The National Diabetes Services Scheme (NDSS) was established in Australia in 1987 to deliver diabetes-related products at subsidised prices and provide information to people with diabetes. Registration of patients is carried out by a medical practitioner or certified diabetes educator. The NDSS captures 80–90% of all Australians with known diabetes . In total, 1,108,420 people were registered with type 2 diabetes on the NDSS between 1997 and 2011. The primary population for this analysis included 744,188 (67.1%) people with type 2 diabetes who had a date of diagnosis of diabetes listed on the NDSS. A secondary analysis included all type 2 diabetes registrants and used date of registration as a proxy for those whose date of diagnosis was missing (1,108,420 people).
Although the NDSS was initially established in 1987, 1997 was chosen as the start date, as this time period followed a unification of state-based registries and consequently an improvement in data quality. Diabetes type is classified by the health practitioner who completes the registration. For the current analysis, type 2 diabetes was assigned to all registrants who were classified as type 2 on the NDSS. Type 1 diabetes status was assigned to registrants who were recorded as type 1 on the NDSS registry, were registered at <45 years of age and were taking insulin. We chose 45 years as the cut-off to minimise the number of people with type 1 diabetes that we would miss without misclassifying significant numbers of people with type 2 diabetes as type 1 diabetes. In addition, registrants who were recorded as having type 2 diabetes on the registry and who were diagnosed before the age of 30 years and were taking insulin within 1 year of diagnosis were reclassified as having type 1 diabetes. All others were classified as having type 2 diabetes.
The NDSS was linked to the National Death Index up to 31 December 2011 by the Australian Institute of Health and Welfare as reported previously [15, 16]. The underlying cause of death was coded according to ICD-10 (www.who.int/classifications/icd/en). In the primary analysis, causes of death were classified as follows: cardiovascular disease (CVD) I20–I25, I60–I69 or ‘diabetes with circulatory complications’ (E10.5, E11.5, E12.5, E13.5, E14.5), cancer C00–C97 and other underlying causes of death (non-CVD and non-cancer). We also explored mortality from ischaemic heart disease (IHD) (I20–I25), stroke (I60–I69), pancreatic cancer (C25–25.9), lung cancer (C33–C35.9) and colorectal cancer (C18–C20.9), since they are among those with the highest cancer death rates in individuals with diabetes in Australia. Deaths with an underlying cause corresponding to ‘uncomplicated diabetes’ (E10.9, E11.9, E12.9, E13.9, E14.9) where a CVD, IHD or stroke ICD code also appeared in the first line of part I on the death certificate were also included in the respective group. We considered these deaths as due to CVD, IHD or stroke, as it is not possible to die from ‘uncomplicated diabetes’. The NDSS dataset linked to the NDI has previously been used to present age-specific mortality rates among those with diabetes .
Statistical analysis and reporting
Individuals were followed from 1 January 1997, or registration date if later, to 31 December 2011 or date of death, whichever occurred first.
Participants’ follow-up (risk time and deaths) was first split into intervals of 6 months by current age (0–100 years), calendar time (1997–2011 incl.) and diabetes duration (0–20 years). Risk time and deaths (overall and by cause) were subsequently tabulated by current age, date of follow-up and diabetes duration, and each cell of the table was assigned age, date and diabetes duration as continuous variables as the midpoint of the 6 month group. Age at diagnosis was calculated as current age minus diabetes duration. Data were analysed using a Poisson model, using log-person-time as the offset variable and spline effects of current age, diabetes duration and age at diagnosis, and with a linear effect of calendar time. The model is formally over-parametrised because age at diagnosis plus diabetes duration equals current age. Therefore, it is not possible to tease out the separate effects of age at diabetes diagnosis and diabetes duration. We only used the model for prediction of rates, and these do not depend on any particular parametrisation. Separate models were fitted for men and women.
The predicted mortality rates by current age for individuals diagnosed at different ages are shown in Fig. 1. To illustrate the effect of age at diagnosis and duration of type 2 diabetes, we computed the mortality rate ratio (also known as HR) between individuals diagnosed at ages 5 years apart as a function of age at follow-up (e.g. age 40 vs 45 at diagnosis, followed from age 45 to 60). Thus, each point on each curve indicates the mortality risk for a person diagnosed at one age compared with a person diagnosed at 5 years older, when both individuals have achieved the same age. Different curves were used to illustrate how this effect varies by age at diagnosis, by comparing individuals diagnosed at ages 45 vs 50, 50 vs 55, etc. The same was done for a 10 year difference in diabetes duration, by comparing individuals diagnosed at ages 40 vs 50, 45 vs 55, etc. The two sets of curves are shown in adjacent graphs in Fig. 2.
On the basis of the underlying cause of death for people with type 2 diabetes, we carried out the same analyses for cause-specific mortality from CVD, cancer and other causes of death.
We report results on two populations: the primary population is that for which age of diabetes diagnosis is available (n = 744,188). For this analysis, after excluding 479 registrants, because the registration date was the same as the date of death or the dates were implausible, the sample size was 743,709. However, given that only 67% of the entire NDSS type 2 diabetes cohort over the time period 1997–2011 had the date of diabetes recorded, the analysis was repeated for all registrants using the registration date as a proxy for those with a missing date of diagnosis (n = 1,108,420). After excluding 743 registrants because the registration date was the same as the date of death, or the dates were implausible, the sample size for the sensitivity analysis was 1,107,677.
All analyses and graphs were generated with R software, version 3.4.2 (www.R-project.org) (R Foundation for Statistical Computing, Vienna, Austria), using packages Epi and popEpi. The study was approved by the Alfred Health Human Ethics Committee and the Australian Institute of Health and Welfare Ethics Committee. All participants gave informed consent for their data to be used in the analyses. A complete account of the data analyses can be found at http://bendixcarstensen.com/IDI/mort/AUS-DM-mort.pdf. The data that support the findings of this study are not publicly available as they were generated under a licence which restricted access to named, approved investigators.