In Canada, more than 22 000 patients were receiving chronic dialysis in 2009, increasing by nearly 20% in five years [1]. Cardiovascular disease (CVD) is the leading cause of mortality among dialysis patients, followed by infections [1]. It is well-established that dialysis patients have a higher mortality rate than the general population (GP) [1–3], and some data suggest that they have a higher proportion of death due to CVD than the GP [4, 5].

Understanding causes of death in a given population is important, as this data may guide prevention efforts, patient management, and clinical or research resources allocation. Many countries, including Canada, rely on registry data to produce estimates of causes of death for their dialysis population. In the GP, death certificates are considered the best practical source of truth at a population level. Since cause of death is measured using registry data in dialysis patients and death certificates in the GP, comparisons of cause of death proportions between dialysis patients and the GP may be biased. Reports comparing death certificates with dialysis registry data have demonstrated poor cause of death concordance [6, 7]. However, the quantification of the difference relative to GP estimates has not been undertaken. Using a single source of truth, such as death certificates, through linkage of databases may produce more reliable estimates and may drive better clinical, research and administrative decisions.

Interestingly, while many CVD-oriented interventions such as use of statins [8, 9] or angiotensin converting enzyme inhibitors [10] have resulted in some improvement in outcomes (such as reduction in atherosclerotic events), none have been demonstrated to lower mortality in the dialysis population. While pathogenic processes for CVD are probably different in the dialysis population compared to the GP, the inability of large studies to demonstrate improved survival, may suggest that attention to CVD deaths alone is insufficient to reduce mortality. Attention to other important contributors, such as infections, may be important [3]. In order to inform decision-making more accurately concerning intervention targets, assessment of causes of death proportions among dialysis patients should be ascertained using the same source of truth as in the GP. Our aim was to compare the proportion of deaths attributed to CVD, malignancy, and infections between patients receiving dialysis and the GP using death certificates for both, and to quantify the magnitude of discrepancy between registry and death certificate estimates in dialysis patients.

As shown in Figure 1, 5997 patients were identified in the CORR cohort and 6567 patients in the RAMQ cohort. After merging both cohorts, 5858 patients remained in the study cohort. Median age was 68 years (interquartile range: 56 – 75), 39.7% were female, and the majority was on hemodialysis (84.2%) versus peritoneal dialysis (15.8%) at three months of dialysis initiation. Median follow-up time was 2.0 years (1.0 – 3.4) using ISQ data.

Among study patients, 1907 deaths were identified using CORR data, while 2074 were identified using ISQ data. In total, 96.4% of patients had a concordant death status: 3778 were still alive at the end of the study according to both data sources and 1872 had a concordant date of death (same or contiguous month). The remaining patients with non-concordant death status were distributed as follow: 6 were still alive in ISQ but deceased in CORR, 29 had a non-concordant date of death (by a median absolute difference of 120 days), and 173 patients were still alive in CORR but deceased in ISQ. Kappa statistics for concordance between dates of death was excellent: 0.92 (95% CI: 0.91, 0.93).

All-cause and CVD mortality rates

All-cause mortality and CVD rates for the dialysis cohort (using CORR or ISQ dates of death) and GP are presented in Figures 2 and 3. Mortality rates were considerably higher in the dialysis population than in the GP. Mortality rate using CORR was lower when using ISQ (10.7%). Differences between CORR and ISQ dates of death also influenced the cumulative survival function, as shown in the Kaplan-Meier curve (Figure 4). CVD mortality rates were lower when using CORR data than using ISQ data, but the gap appeared smaller than with all-cause mortality.

This study reports on the comparison of cause of death proportions between dialysis patients and the GP. It demonstrated poor concordance between registry data and death certificates. This discrepancy between registry data and death certificates led to different conclusions when proportions of death attributed to CVD were compared between the dialysis population and the GP. While CORR showed a higher CVD death proportion among dialysis patients, ISQ showed almost no difference compared to the GP. While the proportion of deaths due to infection was also higher among dialysis patients using CORR data than when compared to underlying cause of death, proportions estimated from direct cause of death in the death certificate showed that they remained an important complication of disease (biological mechanism) leading to death. Finally, cause of death was missing in more than a quarter of deaths in CORR, greatly diminishing its usefulness.

Many countries and regions collect and report mortality data from renal dialysis programs using registries. Examples include the United States Renal Data System (USRDS), United Kingdom Renal Registry (UKRR), European Dialysis and Transplant Association/European Renal Association (EDTA/ERA), and Australia and New Zealand Dialysis and Transplant Registry (ANZDATA) [18]. Most dialysis registries use an internal coding system to classify cause of death instead of ICD-10 [2, 19]. Reports of distribution of causes of death for patients on dialysis vary between countries, influenced in part by different age distributions and case-mix among populations. CVD remains the most frequent cause in all (31 to 55% of deaths), followed in most countries by infections (11 to 18%), and malignancies (4 to 13%) [19–21]. Of note, distribution of causes of death reported for the GP are age and sex adjusted to the dialysis cohort. Therefore, causes of death associated with older age have a higher weight, explaining why malignancies are the leading cause of death at 36.1% in the GP.

Our finding of only fair concordance of cause of death between ISQ and CORR is consistent with results from previous reports using USRDS and ANZDATA (overall kappa from 0.22 to 0.24) [6, 7]. One may hypothesize that, in most of those dialysis populations, reported CVD death proportions are likely higher than what would have been reported using DC. To our knowledge, no study has quantitatively evaluated the proportion of deaths attributable to different causes in comparison to GP, using the same source of information (DC) for both populations.

In the UKRR, part of their data (England and Wales) is linked with national vital statistics [18, 22]. In USRDS, only 1% of deaths are ascertained through national vital statistics [2]. However, reporting death to USRDS is mandatory. CORR is a voluntary registry with no linkage to vital statistic registries. This may account for some of the discrepancies observed in our analysis. Given that information from registries (such as cause of death) is precious to guide policy, planning and resources we insist that better identification of true causes of death in dialysis populations is important, and that we do have an opportunity to remedy to the current situation in Canada.

While validity of death occurrence and its date is fairly high in vital statistic registries such as ISQ [23], misclassification of the cause of death is known to occur [7, 24, 25]. Of note, neither CORR nor ISQ is a true gold standard for cause of death, which will always be subject to some misclassification. DC, obtained from the ISQ data, constitute the most valid practical tool available to determine the cause of death on a provincial level. This study did not evaluate differences between registry and death certificates coding variation by country. However, as mentioned earlier poor concordance was reported in two other countries, so it is likely that this is a true phenomenon.

In conclusion, evaluation of causes of death using DC in dialysis patients in Quebec revealed that they do not have substantially different proportion of death due to CVD or infections than the GP. However, infections appeared to be a frequent complication or biological mechanism leading to death. While all-cause and CVD absolute mortality rates remain many times higher among dialysis patients compared to the GP (no matter which source of data), use of registry data alone would lead health care practitioners and researchers to conclude that there is more than 10 percentage points higher proportion of deaths attributed to CVD in dialysis populations than is really the case. In an era of increasing need for accuracy in medical records, we believe that validation of registry data, or development of better linkage with vital statistics in Canada, should be undertaken for dialysis populations. In this way, we may be able to better establish priorities for prevention and therapeutic efforts, or clinical and research resources allocation. While CVD remains an important issue among dialysis patients, infections appear to be an important target to consider as a means to reducing mortality in that population.