Qun Miao†, Ying Jiang*, Miu Y. Lam†, Anne-Marie Ugnat*, David Holland‡

* Public Health Agency of Canada, Ottawa, Ontario, Canada

† Department of Community Health and Epidemiology, Queens’ University, Kingston, Ontario, Canada

‡Department of Medicine, Queens’ University, Kingston, Ontario, Canada

Qun Miao
10 Stuart Street, 2nd level
Division of Cancer Care Epidemiology
Cancer Research Institute at Queen's University
Queen’s University
Kingston, Ontario K7L 3N6

Background: Inconsistency related to of mortality between hemodialysis and peritoneal dialysis among end stage renal disease (ESRD) patients has been reported. Our study was conducted to compare the mortality of ESRD patients between these two predominant dialysis modalities.

Methods: This is a nationwide population-based, retrospective cohort study with 38,842 ESRD patients from the Canadian Organ Replacement Register (CORR) between 1981 and 1998. Deaths were identified through a record linkage to the Canadian Mortality Database. Survival analysis, including Kaplan-Meier estimates, log-rank test and Cox multivariate regression models was applied to compare the mortality of ESRD patients undergoing hemodialysis (HD) and peritoneal dialysis (CAPD/CCPD). Two analytical strategies (“intent-to-treat” and “as-treated”) were used to deal with the problem of changes in treatment modality.

Results: The 5-year survival rates for HD vs CAPD/CCPD patients were 0.43 vs. 0.40 in the as-treated (AT) analysis, and 0.42 vs. 0.38 in the intent-to-treat (ITT) analysis. In multivariate Cox models, the overall hazard ratio for HD vs. CAPD/CCPD was 1.10 (95% CI: 1.04 to 1.18) using the AT method and 0.99 (95% CI: 0.94 to 1.04) using the ITT method. The hazard ratios for HD vs. CAPD/CCPD during the first 2, 2-4, and after 4 years of dialysis were respectively 1.22 (95% CI: 1.15 to 1.32), 0.79 (95% CI: 0.68 to 0.91) and 0.76 (95% CI: 0.62 to 0.93) with AT, and 1.08 (95% CI: 1.01 to 1.14), 0.82 (95% CI: 0.73 to 0.92) and 0.83 (95% CI: 0.73 to 0.95) with ITT.

Conclusion: The difference of mortality between HD and CAPD/CCPD varied by follow-up periods. The mortality on CAPD/CCPD was lower than HD during the initial period but worsened over the long-term.

Introduction

Among end stage renal disease (ESRD) patients, transplantation is the preferred treatment modality, consistently demonstrating greater survival probability compared to dialysis (1,2). However, organ donation rates in Canada are poor, well below those of most developed countries (3). As a result, most Canadian ESRD patients frequently have to choose dialysis, at least as the initial form of treatment, to cope with the problem of a limited supply of kidneys and a long waiting list for transplantation (3).

The two predominant dialytic modalities are hemodialysis (HD) and continuous ambulatory or continuous cycling peritoneal dialysis (CAPD/CCPD), both of which are imperfect (1-4). There is no consensus about which form of dialysis offers patients the best chances for survival (4-7). Previous studies have yielded conflicting results due to patient selection bias and/or inappropriate statistical methods (4-7). In addition, most relevant studies were from the United States and other countries. Health care in Canada is publicly funded and universally accessible to all Canadians under the Canada Health Act. Reports from other countries are of limited value, due to international differences in the structure and database of the registries, the genetic background and demography of the populations studied, and methods of health care delivery (8).

This study was conducted to compare the mortality of Canadian patients with ESRD undergoing HD and/or CAPD/CCPD. Two analytical strategies (“intent-to-treat” (ITT); and “as-treated” (AT)) (8) were introduced to deal with the problem of treatment modality switches. In multivariate survival analysis, time-dependent variables were introduced to compare survival differences at different follow-up periods while controlling for patient-related and disease-related factors.

Study design and methods

This is a nationwide population-based, retrospective cohort study. The study population was all ESRD patients who initiated their treatment and were captured by the Canadian Organ Replacement Register (CORR) between January 1, 1981 and December 31, 1998. Patients were excluded if they were recorded as: a) having recovered function of their own kidney; b) only having received organ transplant other than kidney or dubious renal transplant; c) before 1981 or after 1998; or d) having multiple primary diagnoses. In total, 38,842 ESRD patients were identified from the CORR dataset. The mortality experience of the cohort patients was determined by linking with the Canadian Mortality Database (CMDB) using a probabilistic linkage procedure referred to the Generalized Record Linkage System (GRLS) (9). The CMDB, maintained by Statistics Canada, contains death data for all Canadian residents from 1950 onwards (9).

