Background

Diabetes has become the most common cause of end-stage renal disease (ESRD) treated by renal replacement therapy (RRT) in most countries around the world; it accounts for 45%, 23%, and 44% of incident cases of RRT requirement in North America [1], Europe [2], and Japan [3], respectively.

Peritoneal dialysis (PD) is valuable for patients newly requiring RRT due to the preservation of residual renal function (RRF), higher quality of life, and hemodynamic stability in comparison with hemodialysis (HD) [4]. Approximately 196,000 patients worldwide underwent PD in 2008, representing 11% of the dialysis population [5]. However, diabetic patients were less likely than non-diabetic patients to receive PD as first RRT in North America (9.0% vs. 10.1%, respectively) [1], Europe (14% vs. 15%, respectively) [2], and Japan (4.9% vs. 6.6%, respectively) [3]. Possible reasons for this therapeutic preference include anxiety regarding worsening of glycemic control, higher prevalence of PD-associated peritonitis, overhydration and rapid RRF decline due to proteinuria and inflammation, and technical problems due to visual disorders and peripheral neuropathy. Additionally, several factors including demographic, medical, social, pre-ESRD, and geographic factors are associated with the selection of dialysis modality [6].

Several reports have provided conflicting results regarding patient survival. Ideally, randomized controlled trials (RCTs) are needed to clarify the survival advantage of PD or HD. Although one such RCT has been conducted, the number of included patients was small and no analysis stratified by diabetes was performed [7]. Couchoud et al. [8] conducted a systematic review based on 25 observational studies published until February 2014, which included 821,783 diabetic patients receiving HD and 106,790 such patients receiving PD. Due to the heterogeneity of study designs and PD and HD practices, they could not provide an evidence-based argument in favor or against the use of either modality as the first dialysis treatment for diabetic patients.

As several advances have been made in therapy for diabetic patients undergoing PD, we conducted a systematic review based on studies published after February 2014 to examine whether incident PD or HD is advantageous with regard to patient survival and other clinical outcomes among patients with diabetes.

Methods

This systematic review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [9] (Additional file 1). The review was performed using a predefined protocol published in PROSPERO (CRD42018104258). No ethical approval was required because this study did not involve the use of confidential personal data or patient interventions.

The MEDLINE, EMBASE, and CENTRAL databases were searched to identify articles published from February 2014 to August 2017 with no language, time, or methodological restriction using focused and highly sensitive search strategies (Additional file 2). We included any type of trial comparing any type of PD (i.e., automated PD or continuous ambulatory PD) with any type of HD (i.e., conventional HD, hemofiltration, hemodiafiltration, daily HD) as first RRT in diabetic patients with ESRD.

Outcomes of interest were all-cause mortality; urinary volume (RRF); major morbid events, including cardiovascular disease (CVD) and infectious disease; and glycemic control.

Studies were excluded (i) if outcomes were not reported separately for diabetic patients, (ii) if they did not provide longitudinal data on any of the abovementioned outcomes, or (iii) if they did not directly compare HD and PD. Case reports, reviews, editorials, and letters were also excluded, although they were screened as potential sources of additional references.

Four reviewers (YM, CH, HI, and KW) independently reviewed the title and abstract of each retrieved publication, and articles were selected for full-text review. The same four reviewers independently screened the reference lists of articles selected for full-text review. The inclusion of full-text articles was finalized after consultation with a fifth reviewer (HT). All disagreements were resolved by consensus.

We used forest plot for comparison of all-cause mortality in diabetic patients receiving incident PD and those receiving incident HD. For this analysis, only publications reporting hazard ratios (HRs) with 95% confidence interval (CIs) for all enrolled diabetic patients were included. We did not conduct a meta-analysis because of the high risk of bias in each study.

Quality assessment

We used the key criteria for limitations of observational studies developed by the GRADE working group (handbook for grading the quality of evidence and the strength of recommendations using the GRADE approach, http://gdt.guidelinedevelopment.org/app/handbook/handbook.html#h.m9385o5z3li7). Two authors of this review independently assessed the items listed below. Disagreements regarding the risk of bias were resolved by consultation with other review authors:

  1. 1.

