project (CMP) broadly aims to understand and provide information about the differences between the NMD and ACD cancer mortality counts and rates. The above-mentioned work in 2022 represented the beginnings of the CMP, and this paper represents the final part of the CMP in its current form.

This paper summarises the nature of the differences between the ACD and NMD cause of death (COD) for selected cancers. In particular, where one dataset records a type of cancer as the cause of death, this investigation summarises how often the other dataset agrees and, where it differs, the nature of the disagreement.

There are two main outputs from this work:

• Summary of findings
• Data visualisation

The dataset used in the investigation comprised the records from the 2019 ACD where the death occurred in 2018 or 2019, the ACD had recorded a COD, and the record could be linked to the National Death Index (which contains the NMD COD).

If a person’s cause of death is a type of cancer, they should have a corresponding diagnosis record on the ACD. The only exceptions to this are when the cancer was diagnosed before the beginning of the ACD, which is 1982, or when the COD is basal or squamous cell carcinoma of the skin because these cancers are not available in the ACD. If there was a corresponding diagnosis, the cause of death was categorised as “viable”. If there was no corresponding diagnosis, the COD was categorised as “no corresponding diagnosis”. The latter kind of COD is less likely to be accurate because Australia’s cancer notification system is very mature and notification is considered near complete (see Quality Statement for 2020 Australian Cancer Database).

For all cancers investigated, the ACD cause of death was much more likely to be viable than the NMD COD. A likely explanation for this lies in the available sources of information on cancer for each dataset. The NMD COD is derived from the Medical Certificate of Cause of Death (MCCD) for doctor-certified deaths and the coronial brief for coroner-certified deaths. It is reliant on the accuracy and specificity of the information recorded by the certifying practitioner on the MCCD. On the other hand, cancer registries have the person’s history of cancer diagnosis information as well as the death certificate from which to derive the cause of death. This can, in some cases, enable a more accurate COD to be determined for the ACD. There is no equivalent opportunity for the NMD.

Summary of findings

Where the ACD and NMD differ in relation to the cause of death, there are recurring patterns in the nature of disagreement that can be generalised. For example, the ACD may record colon cancer as the COD but the NMD often records a nearby site such as the rectum or rectosigmoid junction.

The following describes different types of disagreement that occur between the ACD and NMD recording of the COD. 

The case when both datasets record a cancer cause of death and the cancer sites are different but very near one another

For most cancers, there is a considerable amount of agreement on the cause of death between the two datasets. Where the NMD and ACD CODs differ, it is quite common for the two cancer sites to be close to one another. 

Example: stomach cancer

Stomach cancer results provide a useful illustration of the above. Here, there were 2,937 records within the analysis where stomach cancer was recorded as the cause of death in either the NMD, the ACD or both. Of these, 1,995 (68%) had stomach cancer as the COD in both the ACD and NMD. A further 728 (25%) had stomach cancer as the COD in the ACD only and 214 (7%) had stomach cancer as the COD in the NMD only.

Most of the disagreement between the datasets occurred where the ACD recorded stomach cancer and the NMD had a different COD. Focussing on these 728 records, 522 (72%) had the NMD COD as "cancer of the oesophagus, part unspecified" and a further 6 records as cancer of the lower third of the oesophagus. Overall, where the NMD and COD disagreed with the ACD COD, the NMD COD was often cancer of the oesophagus, which is a site next to the stomach.

Of the 728 ACD records where the ACD recorded stomach cancer as the COD and the NMD recorded a different COD, the ACD COD was viable in every case. Conversely, the NMD COD was viable in only 14 (2%) of these records.

Of the 528 deaths coded to oesophageal cancer in the NMD for which the ACD recorded the COD as stomach cancer, 513 (97%) of the ACD deaths were coded to “cancer of cardia”. These were likely carcinomas of the gastro-oesophageal junction, this site being where the oesophagus joins the stomach.

For stomach cancer, the ACD is likely to be the more accurate dataset due to some of these deaths being categorised as oesophageal cancers in the NMD. For a broader reporting group such as “cancers of the digestive system”, the coding of some stomach cancers to oesophageal cancer is not of concern.

