We performed a population-based cohort study of all adults and children starting RRT for treatment of ESKD in Australia (1 January 1980 to 31 December 2013) and New Zealand (1 January 1988 to 31 December 2012). Australia (population 24.8 million in 2017) and New Zealand (population 4.7 million in 2017) are broadly comparable in racial make-up and population demographics, where approximately 75% are from a European, Australian or New Zealand cultural background.11 12 Both countries have a universal healthcare system that provides free medical care in public health systems. All deaths occurring in each country are reported to the Births, Deaths and Marriages Registry with a medical certificate for the cause of death completed by a medical doctor which is then coded to the international standard, the International Statistical Classification of Diseases and Related Health Problems (ICD), codes by a qualified clinical coder, detailing both the underlying and secondary causes of death. The underlying cause of death is considered as the disease or condition which initiated the sequence of events resulting in death. Secondary or contributing causes of death are other diseases or conditions that contributed to the death but were not the underlying cause.

We used retrospective registry data for this study, hence study participants were not involved in the recruitment or design of the study. Patient advisers were also not involved in this study.

We summarised the main causes of death by deriving the leading causes of death, defined by ICD-10-AM codes.17 We also estimated the crude mortality rates for stroke and non-stroke deaths, overall and stratified by patient demographics.

We used Fine and Gray competing risks regression models to evaluate risk factors associated with stroke death.18 Time was measured from the start date of ESKD treatment until a participant died or 31 December 2013 (Australian participants) or 31 December 2012 (New Zealand participants), whichever came first. As probabilistic record linkage may lead to a small proportion of incorrect links,19 Australian patients were censored at the ANZDATA date of death if ANZDATA considered the patient had died and the national death registry had not captured any death. For the analysis of stroke deaths, all non-stroke deaths were considered as competing events. For comparison, we also similarly conducted an analysis of non-stroke deaths, with stroke deaths considered as competing events. For both the stroke and non-stroke analyses, we used the same covariates, selected a priori including: age at initiation of ESKD treatment, sex, year of initiating ESKD treatment, country, racial background, BMI category (underweight, ≤18.4 kg/m²; normal, 18.5–24.9 kg/m²; overweight, 25.0–29.9 kg/m² and obese, ≥30.0 kg/m²), initial ESKD treatment modality, comorbidities, smoking status and cause of ESKD. Univariable models were fitted for each covariate, and the multivariable models included all covariates. The cumulative incidence was estimated from the final multivariate model, by sex and year of initiating ESKD treatment, adjusting for all other covariates except BMI.

There were no missing data, except for BMI. Missing values for BMI were imputed using chained equations with five iterations. To impute, we used the same variables as those used in the competing risks analyses, then converted those BMI values into categories for the regression models.

We conducted a further three sensitivity analyses to determine: (1) whether risk factors for any stroke death were similar to risk factors for fatal ischaemic stroke and fatal haemorrhagic stroke; (2) whether risk factors were the same when stroke death was defined as underlying cause, or secondary cause when kidney failure was indicated as the underlying cause and (3) whether ever transplanted with a kidney as a time-varying covariate produced different model estimates. For these sensitivity analyses, we used cause-specific Cox regression models to evaluate associated risk factors and used the same covariates and definition of time at risk as described in the main analysis.

Data were analysed using Stata V.14 (Stata).

We identified a total of 62 296 people with incident ESKD in Australia and New Zealand between 1980 and 2013. Of these, 1473 New Zealand patients initiated RRT prior to 1988 or after 2012, and so could not be linked. The remaining 60 823 patients with ESKD were included in the analysis (figure 1). The fact of death for the majority of patients was similar between ANZDATA and the national death registries, where disagreement occurred for 7.6% of Australian patients and 5.5% of New Zealand patients. There were 467 patients considered to be lost to follow-up in ANZDATA, where 57 patients were considered to have died in the national death registry. There were 381 874 person-years (PYs) of follow-up, with a median follow-up time of 4.0 years (IQR 1.7–8.4 years).