    Failure to develop and apply appropriate eligibility criteria (inclusion of a control population)

    • Under- or overmatching in case-control studies

    • Selection of exposed and unexposed groups from different populations in cohort studies

  2. 2.

    Flawed measurement of both exposure and outcome

    • Differences in the measurement of exposure (e.g., recall bias in case-control studies)

    • Differential surveillance for outcome in exposed and unexposed groups in cohort studies

  3. 3.

    Failure to adequately control confounding factors

    • Failure to accurately measure all known prognostic factors

    • Failure to match prognostic factors and/or adjust statistical analysis

  4. 4.

    Incomplete or inadequately short follow-up

    • Especially for prospective cohort studies, both groups should be followed for the same amount of time.

Results

Study selection

Figure 1 summarizes the search strategy that was used. The initial search yielded 766 articles, of which 620 articles were excluded after review of the titles and abstracts. A total of 146 articles underwent full-length review, and 16 studies were included in the qualitative analysis.

Fig. 1
figure 1

PRISMA flow diagram showing study selection

Characteristics of studies

The characteristics of the 16 studies are summarized in Table 1 [10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]. All studies were observational and were conducted using registry or cohort databases. One study included only diabetic patients receiving incident PD [14]; the percentages of diabetic patients ranged from 10.3 [12] to 70.3% [16] in the other studies. The total numbers of diabetic patients included were 50,298 receiving PD and 71,532 receiving HD. Eight studies were from Asia [10, 11, 14, 16, 18, 19, 21, 23], three were from Australia and New Zealand [15, 20, 22], three were from Europe [17, 24, 25], one was from North America [13], and one was from South Africa [12]. Several studies were based on the same registry or cohort databases, such as the Australia and New Zealand Dialysis and Transplant Registry (ANZDATA) [15, 20, 22], the Korean Health Insurance Review and Assessment Service (HIRA) database [10, 19, 21, 23], and the National Health Insurance Research Database (NHIRD) of Taiwan [11, 16].

Table 1 Characteristics of studies

Risk of bias

Table 2 shows the quality of the studies included in the analysis. As all were retrospective observational studies, the risk of bias, especially with regard to the failure to adequately control confounding factors, was high.

Table 2 Quality assessment

All-cause mortality

Fifteen studies [10, 12,13,14,15,16,17,18,19,20,21,22,23,24,25] investigated all-cause mortality among diabetic patients undergoing PD and HD (Table 3). Several studies investigated differences in survival between diabetic and non-diabetic patients in subgroup analyses. Figure 2 shows a forest plot comparing all-cause mortality between diabetic patients receiving incident PD and those receiving incident HD. For this analysis, only publications reporting HRs with 95% CIs for whole populations of enrolled diabetic patients were included. We did not conduct a meta-analysis of all-cause mortality data due to the high risk of bias in each study.

Table 3 Mortality of diabetic PD patients
Fig. 2
figure 2

Forest plot comparing all-cause mortality in diabetic patients receiving incident PD and those receiving incident HD. For this analysis, only publications reporting hazard ratios (HRs) with 95% confidence interval (CIs) for all enrolled diabetic patients were included. We did not conduct a meta-analysis because of the high risk of bias in each study. The Marshall 2014 and Marshall 2016 studies used the same database, i.e., the Australia and New Zealand Dialysis and Transplant Registry (ANZDATA). In Marshall 2014, diabetic patients were divided into those with type 1 and those with type 2 diabetes, and we extracted the HR and 95% CI from the type 2 diabetes group. The Kim 2014, Kim 2015, and Kim 2017 studies were conducted using the same database, i.e., the Korean Health Insurance Review and Assessment Service (HIRA) database. In Nesrallah 2016, the control patients received home HD rather than conventional HD

