Survival of propensity matched incident peritoneal and hemodialysis patients in a United States health care system

Transcription

1 & 2014 International Society of Nephrology see commentary on page 884 of propensity matched incident peritoneal and hemodialysis patients in a United States health care system Victoria A. Kumar 1, Margo A. Sidell 2, Jason P. Jones 2 and Edward F. Vonesh 3 1 Division of Nephrology, Department of Internal Medicine, Southern California Permanente Medical Group, Los Angeles, California, USA; 2 Research and Evaluation, Southern California Permanente Medical Group, Pasadena, California, USA and 3 Department of Preventive Medicine, Feinberg School of Medicine, Northwestern University, Chicago, Illinois, USA We sought to compare survival among incident peritoneal dialysis (PD) patients to matched hemodialysis (HD) patients who received pre-dialysis care, including permanent dialysis access placement. Patients starting PD were propensity matched to those starting HD. HD patients who used a central venous catheter during the first 90 days of dialysis were excluded. Stratified Cox proportional hazards models were used to compare patient survival using both intent-totreat and as-treated analyses. In the intent-to-treat analysis, patients were followed from the date of first dialysis until death and censored at the earliest of the following: renal transplantation, death, renal recovery, loss to follow-up or study end. In the as-treated analysis, patients were also censored at the time of modality change. A total of 1003 matched pairs were obtained from 11,301 incident patients (10,298 HD and 1003 PD). The cumulative hazard ratio for death at one year was 2.38 (95% CI ) and 2.10 ( ) for HD relative to PD patients in the as-treated and intent-to-treat analyses, respectively. The cumulative risk of death, as estimated by the cumulative hazard ratio, favored PD for almost up to 3 years of follow-up in the as-treated analysis and nearly 2 years of follow-up in the intent-to-treat analysis with no differences thereafter. The higher adjusted rate of death observed for HD patients cannot be attributed to initial use of central venous catheters or lack of pre-dialysis care. Kidney International (2014) 86, ; doi: /ki ; published online 2 July 2014 KEYWORDS: ESRD; hemodialysis; mortality risk; peritoneal dialysis; survival Correspondence: Victoria A. Kumar, Department of Internal Medicine, Division of Nephrology, Southern California Permanente Medical Group, 4700 Sunset Blvd., Los Angeles, California 90027, USA. Victoria.a.kumar@kp.org Received 1 October 2013; revised 29 April 2014; accepted 1 May 2014; published online 2 July 2014 The impact of dialysis modality on patient survival has been debated over the previous two decades. Several studies from the 1990s demonstrated a lower, comparable, or even higher relative risk of mortality for peritoneal dialysis (PD) patients compared with similar hemodialysis (HD) patients in the US and Canada. 1 3 Differences in survival among large-scale studies such as these have been attributed to the degree of case-mix adjustment performed and the use of different subgroups. 4 Moreover, results of survival studies have been shown to vary depending upon the analytical methods used. 5,6 A time-dependent trend in death risk has been consistently demonstrated by multiple studies, with PD patients experiencing better survival during the first 1 2 years of dialysis. 1 3 More recent studies have attempted to address the issue of case-mix and have detected no survival differences among incident PD and HD patients. 7 9 Favorable clinical conditions among patients who initiate PD likely affect the results of mortality studies. 5,7 Patients who initiate HD using a central venous catheter have a higher risk of death compared with those who are started with an arterio venous fistula or graft A recent analysis of Canadian registry data found that 1-year mortality was similar for PD patients compared with HD patients who initiate dialysis using an arterio venous fistula or graft. 13 The authors concluded that the use of central venous catheters in incident HD patients largely accounts for the early survival benefit seen in PD patients. Other variables that could potentially confound survival analyses include body mass index, residual renal function, and pre-dialysis care. 20 The majority of survival studies have been performed using large end-stage renal disease (ESRD) registries in the United States and Canada. The United States Renal Data System (USRDS) database is based on Medicare claims data and allows for large-scale study of a heterogeneous population, but under-reporting of patient-specific data including comorbid conditions could potentially bias study results. 21 We have previously reported superior outcomes for PD patients at our organization, which uses an accountable-care organization model to deliver standardized quality care to 1016 Kidney International (2014) 86,