Note: For blood cancers, when the ACD and NMD have different cancer CODs it is often the case that both CODs are blood cancers (for example, different types of leukaemia may be recorded). There are exceptions, but in general where the two datasets disagreed on the cancer COD the two would agree that it is a blood cancer.

The case when the NMD records an imprecise cause of death

For most cancers, some disagreement between the ACD and NMD COD occurs where the ACD records a specific cancer site but the NMD records an imprecise site. In these instances, the NMD COD was categorised as an “imprecise” cause of death rather than “no corresponding diagnosis”.

When deciding upon the COD, if the precise site of the cancer is unknown, a less precise site code can be used. The NMD COD is derived without access to the person’s cancer diagnosis records. The more limited information is more likely to lead to the recording of a less precise cancer COD than in the ACD.

Where the ACD records an imprecise cause of death, it is because there is no diagnostic information to inform a more precise site. Where the NMD records an imprecise site, it may be that there are no diagnostic results to inform a more precise site, but it may also be that the certifying doctor was not aware of, or did not have access to, existing information that stated a more specific site.

Example: hypopharyngeal cancer

There were 198 records where hypopharyngeal cancer was recorded as the cause of death in either the NMD, the ACD or both. Of these, 74 (37%) had hypopharyngeal cancer as the COD in both the ACD and NMD, 94 (47%) had hypopharyngeal cancer as the COD in the ACD only and 30 (15%) had hypopharyngeal cancer as the COD in the NMD only.

Of the 94 records where hypopharyngeal cancer was the COD in the ACD only, the ACD COD was viable in 99% of records and the NMD non-hypopharyngeal cancer COD was viable in 4% of records. However, 37 (39%) of the NMD records were categorised as an imprecise COD. There were several types of imprecise COD in this analysis. The most common was where the NMD recorded the COD as “cancer of pharynx, part unspecified” (23 records, 24%), followed by “cancer of unknown primary site” (8 records, 9%) and “cancer of head, face, and neck” (6 records, 6%).

“Cancer of the pharynx, part unspecified” may be used where it is known that the primary site of the cancer is within the pharynx but nothing more specific is known in relation to the site. Similarly, “cancer of the head, face, and neck” is a general COD where it is known only that the primary site is in the head and neck region. “Cancer of unknown primary site” is used when there is no information at all about the location of the primary site.

For hypopharyngeal cancer, the ACD is likely to be the most accurate dataset due to some of these deaths being coded to “cancer of the pharynx, part unspecified” in the NMD. For a broader reporting group such as “cancers of the lip, oral cavity and pharynx”, the coding of some hypopharyngeal cancer deaths as pharyngeal cancer is not of concern.

Both sources record a different cancer cause of death and both are viable

Overall, where the NMD and ACD recorded a different cancer COD, there were generally only a small proportion of records in which both CODs were viable. This situation can occur if a person has had multiple cancers diagnosed and the different datasets record a different but viable COD.

When a person has multiple primary cancers, the certifying doctor may not be able to determine with great certainty which cancer was the underlying cause of death. Nevertheless, the doctor will assess the case to the best of their knowledge based on the information available to them and give their opinion as to the underlying cause. The cancer registry may reach a different conclusion if it has access to records which the doctor did not. This could happen, for example, if the patient’s recent medical history was spread across more than one hospital. In any case, it is not possible to determine whether the ACD or NMD is more likely to be correct in this situation (without a review of the information available to cancer registries). This investigation has shown that this situation is uncommon.

The case when one dataset records a cancer cause of death while the other source records a non-cancer cause of death

There were occasions where one dataset recorded a cancer cause of death while the other dataset did not. This investigation cannot determine which dataset is more likely to be correct.

Example: myelodysplastic syndrome

There were 1,462 records within the analysis where myelodysplastic syndrome (a blood cancer) was recorded as the COD in either the NMD, the ACD or both. Of these, 728 (50%) had myelodysplastic syndrome as the COD in both the ACD and NMD, 530 (36%) had myelodysplastic syndrome as the COD in the ACD only and 204 (14%) had myelodysplastic syndrome as the COD in the NMD only.