Differences in mortality favoring HD were observed in nine studies [10, 12, 13, 15, 16, 19, 21,22,23]. Marshall et al. [15] reported an HR of 1.17 (95% CI 1.11–1.25) for death among patients receiving incident PD relative to those receiving incident HD, based on ANZDATA data. Among patients receiving incident PD, they found that the risk of death was higher for elderly diabetic patients [22]. Wang et al. [16] reported an HR of 1.22 (95% CI 1.05–1.43) for patients receiving incident PD in a propensity score-matched cohort, based on NHIRD data. Based on HIRA data, Kim et al. [10] reported that the HR for death, calculated by multivariate Cox proportional hazards regression, among patients undergoing PD was 1.27 (95% CI 1.19–1.35), and Kim et al. [19] reported that the adjusted relative risk of death, calculated by multivariate Poisson regression, was 1.29 (95% CI 1.19–1.40). Ryu et al. [21] and Kim et al. [23] reported similar results, and they found that the risk of death among patients receiving incident PD was high for elderly diabetic patients. Tamayo Isla et al. [12] reported HRs for diabetic patients receiving PD and HD of 4.99 (95% CI 2.13–11.71) and 1.02 (95% CI 0.43–2.50), respectively, in comparison with non-diabetic patients receiving HD among 340 patients receiving incident dialysis, based on data from a South African single-center database. Nesrallah et al. [13] reported an unadjusted HR of 1.16 (95% CI 0.99–1.39), based on data for a propensity score-matched cohort of 5336 patients receiving incident dialysis extracted from the US Renal Data System.

Differences in mortality favoring PD were observed in two studies [14, 24]. Lee et al. [14] extracted data on 902 diabetic patients who started dialysis between 2008 and 2013 from a nationwide prospective cohort in Korea, and found that PD was associated with a lower risk of death than was HD, not only in the whole cohort (HR 0.65, 95% CI 0.47–0.90), but also in the group with available hemoglobin A1c (HbA1c) data (HR 0.64, 95% CI 0.46–0.91). In addition, they found that PD had a significant survival advantage over HD in patients with HbA1c < 8.0% (HR 0.59, 95% CI 0.37–0.94), but not in the poor glycemic control group (HbA1c ≥ 8.0%: HR 1.21, 95% CI 0.46–2.76). Heaf and Wehberg [24] extracted data on 12,095 diabetic patients who started dialysis between 1990 and 2010 from the Danish Nephrology Registry, and found that PD was associated with a lower risk of death compared with HD, with a more pronounced difference in recent years (1990–1999: HR 0.97, 95% CI 0.84–1.12; 2000–2010: HR 0.86, 95% CI 0.75–0.97).

Statistical analyses in all of the abovementioned studies were conducted using only an intention-to-treat approach. However, Waldum-Grevbo et al. [17] examined survival using both intention-to-treat and as-treated analyses with data extracted from the Norwegian Renal Registry. They reported that the 2-year mortality rate tended to be higher (intention-to-treat analysis: HR 1.22, 95% CI 0.80–1.86; as-treated analysis: HR 1.20, 95% CI 0.76–1.91), whereas the 5-year mortality rate tended to be lower (intention-to-treat analysis: HR 0.90, 95% CI 0.65–1.25; as-treated analysis: HR 0.99, 95% CI 0.69–1.42) in diabetic patients receiving PD compared with those receiving HD.

Major morbid events, including cardiovascular disease and infectious disease

Two studies investigated the effects of dialysis modality on new-onset CVD among diabetic patients [11, 19]. Shen et al. [11] reported that the risk of new-onset atrial fibrillation was higher in the incident PD group (HR 1.76, 95% CI 1.13–2.75 vs. controls without ESRD) than in the incident HD group (HR 1.52, 95% CI 1.33–1.75vs. controls without ESRD) among 7844 patients with diabetes. Kim et al. [19] reported that the risk of developing major adverse cardiac and cerebrovascular events, including all-cause mortality, non-fatal acute myocardial infarction, target vessel revascularization including percutaneous coronary intervention and coronary artery bypass grafting, and non-fatal stroke, was higher in patients receiving incident PD than in those receiving incident HD among 14,812 patients with diabetes (HR 1.15, 95% CI 1.07–1.24).

None of the 16 studies investigated the effects of dialysis modality on urinary volume or RRF, infectious disease, or glycemic control among diabetic patients.