2 VA Kumar et al.: in US peritoneal and hemodialysis patients clinical investigation patients. 22 We aimed to use our ESRD registry to analyze survival among incident PD patients over the previous decade compared with matched, incident HD patients from the same time period. We excluded HD patients who initiated dialysis urgently with a catheter in an effort to reduce case-mix bias. We also attempted to control for differences in neighborhood socioeconomic status by use of our geocoding database. RESULTS A total of 11,301 patients (10,298 HD and 1003 PD) initiated dialysis during the study period and met the inclusion criteria. The final analysis included 1003 matched pairs of patients. Case-mix differences in patient characteristics at initiation of dialysis are listed in Table 1 for all incident PD and HD patients. Baseline patient characteristics for propensity score-matched PD and HD patients (1003 pairs) are listed in Table 2 for comparison. A total of 67 patients (16 PD, 51 HD) died during the first 90 days after initiation of dialysis, although none were excluded from the analysis. Shown in Figure 1 and Table 3 are population-averaged adjusted patient survival curves from a stratified Cox proportional hazards model for matched PD and HD patients through 9 years of follow-up in the as-treated and intent-to-treat analyses, respectively. 23 The corresponding adjusted median survival times were 8.2 years for PD patients and 7.9 years for HD patients in the as-treated analysis and 7.7 years for PD patients and 7.9 years for HD patients in the intent-to-treat analysis. The propensity score-matched 1- and 2-year adjusted patient survival was, respectively, 95 and 87% for PD patients vs. 89 and 83% for HD patients in the Table 1 Baseline patient demographics for all incident HD and PD patients HD (n ¼ 10,298) PD (n ¼ 1003) P Standardized differences a Mean age in years 65.2± ±14.2 o Males (%) 5988 (58.1) 543 (54.1) Mean CCI 4.4± ±1.6 o Race (% PD) o African American 2302 (22.4) 143 (14.3) Asian 981 (9.5) 142 (14.2) Hispanic 2890 (28.1) 272 (27.1) Other/unknown 502 (4.9) 135 (13.5) White 3623 (35.2) 311 (31.0) Cause of ESRD (%) o Cystic kidney disease 200 (1.9) 55 (5.5) Glomerulonephritis 730 (7.1) 156 (15.6) Hypertension 2176 (21.1) 184 (18.3) Diabetes mellitus 5418 (52.6) 486 (48.5) Other 1428 (13.9) 70 (7.0) Other urologic 148 (1.4) 13 (1.3) Unknown/missing 198 (1.9) 39 (3.9) Abbreviations: CCI, Charlson comorbidity index; ESRD, end-stage renal disease; HD, hemodialysis; PD, peritoneal dialysis. a Standardized difference ¼ difference in means or proportions divided by standard error; imbalance defined as absolute value greater than 0.20 (small effect size). Table 2 Baseline patient demographics for propensity score-matched PD and HD patients (1003 pairs) PD (n ¼ 1003) HD (n ¼ 1003) Standardized differences a Mean age in years 57.4± ± Males (%) 543 (54.1) 543 (54.1) 0 Mean CCI 3.6± ± Race (%) 0 African American 143 (14.3) 143 (14.2) Hispanic 272 (27.1) 272 (27.1) Asian 142 (14.2) 142 (14.2) Other/unknown 135 (13.4) 135 (13.4) White 311 (31.0) 311 (31.0) Cause of ESRD (%) 0 Cystic kidney disease 55 (5.5) 55 (5.5) Glomerulonephritis 156 (15.6) 156 (15.6) Hypertension 184 (18.3) 184 (18.3) Diabetes mellitus 486 (48.5) 486 (48.5) Other 70 (7.0) 70 (7.0) Other urologic 13 (1.3) 13 (1.3) Unknown/missing 39 (3.9) 39 (3.9) Neighborhood family income per year (%) $0-$25, (3.4) 31 (3.2) $25,001 $40, (17.1) 190 (19.4) $40,001 $60, (29.8) 294 (29.9) $60,001 $80, (24.2) 248 (25.3) 4$80, (25.5) 219 (22.3) 0.06 Neighborhood education level b (%) 0.06 o50% with HS education or 138 (13.9) 151 (15.4) higher 50 75% with HS education or 315 (31.8) 325 (33.1) higher 475% with HS education or higher 537 (54.2) 506 (51.5) Abbreviations: CCI, Charlson comorbidity index; ESRD, end-stage renal disease; HS, high school; HD, hemodialysis; PD, peritoneal dialysis. a Standardized difference ¼ difference in means or proportions divided by standard error; imbalance defined as absolute value greater than 0.20 (small effect size). b 21 HD and 13 PD patients did not have geocoded information available for income and education level. Population-averaged adjusted survival As treated PD HD P < Figure 1 Population-averaged adjusted patient survival for propensity score-matched hemodialysis (HD) and peritoneal dialysis (PD) patients by method of analysis. The shaded areas reflect where the point-wise differences in the two survival curves differ (Po0.05). ITT Kidney International (2014) 86,

