Data sources and study population
Data were obtained from the national Canadian dialysis registry, the Canadian Organ Replacement Register (CORR), and the provincial health services administrative databases of the province of Québec, Canada. CORR provides descriptive statistics on dialysis incidence, prevalence and patients’ characteristics, and its data have been used successfully in numerous scientific publications [11–15].
All Québec residents, more than 8 million inhabitants, are covered for their physician and hospital services by a universal single-payer health care system (Régie de l’assurance maladie du Québec – RAMQ). The RAMQ physician claim databases include all visits, diagnosis codes and procedures during in- or outpatient encounters. RAMQ also hosts the hospital discharge summary databases. The Institut de la statistique du Québec (ISQ) holds official governmental vital statistic databases, which include dates and causes of death as reported on the death certificate. Information on data sources is summarized in Table 1.
From CORR, RAMQ and ISQ, data were obtained for all patients initiating chronic dialysis (without a prior kidney transplant) between January 1st, 2001 and December 31st, 2007 in the province of Québec. Patients with less than 90 days of dialysis were excluded. The study cohort consisted of all patients who were present in both the CORR and RAMQ databases as incident dialysis patients. An incident cohort was used, since comorbidities and causes of death may highly depend on dialysis vintage. Patients were followed from day 90 after dialysis initiation until date of death or end of the study period.
Mortality rates in the GP of Québec were obtained from the ISQ website for the years 2001 to 2007 [16].
Measurement of dates and causes of death
CORR data provided a date of death (month and year) and a cause of death using an internal classification (78 elements). The cause of death is usually coded by the registered nurse responsible in each dialysis unit.
ISQ also provided a date of death (month and year) and a cause of death coded using the International Classification of Diseases, 10th Revision (ICD-10). Death certificates are filled by physicians and then coded by trained archivists at ISQ. For the evaluation of dates of death concordance, the date of death provided by RAMQ-ISQ was considered the source of truth.
The cause of death is mandatory on the death certificate (ISQ) and includes different fields: 1) underlying disease that eventually led to death; 2) diseases in the pathway to death (“secondary causes”); and 3) the disease or complication that directly led to death (“direct cause”). For example, a patient may have the following pathway: had an acute myocardial infarction (underlying cause), followed by a cardiogenic shock (secondary cause), and dies after a ventilator-associated pneumonia in the intensive care unit (direct cause).
Causes of death were classified in four mutually exclusive categories: CVD (ICD-10: I00-I99), infection (A00-B99, J10-J18), malignancy (C00-D48), and other. Among the “other” category, kidney failure (N17-N19) and diabetes (E10-E14) were identified using death certificate, but those categories had no code using CORR internal scheme.
Statistical analysis
Dates of death from CORR and ISQ were considered concordant if they occurred in the same month, or in a contiguous month. Concordance was measured using kappa statistics.
Mortality rates were calculated by dividing the number of deaths by the total patient-years of follow-up. 95% confidence intervals (CI) for rates were calculated using a Poisson distribution. Mortality rates for the GP were indirectly standardized using the study cohort age and sex structure. Cumulative survival function was calculated using Kaplan-Meier method.
Causes of death were considered concordant if they fell within the same category. Two concordance analyses were done for causes of death: 1) CORR versus ISQ underlying cause and 2) CORR versus ISQ direct cause. Concordance was measured by non-weighted kappa statistics for categorical variables [17], and by a Chi-square test for proportions.
Sensitivity analyses
Some codes in CORR classification system are broad and may include various categories: Cardiac Arrest, Cause Unknown; Patient Refused Further Treatment; Multi System Failure; or Other Identified Cause of Death. In the main classification, theses codes were classified as “Other” except Cardiac Arrest, Cause Unknown that was classified as CVD. To test the impact of this decision on the results, a sensitivity analysis was conducted by excluding patients who had one of these codes. Also, because a large proportion of causes of death were missing in CORR, two sensitivity analyses were conducted where all missing causes were attributed to 1) CVD or 2) other causes.
Ethical considerations
Permission was obtained to conduct this study by the Government of Québec ethics committee (Commission d’accès à l’information), CORR internal review committee, and Maisonneuve-Rosemont Hospital ethics committee. Informed consent was waived.