Of the 204 records with myelodysplastic syndrome as the COD in the NMD only, the ACD COD was non-cancer for 120 records (59%), which is an unusually high proportion of records of this type. Another unusual finding was that the NMD COD was viable for 94 (46%) of the 204 records. This is a higher proportion of records of this type than occurs for the other cancers investigated. We are unable to explain these findings at this time.

Acute myeloid leukaemia

The results for acute myeloid leukaemia (AML) were quite different from those of the other cancers analysed. The most likely main reason for this is that some blood cancers can transform into AML. Transformation means that:

• the cancer does not satisfy the definition of AML when it is first diagnosed but does satisfy the definition of another blood cancer, and
• the characteristics of the cancer change over time and eventually satisfy the definition of AML.

If a cancer transforms into AML, the AML is not considered to be a “new” cancer and the event does not contribute to the count of new cancers diagnosed in that year. The fact and date of transformation may be recorded by the cancer registry (this varies) but the type of cancer and date of diagnosis remain as they were originally. 

However, these deaths are often recorded differently between the ACD and NMD. The ACD records the original blood cancer as the COD, which is in line with coding guidelines for ICD-O. On the other hand, the NMD records the AML as the COD, which is in line with international coding rules for mortality. In most cases the international coding rules for mortality identify the originating condition leading to death as the COD. However, there are some cases, such as transformations to AML, where a different cause of death is preferred for public health purposes. While this leads to differences in the ACD and NMD from a COD perspective, the coding in the NMD is consistent with that in other countries’ mortality datasets.

There were 2,216 records within the analysis where AML was recorded as the cause of death in either the NMD, the ACD or both. Of these, 1,402 (63%) had AML as the COD in both the ACD and NMD, 166 (7%) had AML as the COD in the ACD only and 648 (29%) had AML as the COD in the NMD only.

When considering the 648 records where the NMD recorded the COD as AML and the ACD did not, AML was not the initial blood cancer diagnosis for 576 (89%) of the records. However, in a large proportion of cases, the ACD COD was a cancer that is known to be capable of transforming into AML. These included 333 records (51%) whose ACD COD was a myelodysplastic syndrome and 126 records (19%) whose ACD COD was a myeloproliferative neoplasm.

When considering the 166 records where the ACD COD was AML and the NMD was a different cancer, the ACD COD was categorised as “no corresponding diagnosis” for 30 cases (18%). This is an unusually high proportion of cases for the ACD COD to not be supported by a corresponding diagnosis. A possible explanation is that the AML had transformed from a prior blood cancer and the cancer registry had recorded AML as the COD rather than the original cancer.

Non-melanoma skin cancer

The comparison of CODs between the ACD and NMD is more difficult when non-melanoma skin cancer (NMSC) is a possible cause of death. This is because the vast majority of cases of NMSC are basal or squamous cell carcinomas, which, as explained earlier, are not recorded in the ACD. If the death certificate asserts that the cause of death was basal or squamous cell carcinoma of the skin, the cancer registry is unlikely to have any record of the cancer or the person. However, in some cases the registry will have a record of the person because they have been diagnosed with other cancers and in some of those cases the registry might have good evidence that one of those cancers is more likely to be the cause of death than NMSC.

For several sites of head and neck cancer and anal cancer, NMSC more commonly appeared as the NMD COD than it did for most of the other cancers investigated. For example, of the 232 records where the ACD had mouth cancer as the COD and the NMD did not, 22% (50 records) had the NMD COD as NMSC. For lip cancer, it was 33% (12 out of 36 records), for tongue cancer 13% (24 out of 183) and for anal cancer 15% (22 out of 146).

Of the 54 records where the NMD recorded parotid gland cancer as the COD and the ACD did not, 43% (23 records) had NMSC as the ACD COD. The NMD COD was categorised as “no corresponding diagnosis” for 94% (51) of the records.

While there are some differences in the classification, the NMD reports on all NMSC deaths while the ACD’s mortality reporting for NMSC is less complete as it excludes basal and squamous cell carcinomas of the skin. 

Eye cancer and melanoma of the skin

There were 166 records within the analysis where eye cancer was recorded as the cause of death in either the NMD, the ACD or both. Of these, 60 (36%) had eye cancer as the COD in both the ACD and NMD, 103 (62%) had eye cancer as the COD in the ACD only and 3 (2%) had eye cancer as the COD in the NMD only.