3 VA Kumar et al.: in US peritoneal and hemodialysis patients Table 3 Adjusted yearly cumulative hazard ratios for comparing survival among propensity score-matched HD and PD patients (1003 pairs) As-treated (AT) Intent-to-treat (ITT) Year %PD %HD Adjusted a time-dependent cumulative hazard ratios (CHR) b SE 95% CI %PD %HD Adjusted a time-dependent cumulative hazard ratios (CHR) b SE 95% CI Abbreviations: CI, confidence interval; HD, hemodialysis; PD, peritoneal dialysis; SE, standard error. a Adjusted for gender, age, race, primary ESRD diagnosis, Charlson comorbidity index, and neighborhood income level. b Time-dependent cumulative hazard ratios calculated using a macro by Ed Vonesh. Cumulative hazard ratio (CHR) As treated CHR(HD:PD) 95% CL Figure 2 Adjusted cumulative hazard ratios for propensity score-matched hemodialysis (HD) and peritoneal dialysis (PD) patients by method of analysis along with 95% confidence intervals. ITT Difference (PD-HD) in PA adjusted survival PD As treated HD 95% CL Figure 3 Point-wise differences in population-averaged adjusted patient survival along with 95% confidence intervals between propensity score-matched hemodialysis (HD) and peritoneal dialysis (PD) patients by method of analysis. ITT as-treated analysis. In the intent-to-treat analysis, the 1- and 2-year adjusted patient survival was 94 and 86% for PD patients vs. 89 and 83% for HD patients, respectively. To account for non-proportional hazards (as reflected by survival curves that cross each other), we estimated the relative risk of death using the cumulative hazard ratio (CHR) from the stratified Cox model adjusted for any residual confounding due to case-mix differences at baseline. 24 As a ratio of baseline cumulative hazard functions, the CHR provides an estimate of the cumulative risk of death for HD relative to PD at any given moment in time. 8,24 Shown in Table 3 and Figure 2 are the adjusted CHRs for death in the propensity score-matched HD and PD patients. The CHR for death after 1 year on dialysis was 2.38 (95% CI , Po0.0001) for HD patients in the as-treated analysis and 2.09 (95% CI , Po0.0001) in the intent-to-treat analysis compared with matched PD patients. At 2 years the CHR was 1.39 (95% CI , P ¼ 0.017) for HD patients in the as-treated analysis and 1.26 (95% CI , P ¼ 0.070) in the intent-to-treat analysis compared with matched PD patients. As reflected in Figure 2, the cumulative risk of death favored PD for nearly 3 years of follow-up in the as-treated analysis and for nearly 2 years of follow-up in the intent-to-treat analysis, with minimal or no significant differences thereafter. This is further reflected in Figure 3 in which the difference in population-averaged adjusted patient survival between PD and HD is plotted along with the corresponding 95% confidence intervals. 23 The ever increasing width of the confidence intervals reflects the uncertainty associated with the reduction in the numbers of patients after 6 years (Table 3) Kidney International (2014) 86,

4 VA Kumar et al.: in US peritoneal and hemodialysis patients clinical investigation DISCUSSION The cumulative risk of death was over two-fold higher over the course of the first year on dialysis among matched incident HD patients compared with incident PD patients in both the adjusted time-dependent as-treated and intent-totreat analyses. The cumulative risk of death tended to favor PD over HD for the first 2 3 years. Specifically, PD was associated with a survival advantage for up to 3 years in the as-treated analysis with no significant difference in adjusted survival thereafter and for up to 2 years in the intent-to-treat analysis with no difference in adjusted survival thereafter (Figures 1 2). Our findings are in contrast to those of Quinn et al., who found no difference in survival between PD and HD patients after adjustment in baseline characteristics. 