REVIEW URRENT C OPINION
Dialysis modality and survival: does the controversy live on? Asad Ali Merchant a, Robert R. Quinn b, and Jeffrey Perl a
Purpose of review Continued debate regarding the relative mortality risk for end-stage renal disease patients treated with either peritoneal dialysis or facility-based three times weekly conventional haemodialysis (CHD) stems from the absence of adequately powered randomized controlled trials, and the reliance on observational studies. These reports have yielded important trends, but also conflicting results. Here, we summarize the contemporary literature on survival comparisons between CHD and peritoneal dialysis, highlighting trends and important differences between studies. Recent findings Large observational studies have not conclusively shown an overall survival advantage of either dialysis modality. Studies have consistently shown an early survival advantage for peritoneal dialysis relative to CHD. New insights including accounting for selection bias and the use of central venous catheters as incident haemodialysis access may explain much of this apparent early mortality difference. The relative mortality risk of peritoneal dialysis versus haemodialysis may be decreasing in more contemporary cohorts. Older patients, diabetic patients, and those with comorbidities may have a relatively worse prognosis on peritoneal dialysis compared to CHD. Summary Overall, survival of incident end-stage renal disease patients is similar for CHD and peritoneal dialysis, but early survival differences may be driven by selection bias. Decisions regarding modality choice should be individualized, considering other important patient outcomes including quality of life. Whereas a future randomized controlled trial is ideally suited to address this question, practical limitations may continue to limit its development. Keywords haemodialysis, mortality, peritoneal dialysis, randomized controlled trials, survival
INTRODUCTION The worldwide incidence and prevalence of endstage renal disease (ESRD) is increasing [1]. Currently, more than 80% of the world’s ESRD patients are treated with conventional, facility-based three times weekly conventional haemodialysis (CHD), whereas peritoneal dialysis remains the most common form of home-based renal replacement therapy (RRT) [1]. Peritoneal dialysis is an attractive treatment option for patients wishing to pursue increased flexibility and autonomy, but more restrictive medical and psychosocial eligibility criteria have traditionally limited its use. Regional variability in the use of peritoneal dialysis is driven by differences in healthcare policy, physician and treatment reimbursement, relative costs, and physician knowledge and attitudes towards peritoneal dialysis [2,3]. Peritoneal dialysis growth is largest across developing countries [4], likely as a means www.co-nephrolhypertens.com
to maximize RRT availability and use, while minimizing increasing dialysis-related healthcare expenditure. Across many countries, peritoneal dialysis remains less costly than facility-based haemodialysis with annualized treatment costs that are 60–70% of those for facility-based haemodialysis [5]. Moreover, peritoneal dialysis utilization has been increasing in the United States, from 6% in 2010 to nearly 10% in 2013 [6]. This resurgence has largely been a Division of Nephrology, and The Keenan Research Centre in the Li Ka Shing Knowledge Institute, St Michael’s Hospital, University of Toronto, Ontario and bDepartments of Medicine & Community Health Sciences, University of Calgary, Calgary, Alberta, Canada
Correspondence to Jeffrey Perl, St Michael’s Hospital, 3-060 Shuter, 30 Bond St, Toronto, Ontario M5B 1W8, Canada. Tel: +1 416 864 6016; fax: +1 416 864 3042; e-mail: [email protected] Curr Opin Nephrol Hypertens 2015, 24:276–283 DOI:10.1097/MNH.0000000000000114 Volume 24 Number 3 May 2015
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Dialysis modality and survival Merchant et al.
