Nephrologists are faced every day with an increasing number of transplant patients in whom the kidney graft fails and must return to dialysis. In recent years, in Spain, between 2 and 4% of people who started dialysis did so due to kidney graft dysfunction. In absolute numbers, this figure represents about a thousand patients per year.1,2

Informing people with non-transplanted kidney failure about which modality of renal replacement therapy best suits their circumstances can be straightforward, but in those with kidney graft failure, this decision is complicated. The evidence base for that advice remains incomplete, as survival comparisons between dialysis modalities are fraught with methodological difficulties, in which a randomized trial is nearly impossible to perform.

There are few publications that evaluate the impact of the dialysis modality on the survival of the transplanted patient with a non-functioning graft who returns to dialysis. Most are local series with few participants in which similar results are reported in relation to survival between peritoneal dialysis (PD) and hemodialysis (HD).3–6 Currently, only one study demonstrates the superiority of the DP in terms of mortality compared to the HD during the first year after graft failure in July. None of these studies have taken into account the presence of competitive events for statistical analysis. By competitive event we understand that which makes it difficult or modifies the possibility of observing the event of interest.8,9 In such a way that, when we are studying death or the rate of hospital admissions in a certain dialysis technique, the abandonment of this by transplant or by transfer to another type of dialysis supposes a competitive event, since the patient ceases to be at risk of dying or having a hospital admission in the dialysis technique under study. In the presence of competitive events, the usual methods of time-to-event analysis such as the Kaplan–Meier survival curves are not suitable as they can overestimate the real event rate.10,11

It is a known fact that transplant patients who are on PD have a high rate of transfer to HD due to the rapid loss of residual renal function (RRF),4,12 therefore, the analysis, taking into account the competitive events, should be a priority when we want to study survival in a certain dialysis modality.

The aim of our study was to analyze the impact of dialysis modality (PD vs. HD) transplant patients returning to dialysis on mortality and morbidity (hospitalizations and death), taking into account the presence of competitive events.

The results of the qualitative variables are expressed by absolute and relative frequency. Continuous variables are shown by the mean ± standard deviation or median and interquartile range (25th and 75th percentile), depending on whether or not they are normally distributed. The comparison of qualitative variables between the PD and HD groups was carried out using the X2 test. Quantitative variables were compared using the Student’s t test or the Mann-Whitney test, depending on whether the distribution was normal or not.

The raw survival of the patient was analyzed by performing Kaplan–Meier curves, comparing the results by means of the Breslow test and exposing the tables of individuals at risk. The graph includes the follow-up time until less than 10% of patients remain at risk.14

To determine the association of the dialysis modality (HD vs. PD) on survival and on hospital admissions (main objective), a competitive risk analysis was carried out, considering as such: the change of dialysis modality, the performance of a transplant and the recovery of renal function, since its appearance competed with that of death or admission.

Different models were carried out where those variables with a clinical sense about the study event and that met the proportional risk assumption were selected. The initial maximum model in each case included a number of variables, together with their possible interactions, that did not exceed 10% of the number of events. The final model was built behind a backwards or backwards strategy. The variables included in the final model for survival were: dialysis modality, age and suboptimal use of the access, and for hospitalization it was the type of dialysis. The criterion for evaluating confusion was a change of more than 10% in the sub Hazard Ratio (sHR) of the variable of interest (dialysis modality) when comparing a model adjusted for the possible confounder and one without adjustment. To control the selection bias dialysis technique performing a propensity index explored, but, due to the limited number of patients and homogeneity of groups, this did not converge properly. To determine the association of the dialysis modality on the appearance of graft intolerance and retransplantation (secondary objectives), a Cox analysis was performed. Due to the limited number of events, one could be included in the final model, in addition to the treatment modality, age and glomerular filtration rate (GFR) to the start of the dialysis.

Values ​​of p < 0.05 were considered significant. The statistical package STATA® 13.1 (StataCorp, College Station, TX) was used for the analysis with competitive risks and the SPSS® 15.0 package (SPSS Inc., Chicago, USA) for the rest of the statistical analysis.

A total of 175 patients who started dialysis after graft failure were analyzed, 89 did so on HD and 86 on PD from January 2000 to December 2013. The mean follow-up of the entire series was 34 ± 33 months, with a median of 24 (IQR 7–50 months).

Association between dialysis modality and survival

There were no differences in survival between the PD group and the HD group at one year or at five years: 95% in PD vs. 92% in HD at one year and 69% at five years in both groups, Breslow test, p = 0.30 (Fig. 1).

Fig. 1.