In both AT and ITT analyses, the end point was defined as the time from enrollment (treatment initiation) to: a) transplantation; b) loss to follow-up; c) death; or d) December 31, 1998, whichever occurred first. However, in AT analysis, patients were censored at day 60 after the first transfer from one dialysis modality to another (8, 11). If patients died within the 60-day period, the deaths were still allocated to the original dialysis modality with the assumption that deaths occurring during this short period were likely associated with the modality of the preceding treatment episode (10, 11). In contrast, in ITT analysis, the modality switches were ignored. All registered deaths during the follow-up period were allocated to the treatment modality at 90 days from the start of relevant treatment (10,11). The rationale for the 90-day rule is that some patients whose final modality was peritoneal dialysis had been placed temporarily on HD until they had completed peritoneal training and catheter placement had been completed (11).

The Kaplan-Meier method and log-rank test were performed to estimate the survival rates among HD and CAPD/CCPD patients without controlling potential confounders. Further, Cox multivariate regression models were applied to compare the mortality of ESRD patients undergoing these two dialytic modalities using AT and ITT to deal with treatment modality switches. Data from 1988 to 1998 were used due to its validity and reliability on relevant variables. Variables included in the models were: treatment, age, sex, primary diagnosis, predialysis comorbid conditions and year-at-initial-treatment. The diabetic status came as the primary kidney disease or as a comorbid condition. The year-at-initial-treatment was grouped as before or after 1990. Comorbid illnesses included cardiovascular conditions (angina, acute myocardial infarction, pulmonary edema, cerebrovascular accident and peripheral vascular disease), chronic obstructive lung disease, malignancies, and any other serious illnesses, which may greatly shorten life expectancy (12). The proportional hazards assumption was checked by graphing the log-log survival curves. It was found that this assumption was questionable for patients in the 0-14 age group, so Cox models were applied to the data excluding this age group. With regard to the treatment effect, the log-log survival curves indicated that treatment effects varied over time. Thus, Cox models were extended to incorporate this time-varying treatment effect defined by the following three time-treatment interaction variables: 1) treatment*(time≤2), 2) treatment*(2In addition, separate models were fitted according to patients’ diabetic status due to the interaction between diabetes and treatment effects (13). All analyses were performed with SAS software, version 8.2(SAS Institute, Cary, NC).

Results

Patients and treatment characteristics

Among 38,842 identified ESRD patients, 24,814 patients started on HD, of whom 14,984 patients had only HD therapy and did not switch to any other treatments. Slightly over one-third of all patients (13,187) began with peritoneal dialysis (including CAPD/CCPD, IPD and NIPD), and 6,455 of them had only peritoneal therapy without any other treatment modalities. The initial treatment distribution is shown in Table 1 and the actual treatment distribution is shown in Figure 1.

Near half of the patients (19,142) died by December 31, 1998. Cardiovascular disease (44.9%, 44.5%, 45.4%, and 35.7% in all, HD, CAPD/CCPD, and renal transplantation patients, respectively) and infectious disease (10.3%, 9.7%, 11.1%, and 20.4%, respectively) were the first two leading causes of death.

Survival Distributions

The 5-year survival rates for HD vs. CAPD/CCPD patients were 0.43 vs. 0.40 in the AT analysis, and 0.42 vs. 0.38 in the ITT analysis (Table 2). In both AT and ITT analysis, the survival rate for patients treated with CAPD/CCPD seemed higher than that of HD during the first two or three years. After that, the survival for HD improved (Figures 2 and 3).

Multivariate Cox Regression Models

Results from both the AT and ITT methods in Cox regression models suggested that the mortality for CAPD/CCPD was lower than HD during the initial 2 years of follow-up, but increased after 2 years. When the time-dependent variable was excluded in the Cox models, the overall survival of patients on HD was higher than that of CAPD/CCPD in the AT method, but no statistical difference was found in the ITT method (Table 3). Further, in both methods, nonwhite and younger patients had better survival. Compared to patients with the primary diagnosis of glomerulonephritis, the survival rates were significantly higher among those with polycystic disease and significantly lower among those with renal vascular disease and diabetes. Survival was worse among patients with co-morbidity before dialysis. In addition, survival has improved during the more recent time period (1990-1998) of initial treatment compared to that before 1990.