KEY POINTS Large observational studies suggest that the survival of patients on peritoneal dialysis compared to haemodialysis is similar, although results are often conflicting. Selection bias due to a ‘sicker’ and more acute haemodialysis population may explain the early survival advantage traditionally associated with the use of peritoneal dialysis. Older patients, diabetic patients, and patients with multiple comorbidities may have a worse prognosis on peritoneal dialysis as compared to haemodialysis, whereas younger patients with no comorbidities enjoy superior survival when treated with peritoneal dialysis as compared to haemodialysis. The prognosis of patients on both peritoneal dialysis and haemodialysis has improved with time; however, survival of patients on peritoneal dialysis has increased further as compared to haemodialysis among more contemporary cohorts.
attributed to a novel climate of favourable reimbursement that, in part, has incentivized the use of peritoneal dialysis over facility-based haemodialysis. In an era of renewed interest in peritoneal dialysis use, increasing attention will be focused on the examination of contemporary outcomes between peritoneal dialysis and CHD, including mortality, hospitalization, and health-related quality of life (HR-QOL). The relative mortality risks of the two modalities have long been debated, with over 30 years of literature published across several national registries worldwide. However, with ongoing evolution in the case-mix of patients, evolving technologies across both peritoneal dialysis and CHD, and the novel application of more sophisticated statistical and epidemiologic techniques aimed at addressing selection bias and other sources of confounding, ongoing re-examination of outcomes between the therapies is necessary. The purpose of the present narrative review is to highlight current and previous efforts aimed at establishing a randomized controlled trial (RCT) to explore outcomes comparing peritoneal dialysis and CHD, and to review contemporary survival differences between CHD and peritoneal dialysis across observational studies. We will review important insights that have emerged in our understanding of the relationship between dialysis modality and survival.
CURRENT AND PREVIOUS RANDOMIZED CONTROLLED TRIALS A RCT would be ideally suited to compare the survival differences across the two dialysis modalities.
Such an endeavour would not only serve to balance important differences in the characteristics of patients selected for peritoneal dialysis and CHD, but would also likely provide valid comparisons among the two groups of patients that are eligible for either peritoneal dialysis or CHD. Up until recently, this has only been attempted once; the Netherlands Cooperative Study on the Adequacy of Dialysis (NECOSAD) was designed to compare the survival and QOL between peritoneal dialysis and haemodialysis patients in an intention-to-treat (ITT) analysis. Unfortunately, recruitment into the trial was prematurely terminated with only 38 patients randomized after 3 years. Therefore, the study was underpowered and failed to find any meaningful and significant survival differences between the two therapies [7]. Although the results of the trial were disappointing, it did serve to underscore the particular difficulty in recruiting and randomizing patients for dialysis trials. The majority of the screened patients in NECOSAD had no contraindications to either modality; however, they indicated a strong preference for a particular modality, and most refused to be randomized [7]. Nonmedical factors such as lifestyle considerations, the ability to perform self-care, and the availability of support and other resources are likely very important in modality decisionmaking [3]. Therefore, if future attempts at RCTs are to be successful across the developed world, a pragmatic approach that retains patient choice will be critical to its design, but such would seem a near impossibility. Notwithstanding, there is a RCT underway comparing survival in Chinese peritoneal dialysis and CHD patients (China Q study – NCT01413074). The inclusion criteria for the trial are any ESRD patients aged 18 or older, and expected to start on RRT within 10 weeks after diagnosis. Patients must be eligible for either peritoneal dialysis or CHD, and those with previous kidney transplantation or already receiving dialysis are excluded. A total of 594 patients were to be randomized from 16 centres and followed for a minimum of 1 year after randomization. To date, 381 patients have been randomized, but mortality rates have been lower than expected, giving rise to concerns regarding the power of the study to compare survival differences between the two modalities. This has led investigators to change the primary outcome from survival to QOL as measured by the Kidney Disease Quality of Life Short Form Questionnaire (KDQoL-SF). The recruitment phase is expected to be complete by 2015 with results anticipated by 2016 after completion of follow-up (Xueqing Yu, Principle investigator of China Q study, personal communication).
