Comparing mortality of peritoneal and hemodialysis patients in the first 2 years of dialysis therapy: a marginal structural model analysis

Abstract

Background and objectives: There are conflicting research results about the survival differences between hemodialysis and peritoneal dialysis, especially during the first 2 years of dialysis treatment. Given the challenges of conducting randomized trials, differential rates of modality switch and transplantation, and time-varying confounding in cohort data during the first years of dialysis treatment, use of novel analytical techniques in observational cohorts can help examine the peritoneal dialysis versus hemodialysis survival discrepancy.

Design, setting, participants, & measurements: This study examined a cohort of incident dialysis patients who initiated dialysis in DaVita dialysis facilities between July of 2001 and June of 2004 and were followed for 24 months. This study used the causal modeling technique of marginal structural models to examine the survival differences between peritoneal dialysis and hemodialysis over the first 24 months, accounting for modality change, differential transplantation rates, and detailed time-varying laboratory measurements.

Results: On dialysis treatment day 90, there were 23,718 incident dialysis-22,360 hemodialysis and 1,358 peritoneal dialysis-patients. Incident peritoneal dialysis patients were younger, had fewer comorbidities, and were nine and three times more likely to switch dialysis modality and receive kidney transplantation over the 2-year period, respectively, compared with hemodialysis patients. In marginal structural models analyses, peritoneal dialysis was associated with persistently greater survival independent of the known confounders, including dialysis modality switch and transplant censorship (i.e., death hazard ratio of 0.52 [95% confidence limit 0.34-0.80]).

Conclusions: Peritoneal dialysis seems to be associated with 48% lower mortality than hemodialysis over the first 2 years of dialysis therapy independent of modality switches or differential transplantation rates.

Figure 1.

Kaplan–Meier survival curves adjusted for age, sex, race, and diabetes examining survival among peritoneal dialysis (PD) and hemodialysis (HD) patients (modality is defined on day 90) for incident dialysis patients initiating dialysis from July of 2001 to June of 2004 (n=23,718).

Figure 2.

Associations between dialysis modality (PD versus HD) and mortality among incident dialysis patients who never changed modalities (n=21,762; left) and patients with at least one modality change over 2 years (n=1956; right). Cox adjustment 1, adjusted for inverse probability of treatment weight (IPTW) predictors only: age, sex, diabetes mellitus (DM), and race; Cox adjustment 2, fully adjusted model adjusted for IPTW predictors and additional confounders: marital status, employment, baseline comorbidities (chronic obstructive pulmonary disease [COPD], cancer, hypertension, ability to ambulate, heart failure, and atherosclerotic heart disease), and baseline serum levels of ferritin, calcium, phosphorus, and normalized protein catabolic rate (nPCR).

Figure 3.

Mortality hazard ratios (HRs) for dialysis modality (PD versus HD) in incident dialysis patients using Cox models (n=23,718) (Supplemental Table 2). Cox adjustment 1, adjusted for IPTW predictors only: age, sex, DM, and race; Cox adjustment 2, fully adjusted model adjusted for IPTW predictors and additional confounders: marital status, employment, baseline comorbidities (COPD, cancer, hypertension, ability to ambulate, heart failure, and atherosclerotic heart disease), and baseline serum levels of ferritin, calcium, phosphorus, and nPCR.

Figure 4.

Mortality HRs for dialysis modality (PD versus HD) in 23,718 incident dialysis patients using a marginal structural model (MSM) taking into account changes in dialysis modality and transplant censorship in the first 2 years (Supplemental Table 3). Model 1, inverse probability weight (IPW): age, baseline modality, and time-dependent modality (TD; stabilized IPW mean=1.06); model 2, IPW: age, sex, race, DM, baseline modality, and TD (stabilized IPW mean=1.06); model 3, IPW: age, sex, race, DM, baseline modality (Alb and Hgb), and TD modality (Alb and Hgb; stabilized IPW mean=1.29); MSM adjustment 1, IPW-adjusted (IPWs were calculated using different sets of variables); MSM adjustment 2, IPW + adding the same variables used to calculate IPWs to control for residual confounding; MSM adjustment 3, same as MSM adjustment 2 + additional confounders: marital status, employment, baseline comorbidities (COPD, cancer, hypertension, ability to ambulate, heart failure, and atherosclerotic heart disease), and baseline serum levels of ferritin, calcium, phosphorus, and nPCR.

Figure 5.

Mortality HRs for dialysis modality (PD versus HD) in 23,718 incident dialysis patients using an MSM taking into account changes in dialysis modality and transplant censorship in the first 2 years stratifying on diabetes status (P value for the interaction=0.07) and age (P value for the interaction=0.26) (Supplemental Table 4). Model 2 IPTWs were used for all stratified models. MSM adjustment 1, IPW-adjusted (IPWs were calculated using different sets of variables); MSM adjustment 2, IPW + adding the same variables used to calculate IPWs to control for residual confounding; MSM adjustment 3, same as MSM adjustment 2 + additional confounders: marital status, employment, baseline comorbidities (COPD, cancer, hypertension, ability to ambulate heart failure, and atherosclerotic heart disease), and baseline serum levels of ferritin, calcium, phosphorus, and nPCR.