BACKGROUND: Patients initiating with peritoneal dialysis (PD) have favorable clinical conditions compared with hemodialysis (HD) patients, which may contribute to the varying results found in studies of mortality across the two therapies. METHODS: National incidence data of end-stage renal disease patients from 1995 to 1997 were used, excluding the first 90 days of treatment and including all patients who were on either PD or HD on day 91. Patients were then followed for a one-year period. A Cox proportional hazards regression analysis was used, separating diabetics and non-diabetics, and two statistical models were applied. Model 1 included race, gender, age, initial modality, and incidence year as explanatory variables. Model 2 added body mass index (BMI), initial levels of serum albumin, creatinine, and blood urea nitrogen. RESULTS: Age was most highly associated with mortality, followed by biochemical variables, BMI, gender, and dialysis modality. In diabetics, the hazard ratio (HR) from Model 1 indicated no difference [1.046, 95% confidence limits (CL) 0.989-1.105; P> 0.1, HD was the reference] in mortality between PD and HD, while Model 2 demonstrated that PD patients had a 13.4% (1.134, CL 1.072-1.100, P < 0.0001) higher chance of death. In non-diabetics, hazard ratios (HRs) from Models 1 and 2 indicated that PD patients had a 23.5% (0.765, 0.722-0.812, P < 0.0001) and 11.9% (0.881, 0.30-0.935, P < 0.0001), respectively, lower likelihood of death than HD patients. CONCLUSION: Our study indicates that the results changed depending on the analytical methods used. We recommend that, due to the unequally distributed clinical conditions of patients at initiation, comparisons of mortality outcomes between dialysis modalities should be made with caution.
BACKGROUND:Patients initiating with peritoneal dialysis (PD) have favorable clinical conditions compared with hemodialysis (HD) patients, which may contribute to the varying results found in studies of mortality across the two therapies. METHODS: National incidence data of end-stage renal diseasepatients from 1995 to 1997 were used, excluding the first 90 days of treatment and including all patients who were on either PD or HD on day 91. Patients were then followed for a one-year period. A Cox proportional hazards regression analysis was used, separating diabetics and non-diabetics, and two statistical models were applied. Model 1 included race, gender, age, initial modality, and incidence year as explanatory variables. Model 2 added body mass index (BMI), initial levels of serum albumin, creatinine, and blood ureanitrogen. RESULTS: Age was most highly associated with mortality, followed by biochemical variables, BMI, gender, and dialysis modality. In diabetics, the hazard ratio (HR) from Model 1 indicated no difference [1.046, 95% confidence limits (CL) 0.989-1.105; P> 0.1, HD was the reference] in mortality between PD and HD, while Model 2 demonstrated that PDpatients had a 13.4% (1.134, CL 1.072-1.100, P < 0.0001) higher chance of death. In non-diabetics, hazard ratios (HRs) from Models 1 and 2 indicated that PDpatients had a 23.5% (0.765, 0.722-0.812, P < 0.0001) and 11.9% (0.881, 0.30-0.935, P < 0.0001), respectively, lower likelihood of death than HDpatients. CONCLUSION: Our study indicates that the results changed depending on the analytical methods used. We recommend that, due to the unequally distributed clinical conditions of patients at initiation, comparisons of mortality outcomes between dialysis modalities should be made with caution.