1062-4821 Copyright ß 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.co-nephrolhypertens.com
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
277
Epidemiology and prevention
OBSERVATIONAL STUDIES There have been numerous observational trials in many countries comparing peritoneal dialysis and haemodialysis mortality rates. The majority are large multicentre observational studies borne out of national and regional registry-based data [8–12,13 ,14–20,21 ,22,23 ,24,25] and have reported inconsistent findings (Table 1). For example, Mehrotra et al. [26] showed no difference in mortality risk among peritoneal dialysis and CHD cohorts in a large US Renal Disease System (USRDS) registry-based study (684 426 incident dialysis patients), while Heaf and Wehberg [21 ] found a persistent survival advantage for peritoneal dialysis patients in a study of 12 095 patients from the Danish Terminal Uremia Register. There are several potential reasons for these conflicting results. It may reflect differences in the case-mix of patients selected for peritoneal dialysis over CHD, and the breadth and scope of case-mix adjustment. For example, peritoneal dialysis patients in the USA are traditionally younger with fewer comorbidities compared to their CHD counterparts, whereas peritoneal dialysis patients in Australia and New Zealand are typically older with more comorbidities than their CHD counterparts. In addition, the majority of these studies have been conducted using large national registry-based datasets that may suffer from variable degrees of bias owing to the degree and extent of data validation. For example, a recent validation exercise of the medical evidence report of the United States Renal Data System revealed systematic underreporting of comorbidities with a sensitivity of comorbidity reporting that was systematically higher for peritoneal dialysis patients than for haemodialysis patients [30]. Differences in the study designs and methodological techniques used may also explain the apparently discrepant findings in observational studies. For example, earlier studies included all prevalent dialysis patients [8,31,32], whereas subsequent studies restricted comparisons to incident dialysis populations. Moreover, some studies included follow-up within the first 90 days of RRT initiation, whereas follow-up across other studies only started after the first 90 days on dialysis and therefore excluded comparisons between patients who died in the first 90 days [10,12,13 ,16,17,19,33]. Weinhandl et al. [18], in 2010, demonstrated that the inclusion or exclusion of the first 90 days was critical to interpretations of modality-related survival, in that, excluding the first 90 days resulted in the loss of an apparent survival advantage of peritoneal dialysis over CHD. Given that modality switches between peritoneal dialysis and CHD are frequent, some investigators &
&
&
&
&
278
www.co-nephrolhypertens.com
have proposed that an as-treated analysis may be more appropriate than an ITT analysis [15]. In an as-treated analysis, patient mortality is attributed to the current modality, which the patient is receiving, while an ITT approach attributes the mortality to the patient’s initial therapy, irrespective of whether or not a change has occurred to another modality. In this regard, ITT approaches better reflect clinical decision-making in that it bases outcomes on what modality is initially received without utilizing any future information on modality switches, which would obviously not have been known or easily predicted at the time of the modality decision. Furthermore, it is likely that an ITT analysis is also a more valid comparison, given that modality switches may be due to worsening health related to the initial therapy, and may even be considered a premorbid event. In some studies, a death that has occurred in a relatively short period after a switch in modality has been assigned as a death attributable to the initial modality serving as an additional sensitivity analysis [28]. Most studies have shown that an ITT analysis attenuates the survival advantage of peritoneal dialysis over haemodialysis [15,18,23 ]. Taken together, it is likely that both ITT and astreated analyses are complimentary, and both provide important and useful information. Another factor may be the differing statistical techniques used to adjust for the case-mix differences between the two populations. Across all statistical techniques there