7 Authors in that study were unable to perform an as-treated analysis because of the inability to reliably capture changes in dialysis modality. Differences in delivery of health care at our organization, as compared with the universal health care system practiced in Ontario, Canada, may explain why PD patients in our study experienced a lower risk of death during the first year on dialysis compared with HD study patients. The results of our analyses demonstrate a statistically significant lower cumulative risk of death associated with PD during the first 2 3 years of dialysis, depending on the method of analysis. In the as-treated analysis, 3 9-year survival was not significantly different between the two modalities (Figures 1 3). Over the same 3 9-year time frame, intent-to-treat survival tended to favor HD over PD. However, except for the narrow window between 4.3 and 4.5 years, neither the CHR nor the point-wise differences in mean survival were statistically significant in the intent-totreat analysis (Figures 1 3). The observed crossing patterns in survival, particularly with the intent-to-treat analysis, may reflect the effect of switching modalities possibly concurrent with the loss of residual renal function over time in the PD cohort, but, unfortunately, we have no data to support this theory. PD may afford a survival advantage to patients with significant residual renal function, but not to patients who lack it. Further insight into the crossing patterns may be gained by examining the cumulative incidence curves associated with death, transplantation, and modality switching in a competing risk-type analysis (Appendix online). Weinhandl et al. reported that cumulative survival probabilities were higher for incident US PD patients 1 and 2 years after dialysis initiation, although the results were no longer significant when the analysis excluded patients who died within 90 days of initiation of dialysis. 25 We did not exclude patients who died within the first 90 days of study because our baseline cohort was designed to exclude highrisk patients who were initiated into dialysis urgently and without adequate pre-dialysis care. Other studies have demonstrated that PD patients experience a lower risk of death in the first 1 2 years after initiation of dialysis. Several explanations have been proposed, including the lack of pre-dialysis care and the use of central venous catheters in HD patients, as well as better preservation of residual renal function in PD patients. Although higher residual renal function during the first few years on PD could explain our results, changes in peritoneal membrane structure over time along with reduced ultrafiltration capacity could explain why PD patients lose their survival advantage after the first few years on dialysis. PD patients tend to be exposed to higher dextrose concentrations and volumes over time, potentially resulting in hypoalbuminemia, hyperglycemia, and changes in peritoneal membrane function. The early survival benefit seen in PD patients could eventually be lost by a decline in nutritional and/or cardiovascular status after the first year or two of PD. Finally, the early survival benefit associated with PD in our study could be attributed to unmeasured confounding factors resulting in selection bias and early mortality of higher risk HD patients. Studies that compare outcomes for PD and HD patients are prone to selection bias due to baseline patient differences. Even when careful matching algorithms are employed, some differences in patient makeup will persist. A future randomized clinical trial to determine the modality that confers better survival to dialysis patients is unlikely, given the cost and feasibility of such a study. A previous attempt to perform such a trial was unsuccessful. 26 The strengths of our study include a large, diverse PD cohort and relatively long follow-up time. Patients appeared well matched in terms of baseline disease burden and demographics. was analyzed on the basis of modality at initiation of dialysis, with death attributed to the modality a patient was on at the time of death (as-treated) or to the modality a patient was started on regardless of whether the patient switched modalities (intent-to-treat), and both methods yielded similar results. Although similar, there are differences in CHRs for death among HD patients in the astreated versus intent-to-treat analysis, and these differences could be reflective of an informative censoring mechanism among those PD patients who were switched to HD early on. Our study utilizes the relatively new and novel CHR to directly compare adjusted PD and HD patient survival in the presence of non-proportional hazards. 8,24 Prior studies have used time-dependent hazard ratios to compare HD and PD outcomes in the presence of non-proportional hazards. 1,3,6,13,16 However, time-dependent hazard ratios do not provide clinicians with a truly prognostic estimate of the cumulative treatment effect of PD vs. HD over time. In contrast, the CHR is a direct function of covariate-specific adjusted patient survival estimates among HD and PD patients. To illustrate this, Figure 4 displays a plot of predicted PD and HD survival curves in our intent-to-treat analysis where survival is adjusted to the average values of the baseline covariates included in the stratified Cox proportional hazards model. Included in Figure 4 is an overlay of the CHR shown in Figure 2 but expressed here in terms of the cumulative risk of death of PD relative to HD. The covariatespecific adjusted survival curves cross each other precisely Kidney International (2014) 86,