Table 4 includes the morality hazard ratios (HD vs CAPD/CCPD) stratified by diabetic status of ESRD patients with dialysis treatment. In the subgroup of non-diabetic patients, the results were similar to those in the whole study population except that no statistically significant difference in survival was found between HD and CAPD/CCPD during the first two years when analyzed using the ITT method. For the subgroup of diabetics, survival ratios of HD vs. CAPD/CCPD were smaller in most subgroups than those in the subgroup of non-diabetics and the whole ESRD patients, although all hazard ratios did not reach statistically significance.

Discussion

There have been a number of studies that examine the issue of differences in mortality between HD and CAPD/CCPD by using ITT and/or AT strategies (6, 10, 11, 13). Our results are similar to those published elsewhere. Murphy et al. conducted a multi-center prospective study in Canada (N=822, 1993-1994). There was no significant difference in overall mortality between HD and CAPD/CCPD by using the ITT method in both Cox proportional hazards regression and Poisson regression models (14). Similarly, Fenton et al. found no overall survival advantage for either group when similar methods and models were applied to analyze the 1990-1994 CORR data (6).

We found that the hazard risk was varied by time in Canadian patients, which is consistent with other studies (6, 10). Fenton et al. reported that the increased mortality on hemodialysis compared with CAPD/CCPD in the first 2 years of follow-up when Poisson regression was applied to 1990-1994 CORR data (10). A study from the Netherlands yielded a similar result with us that the survival of CAPD/CCPD was better in the early stage of follow-up although a statistical significance was not reached and survival was lower than that of HD over the long-term (11).

The reported discrepancy between AT and ITT analysis was in accordance with other studies (10, 11, 13). In ITT analysis, any shift to another modality after 90 days was overlooked, thus the effects of this method were in fact “pooled” (10, 11, 13, 14). If a true difference in mortality rates exists, the estimated relative risk of two modalities in ITT analysis would be attenuated due to the misclassification of patients by modality (11, 13,14). On the other hand, AT analysis accounts for modality switches and allows one to compare purely HD to purely PD mortality rates (10). Some researchers may argue that if a patient undergoing peritoneal dialysis with severe fungal peritonitis is placed on HD and then dies, ignoring this in AT analysis can produce a biased comparison (11). For most situations, the bias can be avoided by simply assigning a transition period, so that death is counted against the initial treatment rather than the new modality (11, 13). However, a study mentioned that 60 days adjustment still tended to overestimate survival for the earlier treatment because of the higher incidence of deaths during several months after the transfer (11). Nevertheless, this bias was minor and can be neglected because we achieved the similar result that the mortality on CAPD/CCPD was better than HD during the initial 2 years of follow-up in both AT and ITT methods.

In our study, approximately 21% of all non-transplant patients switched from one modality of dialysis to another at least once. Mostly, patients transferred from CAPD/CCPD to HD because of clinical and technical problems (11). Occasionally, patients transferred from HD to CAPD/CCPD due to difficulties with vascular access (12). The high rate of switching indicated that an integrated program is necessary for large groups of patients (11).

Our findings on the effects of other risk factors for ESRD patients were the same as those in most other studies (6, 7, 13, 14). Survival for patients whose treatment was initiated after 1990 was better than that for those treated before 1990. This reflects the introduction of new technology and methods in clinical practice in the 1990s (7, 13 and 14).

There are several strengths to this study. This is a nationwide, population-based cohort study with over 20,000 patients and long-term follow up. It provides an opportunity to more precisely estimate survival among patients with different dialyses. AT and ITT methods were introduced to deal with the problem of treatment switches in ESRD patients. Also the time dependent variables defined by graph method were added to the final multivariate Cox models to measure the difference on survival between HD and CAPD/CCPD over various follow-up periods. On the other hand, due to lack of information related to patients’ residual kidney function, nutrition status, biochemical parameters, and severity of each co-morbid in the CORR dataset, this study could not determine the relationship between these potential risk factors and survival. In addition, there is a potential for selection bias because PD patients tended to be healthier than the HD patients. However, our finding was consistent with other studies, in which all these confounders were controlled (7, 11). It is not clear if patient compliance influenced our results. Kutner mentioned that noncompliance with prescribed therapy had a significant impact on dialysis patient care and outcome (16). Thus, this potential confounder should be considered in future studies.