is the concern that unaccounted and unmeasured confounding variables that may significantly bias the results. Recent investigators have used increasingly sophisticated analytic models, such as propensity score matching. However, using the Dutch End-Stage-Renal Disease Registry, Liem et al. [34] demonstrated similar findings using propensity score models versus a traditional multivariable-adjusted model in assessing the impact of modality on survival. Moreover, in many cases, propensity score matching excludes either haemodialysis or peritoneal dialysis patients from the analyses which remain incomparable, thus creating a ‘fantasy’ population that increasingly veers away from ‘real world’ comparisons of populations of dialysis patients. Marginal structural models (MSMs), which, in many cases, use inverse proportional treatment and censoring weighting (IPTCW) to create regression models that account for time-dependent confounders, and balance treatment-specific known covariate distributions, also allow handling of informative censoring for events such as kidney transplantation rates. [13 ,26,28,29]. This is especially important as transplantation rates are traditionally disproportionately higher in some peritoneal dialysis cohorts relative to &
&
Volume 24 Number 3 May 2015
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
1062-4821 Copyright ß 2015 Wolters Kluwer Health, Inc. All rights reserved. 8273
Australia/New Zealand (ANZDATA)
Canada (CORR)
USA (DaVita/USRDS)
Denmark (Danish Terminal Uraemia Register)
Marshall et al. (2011) [29]
Yeates et al. (2012) [25] Lukowsky et al. & (2013) [13 ]
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved. 1003
21 184
31 100
4538
1003
3822
1358
14 308
11 532
7412
2035
64 406
6337
PD
2001–2013
1990–2010
2001–2004
1991–2004
1996–2007
2001–2008
1998–2006
1996–2004
2003
Years of study
Patients with CVC use in the first 90 days excluded
0.42 (0.29–0.60)
1.20 (0.81–1.75)
0.94 (0.87–1.01)
1990–1999
6–12 mo: 0.69 (0.56–0.86)
PS-matched pairs
0.83 (0.77–0.89)
0–6 mo: 0.54 (0.44–0.66)
Analysis stratified by era of dialysis initiation
2000–2010
0.52 (0.34–0.80)
1.08 (1.04–1.11)
1.10 (1.06–1.16)
1.0 (0.91–1.0)
0.97 (0.88–1.07)
1.03 (0.99–1.06)
2002–2004
1.21 (0.85–1.70)
2000–2010
0.59 (0.44–0.78)
n/a
0.80 (0.73–0.87)
0.67 (0.59–0.71)
n/a
n/a
0.90 (0.76–1.06)
Overall adjusted HR (95% CI) (ITT analysis)
Prospective cohort
Analysis stratified by era of dialysis initiation MSM
MSM
Additional MSM analysis performed
PD compared with HD-CVC and HD-AVF/G separately [HD-AVF/G (n) ¼ 6663]
Patients with
Dialysis modality and survival: does the controversy live on? Asad Ali Merchant a, Robert R. Quinn b, and Jeffrey Perl a
Purpose of review Continued debate regarding the relative mortality risk for end-stage renal disease patients treated with either peritoneal dialysis or facility-based three times weekly conventional haemodialysis (CHD) stems from the absence of adequately powered randomized controlled trials, and the reliance on observational studies. These reports have yielded important trends, but also conflicting results. Here, we summarize the contemporary literature on survival comparisons between CHD and peritoneal dialysis, highlighting trends and important differences between studies. Recent findings Large observational studies have not conclusively shown an overall survival advantage of either dialysis modality. Studies have consistently shown an early survival advantage for peritoneal dialysis relative to CHD. New insights including accounting for selection bias and the use of central venous catheters as incident haemodialysis access may explain much of this apparent early mortality difference. The relative mortality risk of peritoneal dialysis versus haemodialysis may be decreasing in more contemporary cohorts. Older patients, diabetic patients, and those with comorbidities may have a relatively worse prognosis on peritoneal dialysis compared to CHD. Summary Overall, survival of incident end-stage renal disease patients is similar for CHD and peritoneal dialysis, but early survival differences may be driven by selection bias. Decisions regarding modality choice should be individualized, considering other important patient outcomes including quality of life. Whereas a future randomized controlled trial is ideally suited to address this question, practical limitations may continue to limit its development. Keywords haemodialysis, mortality, peritoneal dialysis, randomized controlled trials, survival
