OBJECTIVES: To compare the risk of death between peritoneal dialysis (PD) and hemodialysis (HD) patients. METHODS: Through the national healthcare insurance (NHI) database, we identified a national cohort of 66,080 patients initiating dialysis therapy between January 1, 1998 and December 31, 2006 and followed a maximum of nine years. To compare the survival functions between PD and HD patients, the Kaplan-Meier life table was applied. Both proportional and non-proportional Cox regression models were conducted to evaluate the relative hazard of death by dialysis modality using the intention-to-treat (ITT) approach. Three propensity score strategies were applied to achieve covariate balance, matching, stratification, and regression adjustment. Subsets analyses were defined by age and diabetes mellitus (DM), and sensitivity analyses were conducted by the logit propensity score. RESULTS: The results showed that the hazard ratios (HRs) of PD and HD patients varied by age and with/without DM. Among patients under age 49 with or without DM, PD was associated with a lower risk of death. Among patients aged 50-59, the HR of PD relative to HD was higher for those with DM but was lower for those without DM. For patients aged over 60 with or without DM, the HRs were higher on PD than on HD. We also found that the HRs for PD and HD were not proportional over time. The risk of death for PD patients was generally lower during the first year or first two years after the onset of dialysis. Thereafter, the risk of death increased on PD patients. Results were consistent in different propensity score models and in sensitivity analyses.   CONCLUSIONS: In conclusion, there was an initial survival advantage of PD compared with HD among younger or non-DM patients. As the increase in age, with the presence of DM, and vintage, this relative survival advantage vanished, and even reversed.