Kaplan–Meier survival curves. They compare crude mortality in the PD group (n = 86, blue line) vs. in the HD group (n = 89, red line). Breslow p = 0.30. The table shows the patients at risk throughout the follow-up.

(0.08MB).

In the univariate analysis, mortality was related to the dialysis modality (sHR[DP/HD] 0.37 95% CI: 0.175 to 0.774, p = 0.008), age (sHR 1.05 95% CI: 0, 06 to 1.09, p = 0.000), a higher Charlson index (sHR 1.59 95% CI: 1.28–1.98, p = 0.000) and the presence of vascular comorbidity (sHR 2.78 CI 95 %: 1.41–5.50, p = 0.003. The predictive model established that the variables that best predicted patient mortality in our series were age with a sHR 1.06 (95% CI: 1.033–1.106, p = 0.000), the non-optimal use of access with an sHR 3.00 (95% CI: 1.507–5.982, p = 0.028) and the type of dialysis, in this case the PD obtained a sHR[PD/HD] 0, 36 (95% CI: 0.148 to 0.890, p = 0.028) (Table 2). The cumulative incidence of death adjusted by the competitive risk model at 1 year was 3% for PD and 11% for HD and at five 8% for PD and 21% for HD, p = 0.028 (Fig. 2).

Table 2.

Analysis by competitive risks, according to the modality of dialysis PD vs. HD (sHR). Ref 1 = HD.

Variables	S HR	CI (95%)	p
Mortality	0.36	0.148−0.894	0.02
Hospitalization	0.52	0.369−0.743	<0.001

Mortality adjusted for: age, vascular comorbidity, suboptimal use of access. Hospital admission adjusted for: age, vascular comorbidity, glomerular filtration at the start of dialysis, and suboptimal use of access. SHR: sub Hazard Ratio.

Fig. 2.

Cumulative incidence of death adjusted by the competitive risk model in the global series of patients. The table shows the patients at risk throughout the follow-up.

(0.07MB).

Association between the dialysis modality and the risk of hospitalization

The hospitalization rate was significantly lower for the PD group, 0.68 admission/patient/year (one admission every 17 months) than for the HD group, 1.01 admission/patient/year (one admission every 12 months), p = 0.0011. We did not find significant differences in the causes of admission between the two dialysis groups (PD and HD) (Table 3).

Table 3.

Description of all causes of admission in the global series and comparative analysis between both dialysis groups (p = 0.257).

Cause entry	Everyone	DP	HD
	n = 431	n = 117	n = 314
Cardiovascular, n (%)	120 (28)	40 (34.1)	80 (25.4)
Infectious, n (%)	98 (22.2)	22 (18.8)	76 (24.2)
Graft intolerance, n (%)	63 (14.3)	15 (12.8)	48 (15.2)
Related to the dialysis technique, n (%)	58 (13.6)	15 (12.8)	43 (13.6)
Others, n (%)	92 (21.5)	25 (21.3)	67 (21.3)

PD: peritoneal dialysis; HD: hemodialysis.

In the multivariate analysis, the initial model included age, suboptimal access use, vascular comorbidity, Charlson index, DM, and dialysis modality. The final model established that the variable that best predicted the probability of hospital admission of the patient in our series was the dialysis modality, which, in the case of PD, presented a sHR[PD/HD] 0.52 (95% CI: 0.369 a 0.743, p = < 0.001) (Table 2 ).

The cumulative incidence of hospital admission adjusted by the competitive risk model was 47% for PD and 63% for HD in the first year, and 71% for PD and 88% for HD at five years, p < 0.001 (Fig. 3).

Fig. 3.

Cumulative incidence of hospital admissions adjusted by the competitive risks model. The table shows the patients at risk throughout the follow-up.

(0.09MB).

Association between the dialysis modality and the risk of graft intolerance syndrome

The cumulative incidence of graft intolerance was 32% at 12 months and 42% at 24 months for HD and 10% at 12 months and 17% at 24 months for PD (Fig. 4). Patients who developed graft intolerance had a higher eGFR at the start of dialysis (16 ± 9 vs. 12 ± 8 mL/min/1.72 m2, p = 0.016) and a higher percentage were on HD (32.8 vs. 11.6%, p = < 0.001), with no significant differenes regarding the time of presentation, median of seven months (IQR 5–12 months) in HD versus 12 months (IQR 6–20 months) in PD, p = 0.257. In the multivariate analysis, after adjusting for age and eGFR at the start of dialysis, PD was associated, less frequently, with the appearance of a graft intolerance syndrome, HR 0.307 (95% CI 0.142 to 0.758, p = 0.009) (Table 4).

