Mode of Dialysis Therapy and Mortality in End-Stage Renal Disease ROBERT N. FOLEY,* PATRICK S. PARFREY,* JOHN D. HARNETT,* GLORIA M. KENT,* REGAN O’DEA,* DAVID C. MURRAY,t and PAUL E. BARRE *Di,ision of Nephrologv, The Health Sciences Centre, Memorial University, St. John ‘s, Newfoundland, Canada; Division of Nephrology, Salvation Army Grace General Hospital, St. John ‘s, Netfoundland, Canada; and Division of Nephrologv, Royal Victoria Hospital, McGill University, Montreal, Quebec, Canada. Abstract. Despite considerable differences in technique and blood purification characteristics, hemodialysis and peritoneal dialysis have been thought to have similar patient outcomes. An inception cohort of433 end-stage renal disease patients was followed prospectively for a mean of 4 1 mo. The outcomes of hemodialysis (HD) and peritoneal dialysis (PD) patients were compared using intention to treat analysis based on the mode of therapy at 3 mo. After adjustment for PD patients less likely to have chronic hypertension and more likely to have diabetes, ischemic heart disease, and cardiac failure at baseline (P < 0.05), a biphasic mortality pattern was observed. For the first 2 yr, there was no statistically significant difference in mortality. After 2 yr, mortality was greater among PD patients with an adjusted PD/HD hazard ratio of 1 .57 (95% confidence interval [CI], 0.97 to 2.53). Both the occurrence (adjusted hazards ratio 6.87 [95% CI, 2.01 to 23.5]) and the direction (toward PD, adjusted hazards ratio 6.25 [95% CI, 1 .54 to 25]) of a therapy switch were subsequently associated with mortality after 2 yr. Progressive clinical and echocardiographic cardiac disease were not responsible for this late mortality. Lower mean serum albumin levels in PD patients in the first 2 yr of therapy (3.5 ± 0.5 versus 3.9 ± 0.5 g/dl, P < 0.0001) accounted for a large proportion of the increase in subsequent mortality. Hemodial- ysis has a late survival advantage over peritoneal dialysis; antecedent hypoalbuminemia is a major marker of the in- creased late mortality in PD patients. (J Am Soc Nephrol 9: 267-276, 1998) The advent of peritoneal dialysis was a major addition to the therapeutic armamentarium available to treat end-stage renal disease (ESRD) ( 1-3). Peritoneal dialysis and hemodialysis are very different in terms of dialysis technique. Peritoneal dialysis is associated with a lower overall clearance of traditional markers of solute removal, such as urea and creatinine; how- ever, clearance with standard peritoneal dialysis is continuous, as opposed to markedly intermittent for hemodialysis (4). Peritoneal dialysis is associated with a slower loss of endoge- nous renal function than hemodialysis (5). As in hemodialysis patients, uremic solute clearance has recently been shown to have a considerable impact on patient outcome in a large prospective study of peritoneal dialysis patients (6), as is the case in hemodialysis patients (7-12). It has been suggested that peak levels of uremic solutes, rather than the time-averaged levels, determine toxicity. Using urea as a marker for other toxins, it has been suggested that the failure to show mortality differences between hemodialysis and penitoneal dialysis re- Received March 13, 1997. Accepted July 20, 1997. Dr. Foley was the 1992-1994 Baxter/Canadian Society of Nephrology/Kidney Foundation of Canada Research Fellow. Correspondence to Dr. Robert N. Foley, Memorial University of Newfound- land, The Health Sciences Centre, St. John’s, Newfoundland, Canada A1B 3V6. 1046-6673/0902-0267$03.00/0 Journal of the American Society of Nephrology Copyright 0 1998 by the American Society of Nephrology fleets the fact that the prehemodialysis urea levels are similar to the relatively constant urea levels of peritoneal dialysis ( 13). There are major technical and metabolic differences between hemodialysis and peritoneab dialysis. Both earlier and more recent studies have shown inconsistent results for comparative mortality ( 14-27). For example, a multicenter study from Italy showed a lower mortality among older patients treated with peritoneal dialysis (22). On the other hand. recent large epide- miologic studies from the United States have shown an excess mortality among older diabetic patients treated with peritoneal dialysis (23), and among peritoneal dialysis patients in general (24,25) compared with their hemodialysis counterparts. Simi- larly, registry data from Australia and New Zealand (26) sug- gest that peritoneal dialysis patients have higher mortality. In contrast, a recent report from the Canadian Organ Replacement Registry suggests that peritoneal dialysis confers a survival advantage (27). All of these studies are observational, and therefore inconsistent results may be due to unavoidable and varying selection biases that are seen in nonrandomized stud- ies. The logistic and ethical barriers to performing a randomized trial to determine whether the treatments differ in patient outcome are many. Consequently, longitudinal epidemiologic studies are necessary. There are several methodological issues to consider if epidemiologic studies are to be used to help us answer this question: (1) potential imbalances in baseline age and comorbidity in groups treated by hemodialysis or penito-