Four notable findings emerged in the present study: (1) the sensitivity for diabetes, dementia, and cancer being recorded as the underlying cause of death was poor but improved substantially if contributing causes of death were taken into account; (2) women with dementia who died before 2006 were less likely to have dementia recorded on their death certificates, compared with those who died during 2006–2011 or 2012–2017. This pattern was the opposite for diabetes and cancer; (3) Sociodemographic characteristics were weakly and inconsistently associated with the sensitivity of these conditions recorded on death certificates; and (4) Diabetes and dementia were under reported particularly on the death certificates of women who died from circulatory diseases or cancer.
In this study, we found that diabetes was recorded as the underlying cause of death for only 12.3% of women with diabetes in their lifetime. This was broadly consistent with the sensitivity observed in an US study which enrolled 11 927 diabetics (10%)  and another US study which used data from the National Mortality Follow-back Survey (NMFS) (10%) , but higher than the sensitivity found in the US Rancho Bernardo Study (6.2%) . The Rancho Bernardo Study, which included 3 209 participants, found that the sensitivity of death certification of diabetes was higher in women than it was in men, which might explain why the authors found a lower sensitivity than we did as our study only had women participants. Whittall et al. investigated the accuracy of death certification of diabetes in 1 084 Caucasian subjects with diabetes who participated in surveys in rural areas in Western Australia from 1978 to 1982, and found 28% of these women had diabetes recorded as the underlying cause of death , which was much higher than our result observed in participants from Western Australia (13.4%). Whether this discrepancy indicates a decreasing trend in the accuracy of death certification of diabetes over the past decades in Western Australia warrants future investigation.
The accuracy of death certification of dementia has been evaluated in previous studies [6,7,8, 14,15,16,17,18,19,20]. We used survey data from a well-designed nationally representative cohort study linked to multiple sources of routinely collected healthcare data . In our study, 52.3% of women with dementia in their lifetime had dementia recorded as the underlying or contributing cause of death, which was similar to the sensitivity observed in cohort studies in South London (53.6%)  and Finland (44.4%, 95% CI: 13.7, 78.8) , but higher than the sensitivity found in other cohort studies in England and Wales (21.0% in the Cognitive Function and Ageing Study I (CFAS I)) , and the US (28.4% and 23.8%) [14, 17]. The inconsistency between our finding and the findings from these studies could be due to different time periods studied. In the CFAS studies, the sensitivity of dementia death certification increased from 21.0% in participants recruited in 1989 (CFAS I) to 45.2% in participants recruited in 2008 (CFAS II) . The inconsistency between the finding of our study and the findings of the US studies could also be explained by the different populations studied or different gold standards used. The sensitivity of dementia death certification in US studies was generally found to be low, even in people with end-stage dementia .
As the age range of the participants in our study was narrow (i.e., five years), the difference in the sensitivity of death certification of dementia across the three time periods (i.e., < 2006, 2006–2011, and 2012–2017) may not necessarily mean that there was a temporal change in the sensitivity of death certification, but could mean that women with dementia who died at an older age were more likely to have dementia recorded as a cause of death. Similarly, the difference in the sensitivity of death certification of cancer across the three time periods could either be because the sensitivity of death certification of cancer decreased over time or because some women who lived longer had recovered from cancer and died from some other cause. Changes to the automated coding of multiple causes of death occurred in 2006 and 2013. The first of these changes probably explains the changes in sensitivity of dementia coding observed before and after 2006 (see explanatory note #77 of the Australian Bureau of Statistics publication ‘Causes of Death, Australia, 2015’ ). However, the 2013 and any subsequent changes in processing, and slight differences between States and Territories, are less likely to explain the temporal changes in sensitivity of recording of causes of death.
Despite diabetes being one of the most important risk factors for cardiovascular diseases, only 40.9% of women with diabetes had diabetes recorded as a cause of death in our study. We observed that diabetes was under reported particularly on the death certificates of women who died from ischaemic heart disease. For these women it is unlikely that diabetes was not related to the ischaemic heart disease and not a contributing cause of death . We also found that diabetes was under reported in women who died from cancer. This finding was in line with the finding of a previous US study conducted among 11 927 people with diabetes . Diabetes has been increasingly recognized as a predictor of deaths from cancer of pancreas, breast, liver, and colon [23, 24].
The reasons for the underreporting of dementia on death certificates are multifaceted. First, cognitive impairment in people with dementia may restrict them from reporting symptoms and seeking medical help , which could explain our finding that unpartnered women were less likely to have dementia recorded on their death certificate than partnered women. Second, the co-existence of more than one health condition (i.e., multimorbidity) is especially common in older people , making it hard to identify dementia as the single underlying cause of death, especially when dementia co-exists with cardiovascular diseases. Third, having dementia on a death certificate could be used to challenge the validity of someone’s will , which may make the certifying doctor less likely to record dementia as a cause of death.
It should be noted that certifier type (i.e., a medical practitioner or a coroner) also affects how multiple cause of death are recorded on death certificates. In Australia, deaths from natural causes (e.g., cancer or diabetes) are usually certified by a medical practitioner, and most deaths caused by unknown or external causes (e.g., suicides) are certified by a coroner. Approximately 86% to 89% of deaths are certified by a medical practitioner , and multiple causes of death are more likely to be recorded for deaths certified by a medical practitioner. So multiple causes are more likely to be recorded for deaths involving diabetes, dementia and cancer than external causes. Previous analysis has shown that this cohort was more socio-economically advantaged and had higher relative survival than other women in Australia born in the same period . However, the results shown in Figs. 1, 2 and 3 do not suggest that socio-economic differences affected the sensitivity of recording of causes of death.
The under reporting of conditions on the death certificates (e.g., diabetes, dementia, or cancer) does not only result in the under estimation of mortality burden associated with these diseases, but also could change the underlying cause of death selected based on coding rules in the automated processing. Several recommendations have been proposed to improve the accuracy of death certification. First, as the current format of death certificates may have limited space, and it has been suggested that adding a series of check boxes for the most common conditions to death certificates could improve the underreporting of some conditions (e.g., diabetes) . In Australia electronic formats of death certificate are being adopted in some States/Territories and this allows for more conditions to be mentioned in Part II. Second, more educational support to help medical practitioners to improve the accuracy of death certificates could be helpful. Our results emphasise the importance of considering how causes of death (including underlying and contributing causes of deaths) are recorded, a challenge highlighted by the need to distinguish between death from COVID and deaths with COVID. Additionally, national statistics reporting both underlying and contributing causes of death would better represent the burden of disease.
Three main limitations of this study should be acknowledged. First, as this study was from ALSWH, only women were included, restricting our capacity to generalize the findings to men. Second, the narrow age range of the participants restricted the generalization of the research findings to all women in Australia. Third, the place of death (e.g., hospital or aged care facility) has been found to be associated with the accuracy of death certification for dementia [6, 7], but we were unable to obtain this information.