Of the 103 records where only the ACD recorded eye cancer as the COD, melanoma of the skin was the COD in the NMD for 86 records (83%), which is a very high proportion.

For these records, it is possible that the death certificate stated the COD to be “melanoma”, rather than “melanoma of the eye”. Melanoma without specification of primary site (and without specification that the primary site is “unknown”) is assigned the same ICD-10 code as melanoma of the skin (C43). One reason that the death certificate might have stated just “melanoma” could be that the certifying doctor was unable to determine the primary site from the available information. Another reason could be that the doctor knew the primary site but did not realise that it was important to state it in the case of melanoma.

If interested in eye cancers, the ACD is likely the most accurate data source. Some of these deaths are categorised as melanoma of the skin in the NMD.

Future work

The Cancer Mortality Project (CMP) began by providing two datasets as sources of mortality data in order to highlight potential issues in cancer cause of death reporting. Commentaries were released to provide a better understanding of, and speculate upon the reasons for, possible differences. The present work concludes the CMP and generally supports the previous recommendations to use the ACD for certain cancers.

It should be remembered that the analysis in this paper focusses only on the cancers where the ACD is the recommended dataset; there are many other cancers where the ACD and NMD mortality results are better aligned.

A common reason for differences between the two datasets is for the NMD to have a COD from a nearby site (or a similar type of blood cancer) or an imprecise COD. The NMD can only record the COD based on the information within the death certificate (unless there is a coroner’s report), which may have limitations in its accuracy. The NMD COD is derived without the benefit of access to the cancer diagnosis information that the cancer registries possess. Without such information it is not possible to verify that the deceased person has ever been diagnosed with the cancer named on the death certificate.

While it is possible that this analysis has produced some results that will be of benefit to cancer registries and/or the ABS in improving the accuracy of the coding of COD in their respective datasets, ongoing differences in the COD between the datasets are to be expected due to differing sources of information about the deceased person.

In the future, AIHW will continue to release cancer mortality data from both the NMD and ACD.

AIHW’s publication of state and territory cancer mortality data was halted while the CMP was underway but recommenced in the August release of Cancer data in Australia. 

At present, the ACD cause of death time series only goes back to 2007. Future work may look at the feasibility of producing mortality data using the ACD for earlier years.

Notes on the method

This investigation only examined the cancers for which the ACD had previously been recommended for use. These are the cancers for which the total number of deaths according to the NMD and ACD were rather different. Some broad cancer reporting groups where the ACD was recommended were excluded from analysis. For these, only the cancers within the group where the ACD was recommended were analysed. For example, head and neck cancers (including lip) were not analysed as a group but the ACD-recommended cancers within the group such as lip, mouth and hypopharyngeal cancer were examined. Cancers where the ACD and NMD more closely align will also have discrepancies in the CODs but these differences will be smaller and both datasets are acceptable for aggregate reporting of cancer mortality.

The determination of viability was based on the usual AIHW reporting groups. For example, if the NMD COD was “C34.1 Cancer of upper lobe of lung” but the ACD said the person had been diagnosed with “C34.2 Cancer of middle lobe of lung”, the NMD COD was considered to be viable. Although the codes do not match exactly, they both belong to the reporting group “lung cancer” and hence the difference in codes does not impact lung cancer mortality statistics.

For efficiency and due to the relatively small number of records involved, an NMD COD of “C76 Cancer of ill-defined site” was considered imprecise unless the only diagnoses for that person were blood cancers. In that situation the NMD COD was classified as “no corresponding diagnosis”.

Where the two CODs differed, up to 8 groups were created to describe how the other dataset recorded the COD. The choice of groups has a degree of subjectivity. For example, there were 505 records where the ACD recorded the COD as extrahepatic bile duct cancer and the NMD did not. Within the NMD, 410 of these records had a COD of intrahepatic bile duct cancer and 7 had other kinds of liver cancer. These 417 records could have been combined into the single AIHW reporting group “liver cancer” but, for this analysis, it was considered important to make the distinction between intrahepatic bile duct and other liver cancer.