INTRODUCTION The worldwide incidence and prevalence of endstage renal disease (ESRD) is increasing [1]. Currently, more than 80% of the world’s ESRD patients are treated with conventional, facility-based three times weekly conventional haemodialysis (CHD), whereas peritoneal dialysis remains the most common form of home-based renal replacement therapy (RRT) [1]. Peritoneal dialysis is an attractive treatment option for patients wishing to pursue increased flexibility and autonomy, but more restrictive medical and psychosocial eligibility criteria have traditionally limited its use. Regional variability in the use of peritoneal dialysis is driven by differences in healthcare policy, physician and treatment reimbursement, relative costs, and physician knowledge and attitudes towards peritoneal dialysis [2,3]. Peritoneal dialysis growth is largest across developing countries [4], likely as a means www.co-nephrolhypertens.com
to maximize RRT availability and use, while minimizing increasing dialysis-related healthcare expenditure. Across many countries, peritoneal dialysis remains less costly than facility-based haemodialysis with annualized treatment costs that are 60–70% of those for facility-based haemodialysis [5]. Moreover, peritoneal dialysis utilization has been increasing in the United States, from 6% in 2010 to nearly 10% in 2013 [6]. This resurgence has largely been a Division of Nephrology, and The Keenan Research Centre in the Li Ka Shing Knowledge Institute, St Michael’s Hospital, University of Toronto, Ontario and bDepartments of Medicine & Community Health Sciences, University of Calgary, Calgary, Alberta, Canada
Correspondence to Jeffrey Perl, St Michael’s Hospital, 3-060 Shuter, 30 Bond St, Toronto, Ontario M5B 1W8, Canada. Tel: +1 416 864 6016; fax: +1 416 864 3042; e-mail: [email protected] Curr Opin Nephrol Hypertens 2015, 24:276–283 DOI:10.1097/MNH.0000000000000114 Volume 24 Number 3 May 2015
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Dialysis modality and survival Merchant et al.
KEY POINTS Large observational studies suggest that the survival of patients on peritoneal dialysis compared to haemodialysis is similar, although results are often conflicting. Selection bias due to a ‘sicker’ and more acute haemodialysis population may explain the early survival advantage traditionally associated with the use of peritoneal dialysis. Older patients, diabetic patients, and patients with multiple comorbidities may have a worse prognosis on peritoneal dialysis as compared to haemodialysis, whereas younger patients with no comorbidities enjoy superior survival when treated with peritoneal dialysis as compared to haemodialysis. The prognosis of patients on both peritoneal dialysis and haemodialysis has improved with time; however, survival of patients on peritoneal dialysis has increased further as compared to haemodialysis among more contemporary cohorts.
attributed to a novel climate of favourable reimbursement that, in part, has incentivized the use of peritoneal dialysis over facility-based haemodialysis. In an era of renewed interest in peritoneal dialysis use, increasing attention will be focused on the examination of contemporary outcomes between peritoneal dialysis and CHD, including mortality, hospitalization, and health-related quality of life (HR-QOL). The relative mortality risks of the two modalities have long been debated, with over 30 years of literature published across several national registries worldwide. However, with ongoing evolution in the case-mix of patients, evolving technologies across both peritoneal dialysis and CHD, and the novel application of more sophisticated statistical and epidemiologic techniques aimed at addressing selection bias and other sources of confounding, ongoing re-examination of outcomes between the therapies is necessary. The purpose of the present narrative review is to highlight current and previous efforts aimed at establishing a randomized controlled trial (RCT) to explore outcomes comparing peritoneal dialysis and CHD, and to review contemporary survival differences between CHD and peritoneal dialysis across observational studies. We will review important insights that have emerged in our understanding of the relationship between dialysis modality and survival.
CURRENT AND PREVIOUS RANDOMIZED CONTROLLED TRIALS A RCT would be ideally suited to compare the survival differences across the two dialysis modalities.