5 VA Kumar et al.: in US peritoneal and hemodialysis patients Covariate-specific adjusted survival 1.0 CHR(PD:HD) 95% CL PD HD Figure 4 Overlay of adjusted cumulative hazard ratios (CHRs) and covariate-specific adjusted patient survival among propensity score-matched hemodialysis (HD) and peritoneal dialysis (PD) patients in the intent-to-treat analysis. whenever the CHR(PD:HD) ¼ 1.00 (the horizontal reference line), which in our study occurs at time t ¼ 3.3 years (the vertical reference line). In addition, the CHR provides one with the ability to directly compare adjusted survival curves between PD and HD at select times of interest. A major limitation of our study is that we do not have data regarding residual renal function at the time dialysis was initiated. The rate of decline in residual renal function is generally higher for HD patients than for PD patients, 27,28 although differences may be less pronounced in today s era of biocompatible membranes and angiotensin blockade The authors cannot exclude that baseline residual renal function was higher in PD patients than in matched HD patients, potentially conferring a survival benefit to the PD cohort. We also have limited data on other key baseline laboratory values as recorded in the ESRD registry starting from Care of the PD patient has changed over the previous two decades. As a result, annual peritonitis rates, hospital days, and mortality rates have declined for US PD patients. 32 Since the advent of the Kidney Dialysis Outcomes Quality Initiative (K/DOQI) in the 1990s, more attention has been given to preservation of residual renal function, nutritional status, anemia, and mineral metabolism, all of which may influence mortality Improvements in management of the PD patient and avoidance of metabolic complications may ultimately lead to longer-term differences in survival in the future. MATERIALS AND METHODS Kaiser Permanente and the ESRD registry The Kaiser Permanente Southern California (KPSC) organization and ESRD registry have been previously described by us. 22 Briefly, KPSC is a large, integrated health maintenance organization that utilizes an accountable-care organization model and serves B3.5 2 Cumulative hazard ratio (CHR) million members in Southern California. The ESRD registry includes data for all KPSC ESRD patients, including patient demographics, cause of ESRD, ESRD start date, modality type, vascular access type, and date of death. Data are reported monthly into a central registry by individual PD and HD units within KPSC and by contracted units. Only a small minority of KPSC PD patients are managed by contracted units. Modality changes, modality-specific quality data, and KPSC system membership status are tracked prospectively and reported into the ESRD registry. The ESRD registry was utilized to identify all patients who initiated PD or HD as their first dialysis modality between 1 January 2001 and 30 June 2013 (study period). Inclusion criteria for patients included age 418 years and 1 year of prior KPSC membership at initiation of dialysis (the latter permitted characterization of baseline comorbid conditions). All HD patients included in the study utilized either an arterio venous fistula or a graft during the first 90 days of study. HD patients who utilized a central venous catheter as vascular access at any time during the first 90 days of dialysis were excluded from the final analysis. PD patients who received temporary HD during the first 90 days of the study period were excluded. Charlson comorbidity index The Charlson comorbidity index was calculated for all study patients using baseline medical diagnoses obtained from the KPSC patient database. 39,40 The index assigns weights for a number of major conditions and represents a measure of the burden of comorbid disease. Geocoding database We used geocoding to determine socioeconomic status for our patients on the basis of each patient s neighborhood. Although geocoding cannot provide exact patient characteristics, it is reliable when applied to a group or population. 