During the study period, 26 505 patients (43%) remained in active follow-up and 34 318 patients (57%) died (table 1). Of the patients who died, 33 377 patients died of non-stroke causes (97.3%) and 941 patients (2.7%) died of stroke. In our entire study population, two-thirds (69%) were aged 50 years or older at ESKD, with a median age of 59 years (IQR 46–69). Most were from Australia (85%), were male (59%), had a Caucasian background (76%) and had started treatment for ESKD in 1996 or later (74%). The majority of patients had a normal (33%) or overweight (27%) BMI, where only 4% were underweight. The BMI was not collected for 13% of patients, the large majority (95%) of whom reached ESKD prior to 1996. The main cause of ESKD was diabetes (28%), followed by glomerulonephritis/IgA nephropathy (27%), other causes of ESKD (26%), hypertension/renal artery disease (12%) and polycystic kidney disease (7%). The reason for other causes of ESKD included uncertain diagnosis (23%), analgesic nephropathy (18%) and reflux nephropathy (16%).

There were a total of 34 318 deaths since initiation of ESKD treatment, with a crude rate of 89.9 (95% CI 88.9 to 90.8) per 1000 PYs. The top four leading causes of death were recorded as diabetes (18%), coronary heart disease (18%), kidney failure (13%) and glomerular disease (5%). Stroke was the sixth leading cause of death since RRT, with a total 941 stroke deaths, corresponding to a crude rate of 2.5 (95% CI 2.3 to 2.6) per 1000 PYs (table 2). Stroke deaths were due to: 108 (11%) ischaemic strokes; 259 (28%) intracerebral haemorrhages; 90 (10%) intracranial haemorrhages; 68 (7%) subarachnoid haemorrhages; 114 (12%) transient cerebral ischaemic attacks and related syndromes and 302 (32%) unspecified strokes (online supplementary appendix 1).

The highest rate of stroke death since initiation of ESKD treatment was among those with known cerebrovascular disease at 9.1 (95% CI 8.0 to 10.3) per 1000 PYs, followed by those aged 75 years or older at 6.6 (95% CI 5.6 to 7.7) per 1000 PYs (table 2). In comparison, the highest rate of non-stroke death since initiation of ESKD treatment was among those aged 75 years or older at 254.5 (95% CI 248.0 to 261.2) per 1000 PYs, followed by those with cerebrovascular disease at 199.0 (95% CI 193.7 to 204.4) per 1000 PYs.

The multivariable model suggested a higher risk of stroke death was associated with older age, female sex, earlier year of ESKD, normal BMI range, presence of pre-existing cerebrovascular disease, no previous malignancy, ESKD treatment with haemodialysis (HD) compared with transplant and hypertension/renal artery disease or polycystic kidney disease as the cause of ESKD compared with glomerulonephritis/IgA nephropathy (table 3).

Older age was associated with a higher risk of stroke death, where patients aged 75 years or older were 92% sub-HR (SHR 1.92, 95% CI 1.45 to 2.55) more likely to die from stroke death compared with those aged 30–49 years (p<0.001). Females were 41% more likely to die of stroke death compared with males (SHR 1.41, 95% CI 1.21 to 1.64; p<0.001). There was a decreasing trend in the risk of stroke death overtime, where patients initiating ESKD treatment in ≤1995 and 1996–2005 were 93% (SHR 1.93, 95% CI 1.56 to 2.39) and 30% (SHR 1.30, 95% CI 1.09 to 1.55), respectively, more likely to die from stroke death compared with those in 2006–2013 (p<0.001). Patients had over two times the increased risk of stroke death if they had pre-existing cerebrovascular disease (SHR 2.39, 95% CI 1.99 to 2.87; p<0.001) and were 36% less likely to die of stroke if they had a previous malignancy (SHR 0.64, 95% CI 0.53 to 0.78; p<0.001). There was no difference in the risk of stroke death for those who initiated PD (SHR 0.85; 95% CI 0.72 to 1.01) compared with HD. However, transplant patients were 73% less likely to die of stroke compared with HD patients (SHR 0.27, 95% CI 0.09 to 0.84). Similar estimates for the risk of stroke death were produced when transplant was considered a time-varying covariate (HR 0.24, 95% CI 0.17 to 0.32; p<0.001) (online supplementary appendix 2). The cause of ESKD was also significantly associated with stroke mortality (p=0.020). Patients with hypertension/renal artery disease or polycystic kidney disease were 39% (SHR 1.50, 95% CI 1.20 to 1.87) and 38% (SHR 1.38, 95% CI 1.00 to 1.90), respectively, more likely of stroke death compared with those with glomerulonephritis/IgA nephropathy. The unadjusted estimated risk of stroke death is presented in online supplementary appendix 3.