In summary, this study provides detailed analyses comparing the mortality of Canadian patients with ESRD undergoing HD vs. CAPD/CCPD. The AT and ITT strategies were performed to address the issue of treatment switch in survival analysis. In this study, the difference of mortality between HD and CAPD/CCPD varied by follow-up periods. The survival for CAPD/CCPD was higher than HD during the initial period but became lower in the long-term follow-up. Many factors may influence patients’ survival. The study suggests that it may be better to choose CAPD/CCPD for new ESRD patients and HD for chronic patients in clinical practice.

ACKNOWLEFGMENTS

Staff from Public Health Agency of Canada, Statistics Canada, Canadian Organ Replacement Register (CORR) and Canadian Health Information Institute.

References

(1) Mallick NP, Jones E, Selwood N. The European (European Dialysis and Transplantation Association-European Renal Association) Registry. American Journal of Kidney Diseases 1995; 25(1):176-187.

(2) Uribarri J. Past, present and future of end-stage renal disease therapy in the United States. Mount Sinai Journal of Medicine 1999; 66(1):14-19.

(3) Mao Y. Mortality and Cancer Incidence Among Organ Failure Patients (Proposal). 2002.

(4) Bloembergen WE, Port FK, Mauger EA, Wolfe RA. A comparison of mortality between patients treated with hemodialysis and peritoneal dialysis. Journal of the American Society of Nephrology 1995; 6(2):177-183.

(5) Collins AJ, Hao W, Xia H, Ebben JP, Everson SE, Constantini EG et al. Mortality risks of peritoneal dialysis and hemodialysis. American Journal of Kidney Diseases 1999; 34(6):1065-1074.

(6) Fenton SS, Schaubel DE, Desmeules M, Morrison HI, Mao Y, Copleston P et al. Hemodialysis versus peritoneal dialysis: a comparison of adjusted mortality rates. American Journal of Kidney Diseases 1997; 30(3):334-342.

(7) Schaubel DE, Morrison HI, Fenton SS. Comparing mortality rates on CAPD/CCPD and hemodialysis. The Canadian experience: fact or fiction?[see comment]. Peritoneal Dialysis International 1998; 18(5):478-484.

(8) D'Amico G. Comparability of the different registries on renal replacement therapy. American Journal of Kidney Diseases 1995; 25(1):113-118.

(9) Fair M. Generalized Record Linkage System – Statistics Canada’s Record Linkage Software. Austrian Journal of Statistics Volume 33 (2004), Number 1&2, 37-53

(10) Vonesh E, Schaubel DE, Hao W, Collins AJ. Statistical methods for comparing mortality among ESRD patients: Examples of regional / international variations. Kidney International - Supplement 2000; 57(Suppl. 74):S19-S27.

(11) Termorshuizen F, Korevaar JC, Dekker FW, Van Manen JG, Boeschoten EW, Krediet RT et al. Hemodialysis and peritoneal dialysis: comparison of adjusted mortality rates according to the duration of dialysis: analysis of The Netherlands Cooperative Study on the Adequacy of Dialysis 2. Journal of the American Society of Nephrology 2003; 14(11):2851-2860.

(12) Alloatti S, Manes M, Paternoster G, Gaiter AM, Molino A, Rosati C. Peritoneal dialysis compared with hemodialysis in the treatment of end-stage renal disease. Journal of Nephrology 2000; 13(5):331-342.

(12) Murphy SW, Foley RN, Barrett BJ, Kent GM, Morgan J, Barre P et al. Comparative mortality of hemodialysis and peritoneal dialysis in Canada. Kidney International 2000; 57(4):1720-1726.

(14) Silins J, Fortier L, Mao Y, Posen G, Ugnat AM, Brancker A et al. Mortality rates among patients with end-stage renal disease in Canada, 1981-86. CMAJ Canadian Medical Association Journal 1989; 141(7):677-682.

(15) Reikes ST. Trends in end-stage renal disease. Epidemiology, morbidity, and mortality. Postgraduate Medicine 2000; 108(1):124-126.

(16) Kutner NG. Improving compliance in dialysis patients: does anything work? Seminars in Dialysis 2001; 14(5):324-327.