Fig. 4.

Cumulative incidence of graft intolerance throughout follow-up, separated by dialysis groups (HD and PD).

(0.11MB).

Table 4.

Cox analysis for the study of risk factors for developing graft intolerance.

Variable	HR	CI (95%)	p
PD dialysis mode (ref 1 = HD)	0.307	(0.142−0.758)	0.009
Age (year)	0.994	(0.962−1.009)	0.22
GFR at the start of dialysis	0.998	(0.962−1.035)	0.89

PD: peritoneal dialysis; HD: hemodialysis; HR: Hazard Ratio; DF: glomerular filtration.

To our knowledge, this is the first study at the national level in which PD has been more advantageous in terms of morbidity and mortality than HD after renal graft failure, taking into account competitive events. In the literature, we find numerous publications that study the morbidity and mortality of transplanted patients who return to dialysis and compare it with those who have not been transplanted,12,15–19 but only five publications relate the impact of the dialysis modality on said morbidity and mortality.3–7 Our results have great clinical relevance, since we should make you reconsider the current trend of not including PD patients after graft failure, as the vast majority of those who return to dialysis do with HD.20

In our study, the patients who were included in PD after graft failure were significantly younger and with less cardiovascular morbidity than those who were admitted to HD, a fact that was consistent with most of the published series.3,6,7 This characteristic represents an important bias that can be partially corrected through statistical analysis adjusted to the clinical and demographic variables available.4,5 However, according to some authors,21 there are characteristics such as self-care capacity, responsibility towards dialysis treatment and compliance with it, which are not included in most studies and are therefore not available at the time of comparative analysis of dialysis groups. These characteristics, considered as “functional” are an important statistical confounder, in such a way that patients who had a greater degree of commitment to their disease, greater capacity for self-care and who enjoyed a better state of health, tend to prefer a technique home dialysis such as PD. On the contrary, individuals who are less disciplined, less compliant and therefore, with a potential for a worse prognosis, would choose HD, thus worsening the overall survival of this group.

In studies with large number of patients, the use of sophisticated statistical tools such as propensity test allows two comparable groups and partially eliminate selection bias technique.7 In our work, the possibility of performing a propensity test was explored, but it did not converge adequately, due to the limited number of patients, so its application was not possible. Consequently, one of the fundamental limitations is the presence of a selection bias that may justify, despite the statistical adjustment, that the group of people on PD present better survival than those included on HD after graft failure.

Another notable difference in our series was that HD patients started dialysis with higher eGFRs than those on PD. As was the case in the study by Perl et al.,7 we consider that this phenomenon could be related to the fact that the subject who begins HD does so more often due to episodes of heart failure derived from a worse clinical state and needs dialysis with higher eGFRs than those who do so on PD. The existence of a functioning internal arteriovenous fistula (iAF) or the ease of placing a central venous catheter and its immediate use could also favor this earlier onset. On the contrary, for those patients who wish to start PD, the implantation of a peritoneal catheter must be scheduled and a minimum time of maturation must be ensured, facts that could delay the start of dialysis for weeks or even months.

Another fact to highlight is that the percentage of diabetic people who start PD is only 13%. This is probably due to the low inclusion of diabetic subjects in PD in the hospitals of the Levante Registry, since, if we analyze the incorporation of patients in PD de novo (not transplanted), we also see that the percentage of diabetics is lower than that reported nationwide (14.9 vs. 25%).1,2

The mean follow-up of the entire series was 24 months, as in previous studies,7,18,22 being shorter in the PD group due to the need to transfer to HD due to the inability to obtain an adequate dose of dialysis. According to some authors, the cause of this insufficient dose could be related to the rapid loss of RRF in transplanted grafts, especially if it is decided to suspend immunosuppression after failure.4,23

In our series, the survival of transplanted patients after the start of dialysis was 93% in the first year and 70% at five years, data similar to those reported in most of the published studies.19,22

The main causes of death in our study were cardiovascular and infectious. These findings confirm that our results are comparable to those described in previous studies, both recent24 and old25 and show that the causes of mortality in these patients have not changed significantly in the last 15 years, indicating which are still the main causes. Challenges in the field of morbidity and mortality in transplanted patients who start dialysis.

The independent risk factors that increased mortality in our analysis were age, described in most of the published series,17,19,24,25 the HD modality with respect to PD, and the initiation of dialysis through non-optimal access.