Such an endeavour would not only serve to balance important differences in the characteristics of patients selected for peritoneal dialysis and CHD, but would also likely provide valid comparisons among the two groups of patients that are eligible for either peritoneal dialysis or CHD. Up until recently, this has only been attempted once; the Netherlands Cooperative Study on the Adequacy of Dialysis (NECOSAD) was designed to compare the survival and QOL between peritoneal dialysis and haemodialysis patients in an intention-to-treat (ITT) analysis. Unfortunately, recruitment into the trial was prematurely terminated with only 38 patients randomized after 3 years. Therefore, the study was underpowered and failed to find any meaningful and significant survival differences between the two therapies [7]. Although the results of the trial were disappointing, it did serve to underscore the particular difficulty in recruiting and randomizing patients for dialysis trials. The majority of the screened patients in NECOSAD had no contraindications to either modality; however, they indicated a strong preference for a particular modality, and most refused to be randomized [7]. Nonmedical factors such as lifestyle considerations, the ability to perform self-care, and the availability of support and other resources are likely very important in modality decisionmaking [3]. Therefore, if future attempts at RCTs are to be successful across the developed world, a pragmatic approach that retains patient choice will be critical to its design, but such would seem a near impossibility. Notwithstanding, there is a RCT underway comparing survival in Chinese peritoneal dialysis and CHD patients (China Q study – NCT01413074). The inclusion criteria for the trial are any ESRD patients aged 18 or older, and expected to start on RRT within 10 weeks after diagnosis. Patients must be eligible for either peritoneal dialysis or CHD, and those with previous kidney transplantation or already receiving dialysis are excluded. A total of 594 patients were to be randomized from 16 centres and followed for a minimum of 1 year after randomization. To date, 381 patients have been randomized, but mortality rates have been lower than expected, giving rise to concerns regarding the power of the study to compare survival differences between the two modalities. This has led investigators to change the primary outcome from survival to QOL as measured by the Kidney Disease Quality of Life Short Form Questionnaire (KDQoL-SF). The recruitment phase is expected to be complete by 2015 with results anticipated by 2016 after completion of follow-up (Xueqing Yu, Principle investigator of China Q study, personal communication).
1062-4821 Copyright ß 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.co-nephrolhypertens.com
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
277
Epidemiology and prevention
OBSERVATIONAL STUDIES There have been numerous observational trials in many countries comparing peritoneal dialysis and haemodialysis mortality rates. The majority are large multicentre observational studies borne out of national and regional registry-based data [8–12,13 ,14–20,21 ,22,23 ,24,25] and have reported inconsistent findings (Table 1). For example, Mehrotra et al. [26] showed no difference in mortality risk among peritoneal dialysis and CHD cohorts in a large US Renal Disease System (USRDS) registry-based study (684 426 incident dialysis patients), while Heaf and Wehberg [21 ] found a persistent survival advantage for peritoneal dialysis patients in a study of 12 095 patients from the Danish Terminal Uremia Register. There are several potential reasons for these conflicting results. It may reflect differences in the case-mix of patients selected for peritoneal dialysis over CHD, and the breadth and scope of case-mix adjustment. For example, peritoneal dialysis patients in the USA are traditionally younger with fewer comorbidities compared to their CHD counterparts, whereas peritoneal dialysis patients in Australia and New Zealand are typically older with more comorbidities than their CHD counterparts. In addition, the majority of these studies have been conducted using large national registry-based datasets that may suffer from variable degrees of bias owing to the degree and extent of data validation. For example, a recent validation exercise of the medical evidence report of the United States Renal Data System revealed systematic underreporting of comorbidities with a sensitivity of comorbidity reporting that was systematically higher for peritoneal dialysis patients than for haemodialysis patients [30]. Differences in the study designs and methodological techniques used may also explain the apparently discrepant findings in observational studies. For example, earlier studies included all prevalent dialysis patients [8,31,32], whereas