41,42 Education and income levels were obtained for all PD patients from the KPSC Geocoding database, which contains aggregate neighborhood income and education data for KPSC members. It is based on the US Census, which is performed once every decade. Data were obtained for KPSC members using the census identification at block, block group, tract, county, or ZIP code level, using the most detailed level available (block level being the most detailed level). Home addresses for PD and HD patients were geocoded and linked to the 2000 Census database, which includes neighborhood socioeconomic characteristics. The 2000 and 2010 Census databases contain the highest level of education attained for all individuals in a given census unit who were at least 25 years of age at the time of the census. Family income was available for the census year, defined as income of family members and/or persons related to the householder living within a given geocode. A family consisted of at least two members (one householder and one family member or related individual). The distribution of neighborhood educational attainment and family income was obtained for residents within the census unit based on the patient s home address. Patient matching After identification of all incident PD patients who met the inclusion criteria, we matched each to one of the incident HD patients who met the inclusion criteria using frequency matching with propensity scores. The propensity scores were estimated using 1020 Kidney International (2014) 86,

6 VA Kumar et al.: in US peritoneal and hemodialysis patients clinical investigation the logistic regression model, which included baseline covariates (age, sex, race, primary cause of ESRD, year of initiation on dialysis, and Charlson comorbidity index). The caliper matching method was used with a maximum tolerance level of Statistics Unmatched continuous variables were compared using the Wilcoxon rank-sum test, whereas unmatched categorical variables were compared using Fisher s exact test. In the as-treated analysis, survival was calculated from the date of first dialysis to the earliest of the following: death, modality change, renal transplantation, renal recovery, loss of KPSC coverage, or study end. PD patients who transitioned to HD temporarily were allowed up to 120 days off modality, after which they were deemed modality switches and censored from the as-treated analysis. Mortal events that occurred within that 120-day window were attributed to PD. We elected not to use a time-dependent internal covariate for treatment modality in the as-treated analysis as our primary goal was to provide estimates of adjusted patient survival over time. In the intent-to-treat analysis, survival was calculated from the date of first dialysis to the earliest of the following: death, renal transplantation, renal recovery, loss of KPSC coverage, or study end. Under this intent-to-treat analysis, mortal events were attributed to the initial therapy regardless of whether patients subsequently switched modalities. Thus, a patient who was on PD for 6 months, then transferred to HD for 12 months, and then died, would have a calculated survival time of 18 months in the intent-totreat analysis. Preliminary analysis based on a piece-wise interval Cox model found that mortality rates between PD and HD significantly violated the proportional hazards assumption. Subsequently, we fit a stratified Cox proportional hazards model to the propensity scorematched PD and HD patients in which separate baseline nonproportional hazard functions are assumed for the two modalities. We did so adjusting for any residual confounding due to baseline differences in age, sex, race, primary cause of ESRD, and Charlson comorbidity index as well as for differences in geocoded variables reflecting education and income levels. Robust standard errors based on score residuals aggregated over matched patients were used to correct for within matched-pairs correlation. The relative risk of death was estimated using a CHR, which is the ratio of the baseline cumulative hazard functions for HD relative to PD adjusted for case-mix differences at baseline. 