In the sensitivity analysis, the risk of fatal ischaemic stroke was not associated with BMI, cerebrovascular disease, previous malignancy or HD compared with transplant (online supplementary appendix 4). White racial background and coronary artery disease were associated with an increased risk of ischaemic stroke death, which were not risk factors for all-cause stroke death. Year of ESKD was also not associated with the risk of fatal ischaemic or haemorrhagic stroke. The risk factors for fatal haemorrhagic stroke were all similar to the risk factors for all-cause stroke death, except for cause of renal failure (online supplementary appendix 4). Cause of renal failure was not associated with the risk of fatal haemorrhagic stroke.

There were an additional 694 stroke deaths when all-cause stroke death was defined as the underlying cause or secondary cause when kidney failure was the underlying cause. All risk factors associated with the extended definition for all-cause stroke were the same as those identified in the main analysis, except for peripheral artery disease (online supplementary appendix 5). In the sensitivity multivariate model, patients with pre-existing peripheral artery disease were at an increased risk for all-cause stroke mortality.

The multivariable model suggested that a higher risk of non-stroke death was associated with older age, female sex, earlier year of ESKD, underweight or normal BMI range, Caucasian background, presence of comorbidities including cerebrovascular disease, coronary artery disease, peripheral artery disease and no previous malignancy, current or former smoker, dialysis compared with transplant and cause of ESKD including diabetes, hypertension/renal artery disease or other causes (table 3).

Patients initiating ESKD treatment with a transplant were associated with a 73% lower risk of non-stroke death compared with those initiating HD (SHR 0.27, 95% CI 0.23 to 0.32; p<0.001). Similar estimates for the risk of non-stroke death were produced when transplant was considered a time-varying covariate (HR 0.30, 95% CI 0.29 to 0.32; p<0.001) (online supplementary appendix 2). The cause of ESKD was also significantly associated with non-stroke mortality (p<0.001). A higher risk of non-stroke mortality was associated with diabetes (SHR 1.60, 95% CI 1.54 to 1.65), hypertension/renal artery disease (SHR 1.24, 95% CI 1.19 to 1.29) or other causes (SHR 1.38, 95% CI 1.33 to 1.42) compared with glomerulonephritis/IgA nephropathy as the cause of ESKD. Those with polycystic kidney disease were 24% less likely (SHR 0.76, 95% CI 0.72 to 0.80) to die of non-stroke compared with those with glomerulonephritis/IgA nephropathy. The unadjusted estimated risk of non-stroke death is presented in online supplementary appendix 3.

The estimated cumulative incidence of stroke mortality since RRT was relatively low at 0.5% at 2 years, while the non-stroke mortality was substantially higher at 15.5% at 2 years. At 5 years, the cumulative incidence increased to 0.9% for stroke and 36.8% for non-stroke mortality.

The cumulative incidence of stroke or non-stroke death differed by sex, where females consistently had a higher cumulative incidence (figure 2A). For stroke mortality, the cumulative incidence at 2 years in males was 0.4% and slightly higher in females at 0.6%. At 5 years, the sex difference increased where the cumulative incidence was 0.8% in males and 1.1% in females. For non-stroke mortality, the cumulative incidence at 2 years was similar in males at 17.6% and females at 17.4%. At 5 years, the cumulative incidence remained similar at 43.3% in males and 42.9% in females.

The cumulative incidence of stroke mortality decreased over time, where those diagnosed with ESKD in 2006–2013 had the lowest cumulative incidence during follow-up (figure 2B). At 5 years, the cumulative incidence increased to 1.6% in ≤1995, 0.9% in 1996–2005 and 0.7% in 2006–2013. The cumulative incidence of non-stroke mortality showed a similar trend. At 5 years, the cumulative incidence increased to 46.3% in ≤1995, 38.0% in 1996–2005 and 30.4% in 2006–2013.