In our research, PD has been associated with lower mortality compared to HD. In the literature, there are publications in both directions, in the majority, mortality is similar in both dialysis techniques when the graft fails3–6 while in others, such as the study by Perl et al.,7 mortality varies according to time on dialysis. In this work, the authors describe that, after the start of dialysis when the graft fails, PD is superior in terms of survival up to the first year, equals HD between the first and second year and, beyond two years, PD would have higher mortality than HD. According to these data, the better survival of the PD group in our series could be explained by the median survival in this group with the technique. Furthermore, based on these findings, one could speculate on the advisability of not delaying the transition to HD after a year and a half so as not to worsen patient survival.

Non-optimal access at the beginning of dialysis, both for HD and PD, behaves in our series, as in24 others, as an independent risk factor for death, a relevant fact, since it is a factor of modifiable risk and depends directly on the coordination between the transplant nephrologist and the dialysis nephrologist. Although in our study, the percentage of patients who started HD through a central venous catheter is lower than that of others,26,27 it is known that initiating HD through the central venous catheter increases the risk of sepsis and favors underdialysis. In PD, the use of a catheter without the appropriate maturation period increases the incidence of peritoneal fluid leaks, catheter exit site infections and, consequently, peritonitis, all of this limiting the possibility of achieving ultrafiltration targets (UF) and dialysis dose (Kt/V).13

With all this, given the irreversible deterioration of the graft function, the referral of the patient to pre-dialysis consultations should be done early to guarantee the creation of an AVF or the implantation of a PD catheter. In this way, we would be ensuring a mature access at the time when the patient must start dialysis and we would avoid a large number of complications. Coordination between the transplant unit and the different areas in charge of creating dialysis access is essential at this time, as Oppenheimer et al.28

The overall rate of hospital admissions in our series was similar to that reported in other cohorts of patients.22 The main reason for entering the study, like other publications,29 was the cardiovascular cause, followed by the infectious one. PD patients showed a lower rate of hospital admissions than patients in HD, and the like has been described in other series of patients transplanted.30 In the PD group there was a higher percentage of admissions due to cardiovascular causes than in the HD group, and the latter presented a higher number of admissions due to graft intolerance than the PD group. In PD, the greater dependence on RRF to maintain adequate dialysis, the greater difficulty in maintaining said RRF in the transplanted graft, and the high peritoneal transport that has been described in these people31,32 could have played a fundamental role in the difficulty in volume management and thus justify the high rate of admissions for cardiovascular events, especially heart failure, compared to HD patients.

The PD group presented a lower incidence of graft intolerance and a reduced time to its appearance, but the absence of the analysis of competitive events in this section could make data interpretation difficult. We think that the lower incidence of graft intolerance in the PD group could be related to the maintenance of low doses of immunosuppression, with the intention of preserving RRF in these patients.33 On the contrary, in HD, the greater rapidity of loss of residual diuresis and the lesser dependence on it for adequate dialysis could encourage a more rapid withdrawal of immunosuppression with the risk of intolerance to the secondary graft. Unfortunately, we do not have data on the management of immunosuppression in our work to be able to confirm this hypothesis.

In our series, as in other publications,34–36 half of the patients who started dialysis after graft failure were included again on the waiting list and 15.4% received a subsequent transplant. The mean time to re-transplantation was 26 ± 24 months (median 22 months), similar to that reported by Portolés and Marcén in their publications on transplant patients starting dialysis.22,37 In multivariate analysis, younger age of the patient proved to be the only variable that predicted access to a new transplant, as occurs in the other series.7 We do not have data regarding immunological factors such as the rate of sensitization to human leukocyte antigens (HLA) that could have a significant influence on the possibility of retransplantation.

One of the fundamental limitations of the study has been the diverse origin of the patients. The PD group was obtained from different hospitals, while all the subjects in the HD group belonged to a single hospital. This diversity in the clinical action protocols regarding fundamental aspects such as the time to start dialysis, the admission criteria, the management of the graft intolerance syndrome or the availability of PD could increase the biases of our study. Another important limitation has been the difference in survival between the two groups. The mean follow-up in the PD group was only 18 months compared to 35 in the HD group. Therefore, with the results of our study, we can affirm that PD is beneficial over HD in the first year and a half after starting dialysis.

With the aforementioned limitations, this is the first study at the national level that has shown that PD compared to HD is associated with better survival in patients who start dialysis after graft failure, as well as a lower rate of hospitalization during the first year and a half after the start of the procedure. This should make us rethink the current inertia of referring these patients to HD without considering PD. We need prospective studies to confirm this result, as well as others to understand the role that maintenance of immunosuppression plays on residual renal function in transplant patients after graft failure.

This work has not received any type of funding.