subsequent studies restricted comparisons to incident dialysis populations. Moreover, some studies included follow-up within the first 90 days of RRT initiation, whereas follow-up across other studies only started after the first 90 days on dialysis and therefore excluded comparisons between patients who died in the first 90 days [10,12,13 ,16,17,19,33]. Weinhandl et al. [18], in 2010, demonstrated that the inclusion or exclusion of the first 90 days was critical to interpretations of modality-related survival, in that, excluding the first 90 days resulted in the loss of an apparent survival advantage of peritoneal dialysis over CHD. Given that modality switches between peritoneal dialysis and CHD are frequent, some investigators &
&
&
&
&
278
www.co-nephrolhypertens.com
have proposed that an as-treated analysis may be more appropriate than an ITT analysis [15]. In an as-treated analysis, patient mortality is attributed to the current modality, which the patient is receiving, while an ITT approach attributes the mortality to the patient’s initial therapy, irrespective of whether or not a change has occurred to another modality. In this regard, ITT approaches better reflect clinical decision-making in that it bases outcomes on what modality is initially received without utilizing any future information on modality switches, which would obviously not have been known or easily predicted at the time of the modality decision. Furthermore, it is likely that an ITT analysis is also a more valid comparison, given that modality switches may be due to worsening health related to the initial therapy, and may even be considered a premorbid event. In some studies, a death that has occurred in a relatively short period after a switch in modality has been assigned as a death attributable to the initial modality serving as an additional sensitivity analysis [28]. Most studies have shown that an ITT analysis attenuates the survival advantage of peritoneal dialysis over haemodialysis [15,18,23 ]. Taken together, it is likely that both ITT and astreated analyses are complimentary, and both provide important and useful information. Another factor may be the differing statistical techniques used to adjust for the case-mix differences between the two populations. Across all statistical techniques there is the concern that unaccounted and unmeasured confounding variables that may significantly bias the results. Recent investigators have used increasingly sophisticated analytic models, such as propensity score matching. However, using the Dutch End-Stage-Renal Disease Registry, Liem et al. [34] demonstrated similar findings using propensity score models versus a traditional multivariable-adjusted model in assessing the impact of modality on survival. Moreover, in many cases, propensity score matching excludes either haemodialysis or peritoneal dialysis patients from the analyses which remain incomparable, thus creating a ‘fantasy’ population that increasingly veers away from ‘real world’ comparisons of populations of dialysis patients. Marginal structural models (MSMs), which, in many cases, use inverse proportional treatment and censoring weighting (IPTCW) to create regression models that account for time-dependent confounders, and balance treatment-specific known covariate distributions, also allow handling of informative censoring for events such as kidney transplantation rates. [13 ,26,28,29]. This is especially important as transplantation rates are traditionally disproportionately higher in some peritoneal dialysis cohorts relative to &
&
Volume 24 Number 3 May 2015
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
1062-4821 Copyright ß 2015 Wolters Kluwer Health, Inc. All rights reserved. 8273
Australia/New Zealand (ANZDATA)
Canada (CORR)
USA (DaVita/USRDS)
Denmark (Danish Terminal Uraemia Register)
Marshall et al. (2011) [29]
Yeates et al. (2012) [25] Lukowsky et al. & (2013) [13 ]
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved. 1003
21 184
31 100
4538
1003
3822
1358
14 308
11 532
7412
2035
64 406
6337
PD
2001–2013
1990–2010
2001–2004
1991–2004
1996–2007
2001–2008
1998–2006
1996–2004
2003
Years of study
Patients with CVC use in the first 90 days excluded
0.42 (0.29–0.60)
1.20 (0.81–1.75)
0.94 (0.87–1.01)
1990–1999
6–12 mo: 0.69 (0.56–0.86)
PS-matched pairs
0.83 (0.77–0.89)
0–6 mo: 0.54 (0.44–0.66)
Analysis stratified by era of dialysis initiation
2000–2010
0.52 (0.34–0.80)
1.08 (1.04–1.11)
1.10 (1.06–1.16)
1.0 (0.91–1.0)
0.97 (0.88–1.07)
1.03 (0.99–1.06)
2002–2004
1.21 (0.85–1.70)
2000–2010
0.59 (0.44–0.78)
n/a
0.80 (0.73–0.87)
0.67 (0.59–0.71)
n/a
n/a
0.90 (0.76–1.06)
Overall adjusted HR (95% CI) (ITT analysis)
Prospective cohort
Analysis stratified by era of dialysis initiation MSM
MSM
Additional MSM analysis performed
PD compared with HD-CVC and HD-AVF/G separately [HD-AVF/G (n) ¼ 6663]
Patients with