The findings are summarized in Table 4.

Table 4 Summary of findings

Discussion

The present systematic review was performed to examine whether PD or HD as the first RRT for diabetic patients with ESRD improved clinical outcomes. The chief findings were that differences in mortality favoring HD were observed in nine studies, whereas those favoring PD were observed in two studies. Although the risk of death tended to be higher among patients receiving incident PD than among those receiving incident HD, we could not confirm the superiority of PD or HD because of conflicting results and a high risk of bias in the included studies, especially with regard to the failure to adequately control confounding factors. These results are similar to those of a previous systematic review conducted by Couchoud et al. [8], which included 25 observational studies published until February 2014.

We conducted this systematic review on the assumption that improved outcomes were expected among patients with diabetes undergoing incident PD because of advances in the management of these patients, including the use of icodextrin-containing PD solutions and dipeptidyl peptidase-4 (DPP-4) inhibitors.

RRF is a strong predictor of patient survival [26, 27] and is preserved better among patients receiving PD than in those receiving HD [28]. Among patients undergoing PD, the rate of RRF loss is higher in diabetic than in non-diabetic patients [28, 29]. Interestingly, fluid overload and impaired RRF are closely linked. Udo et al. [30] reported that diabetic patients electively starting PD showed greater extracellular water retention 6–10 weeks after starting PD than did non-diabetic patients, despite similar peritoneal function, as determined by the peritoneal equilibration test. In addition, Kim et al. [31] reported that an increase in body weight during the first year and diabetes were associated independently with a rapid decline of RRF. Icodextrin-containing solutions improve peritoneal ultrafiltration and mitigate uncontrolled fluid overload. Icodextrin became commercially available in 1997 in Europe, in 2001 in Korea, in 2002 in Australia and New Zealand, in 2003 in the USA and Japan, and in 2004 in Taiwan. In a recent systematic review, the use of icodextrin was shown to uniformly result in improved peritoneal ultrafiltration compared with glucose exchange, especially among patients with higher peritoneal transport characteristics, and to reduce reported episodes of uncontrolled fluid overload [32]. However, icodextrin had no appreciable impact on RRF, technical failure, or death [32].

Several advances have been made in the treatment of diabetes, including the development of DPP-4 inhibitors. These drugs, which were approved for clinical use in 2006, provide an effective therapeutic option without the drawback of inducing hypoglycemia and can be used safely in patients receiving dialysis. DPP-4 inhibitor use was found to significantly improve the HbA1c level and hyperglycemia in patients receiving PD [33, 34]. Glycemic control is known to influence clinical outcomes, including mortality, in patients with chronic kidney disease who are and are not receiving dialysis. Duong et al. [35] reported that poor glycemic control (HbA1c ≥ 8% or serum glucose ≥ 300 mg/dl) was associated with decreased survival in a population of 2798 diabetic patients receiving PD. Furthermore, Lee et al. [14] reported a significant survival advantage of PD in patients with HbA1c < 8.0%, but no significant difference in the survival rate according to dialysis modality (PD or HD) in the poor glycemic control group (HbA1c ≥ 8.0%). Unfortunately, the details of diabetes treatment were not provided in all cited studies.

In this systematic review, the results differed among studies due to the heterogeneity of dialysis practices; diabetes treatments; patient backgrounds, including educational and social insurance statuses; and the timing of referral to a nephrologist. Unfortunately, these factors were not clarified in all of the included studies. In addition, no report from Japan was included. We recently reported that technical and patient survival did not differ between diabetic and non-diabetic patients receiving incident PD and that the presence of diabetes did not affect either survival measure in multivariate analyses [36].

Conclusions

In the present systematic review, the risk of death tended to be higher among diabetic patients with ESRD receiving incident PD as RRT than among those receiving incident HD. However, we could not determine definitively whether PD or HD was superior with regard to patient outcomes because of the high risk of bias and the diversity of management of diabetic patients undergoing dialysis. Further studies are needed to clarify the advantages of RRT with PD and HD in diabetic patients with ESRD.