24 It is a direct function of covariate-specific adjusted patient survival estimates for HD and PD and is a measure of the cumulative risk of death for HD relative to PD or of PD relative to HD depending on the reference group one wishes to choose. Although covariate-specific adjusted survival curves like those shown in Figure 4 are readily available with most software packages including SAS, they can be difficult to interpret when some of the covariates are strictly categorical. For example, the average value for gender in our data set is the fraction of males, which was 41. This is the mean value (i.e., proportion) used for gender when plotting the covariate-averaged survival curves. An alternative approach is to plot population-averaged or direct-adjusted PD and HD survival curves, which are obtained by averaging the estimated survival curves of patients on the basis of their unique set of covariate values. 23 A population-averaged survival curve corresponds to the average adjusted survival in the population and results in a more representative and interpretable survival curve. This is demonstrated in Figure All analyses were carried out using SAS version 9.2. All tests were two-sided and differences were considered significant at Po0.05. The study was approved by the Institutional Review Board. DISCLOSURE All the authors declared no competing interests. ACKNOWLEDGMENTS This manuscript was supported by a grant from Baxter Healthcare Incorporated. EFV, PhD, is a consultant for Baxter Healthcare Incorporated but agreed to serve as an adviser and co-author for this manuscript and received no financial compensation for his work. None of the authors have a financial interest in Baxter Healthcare Incorporated. The authors thank James Sloand, MD of Baxter Healthcare Corporation, for reviewing the manuscript prior to submission. His many comments were invaluable to the manuscript. The authors also thank Scott Rasgon, MD, chief of the Nephrology Department at KPSC, Los Angeles Medical Center, whose support allowed for this project to take place. REFERENCES 1. Fenton SS, Schaubel DE, Desmeules M et al. Hemodialysis versus peritoneal dialysis: a comparison of adjusted mortality rates. Am J Kidney Dis 1997; 30: Bloembergen WE, Port FK, Mauger EA et al. A comparison of mortality between patients treated with hemodialysis and peritoneal dialysis. JAm Soc Nephrol 1995; 6: Collins AJ, Hao W, Xia H et al. Mortality risks of peritoneal dialysis and hemodialysis. Am J Kidney Dis 1999; 34: Vonesh EF, Snyder JJ, Foley RN et al. Mortality studies comparing peritoneal dialysis and hemodialysis: what do they tell us? Kidney Int Suppl 2006; 103: S Xue JL, Everson SE, Constantini EG et al. Peritoneal and hemodialysis: II. Mortality risk associated with initial patient characteristics. Kidney Int 2002; 61: Jaar BG, Coresh J, Plantinga LC et al. Comparing the risk for death with peritoneal dialysis and hemodialysis in a national cohort of patients with chronic kidney disease. Ann Intern Med 2005; 143: Quinn RR, Hux JE, Oliver MJ et al. Selection bias explains apparent differential mortality between dialysis modalities. J Am Soc Nephrol 2011; 22: Yeates K, Zhu N, Vonesh E et al. Hemodialysis and peritoneal dialysis are associated with similar outcomes for end-stage renal disease treatment in Canada. Nephrol Dial Transplant 2012; 27: Mehrotra R, Chiu YW, Kalantar-Zadeh K et al. Similar outcomes with hemodialysis and peritoneal dialysis in patients with end-stage renal disease. Arch Intern Med 2011; 171: Moist LM, Trpeski L, Na Y et al. Increased hemodialysis catheter use in Canada and associated mortality risk: data from the Canadian Organ Replacement Registry Clin J Am Soc Nephrol 2008; 3: Lacson E Jr, Wang W, Hakim RM et al. Associates of mortality and hospitalization in hemodialysis: potentially actionable laboratory variables and vascular access. Am J Kidney Dis 2009; 53: Pisoni RL, Arrington CJ, Albert JM et al. Facility hemodialysis vascular access use and mortality in countries participating in DOPPS: an instrumental variable analysis. Am J Kidney Dis 2009; 53: Perl J, Wald R, McFarlane P et al. Hemodialysis vascular access modifies the association between dialysis modality and survival. J Am Soc Nephrol 2011; 22: Fleischmann E, Teal N, Dudley J et al. Influence of excess weight on mortality and hospital stay in 1346 hemodialysis patients. Kidney Int 1999; 55: Kato A, Odamaki M, Yamamoto T et al. Influence of body composition on 5 year mortality in patients on regular haemodialysis. Nephrol Dial Transplant 2003; 18: Stack AG, Murthy BV, Molony DA. differences between peritoneal dialysis and hemodialysis among "large" ESRD patients in the United States. Kidney Int 2004; 65: van der Wal WM, Noordzij M, Dekker FW et al. Netherlands Cooperative Study on the Adequacy of Dialysis Study Group (NECOSAD). Full loss of Kidney International (2014) 86,

7 VA Kumar et al.: in US peritoneal and hemodialysis patients residual renal function causes higher mortality in dialysis patients; findings from a marginal structural model. Nephrol Dial Transplant 2011; 26: Wang AY, Woo J, Wang M et al. Important differentiation of factors that predict outcome in peritoneal dialysis patients with different degrees of residual renal function. Nephrol Dial Transplant 2005; 20: Shafi T, Jaar BG, Plantinga LC et al. Association of residual urine output with mortality, quality of life, and inflammation in incident hemodialysis patients: the Choices for Healthy Outcomes in Caring for End-Stage Renal Disease (CHOICE) Study. Am J Kidney Dis 2010; 56: Slinin Y, Guo H, Gilbertson DT et al. Prehemodialysis care by dietitians and first-year mortality after initiation of hemodialysis. Am J Kidney Dis 2011; 58: Longenecker JC, Coresh J, Klag MJ et al. Validation of comorbid conditions on the end-stage renal disease medical evidence report: the CHOICE study. Choices for Healthy Outcomes in Caring for ESRD. JAm Soc Nephrol 2000; 11: Kumar VA, Sidell MA, Yang WT et al. Predictors of peritonitis, hospital days and technique survival for peritoneal dialysis patients in a managed care setting. Perit Dial Int 2013; 34: Zhang X, Loberiza FR, Klein JP et al. A SAS macro for estimation of direct adjusted survival curves based on a stratified Cox regression model. Comput Methods Programs Biomed 2007; 88: Wei G, Schaubel DE. Estimating cumulative treatment effects in the presence of nonproportional hazards. Biometrics 2008; 64: Weinhandl ED, Foley RN, Gilbertson DT et al. Propensity-matched mortality comparison of incident hemodialysis and peritoneal dialysis patients. J Am Soc Nephrol 2010; 21: Korevaar JC, Feith GW, Dekker FW et al. NECOSAD Study Group. Effect of starting with hemodialysis compared with peritoneal dialysis in patients new on dialysis treatment: a randomized controlled trial. Kidney Int 2003; 64: Lysaght MJ, Vonesh EF, Gotch F et al. The influence of dialysis treatment modality on the decline of remaining renal function. ASAIO Trans 1991; 37: Cancarini GC, Brunori G, Camerini G et al. Renal function recovery and maintenance of residual diuresis in CAPD and hemodialysis. Perit Dial Bull 1986; 6: Li PK, Chow KM, Wong TY et al. Effects of an angiotensin-converting enzyme inhibitor on residual renal function in patients receiving peritoneal dialysis. A randomized, controlled study. Ann Intern Med 2003; 139: McKane W, Chandna SM, Tattersall JE et al. Identical decline of residual renal function in high-flux biocompatible hemodialysis and CAPD. Kidney Int 2002; 61: McCarthy JT, Jenson BM, Squillace DP et al. Improved preservation of residual renal function in chronic hemodialysis patients using polysulfone dialyzers. Am J Kidney Dis 1997; 29: US Renal Data System, USRDS 2012 Annual Data Report: Atlas of Chronic Kidney Disease and End-Stage Renal Disease in the United States, National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases; 2012 (Vol. 2). 33. Acchiardo SR, Moore LW, Latour PA. Malnutrition as the main factor in morbidity and mortality of hemodialysis patients. Kidney Int Suppl 1983; 16: S199 S Iseki K, Kawazoe N, Fukiyama K. Serum albumin is a strong predictor of death in chronic dialysis patients. Kidney Int 1993; 44: Shemin D, Bostom AG, Laliberty P et al. Residual renal function and mortality risk in hemodialysis patients. Am J Kidney Dis 2001; 38: Shemin D, Bostom AG, Lambert C et al. Residual renal function in a large cohort of peritoneal dialysis patients: change over time, impact on mortality and nutrition. Perit Dial Int 2000; 20: Foley RN, Parfrey PS, Harnett JD et al. The impact of anemia on cardiomyopathy, morbidity, and and mortality in end-stage renal disease. Am J Kidney Dis 1996; 28: Block GA, Hulbert-Shearon TE, Levin NW et al. Association of serum phosphorus and calcium x phosphate product with mortality risk in chronic hemodialysis patients: A national study. Am J Kidney Dis 1998; 31: D Hoore W, Bouckaert A, Tilquin C. Practical considerations on the use of the Charlson comorbidity index with administrative data bases. J Clin Epidemiol 1996; 49: Hemmelgarn BR, Manns BJ, Quan H et al. Adapting the Charlson Comorbidity Index for use in patients with ESRD. Am J Kidney Dis 2003; 42: Krieger N, Chen JT, Waterman PD et al. Geocoding and monitoring of US socioeconomic inequalities in mortality and cancer incidence: does the choice of area-based measure and geographic level matter?: the Public Health Disparities Geocoding Project. Am J Epidemiol 2002; 156: Krieger N, Waterman P, Lemieux K et al. On the wrong side of the tracts? Evaluating the accuracy of geocoding in public health research. Am J Public Health 2001; 91: Kidney International (2014) 86,