Dialysis Modality and Mortality in the Elderly: A Meta-Analysis

Abstract

Background and objectives

Identifying the appropriate choice between hemodialysis (HD) and peritoneal dialysis (PD) is an unresolved issue in elderly patients with ESRD, who are at high risk for death but have a low chance of receiving kidney transplantation.

Design, setting, participants, & measurements

Data on 13,065 incident dialysis Korean patients (age≥65 years) receiving HD (n=10,675) or PD (n=2390) were obtained from the Korean Health Insurance dataset. Multiple statistical approaches, including the multivariate Cox model, were used to compare mortality between Korean patients receiving PD and those receiving HD. Subsequently, meta-analysis of previous comparison studies (published since the year 2000; population-based studies) and the Korean dataset was performed.

Results

During a mean duration of 1.8±1.3 years (maximum of 5 years), the Korean PD group had a higher mortality rate than the Korean HD group (hazard ratio [HR], 1.20 [95% confidence interval (95% CI), 1.13 to 1.28]; P<0.001 by multivariate Cox model). The discrepancy between the two modalities was greater in the presence of certain conditions, such as diabetes mellitus or longer dialysis duration. In the meta-analysis, 15 studies involving >631,421 elderly patients were reviewed. Compared with HD, the pooled HR with PD was 1.10 (95% CI, 1.01 to 1.20). When the meta-analysis was stratified by confounding factors, the survival benefit from HD was particularly strong in subgroups that had diabetes mellitus, had long dialysis duration (>1 year), or contained cohorts starting dialysis in the 1990s.

Conclusions

A meta-analysis that included results in Korean patients suggests a higher risk for death in elderly patients receiving PD than in those receiving HD.

Introduction

Patients with ESRD are at higher risk of morbidity and mortality (1), and, thus, their global social and economic burden is high (2). Although patients with ESRD have received intensive attention for this reason, the prevalence of ESRD has not decreased; rather, it has continued to grow at a rate of approximately 3% per year (3). This trend is much more prominent in elderly patients, who make up to half of the dialysis population (4). The elderly have a higher prevalence of comorbidities, which increases the burden of dialysis via a substantially higher mortality rate compared with that in their younger counterparts (5,6). An additional problem is a lower chance of receiving transplantation in elderly patients, even though kidney transplantation provides a survival advantage over dialysis (7). Accordingly, it is necessary to identify risk factors for death in this population before and after initiation of dialysis.

Several observational studies have previously compared hemodialysis (HD) and peritoneal dialysis (PD) in the ESRD population. To date, only one randomized controlled trial attempted to compare the mortality risk between the two dialysis modalities, but it was halted early because of insufficient enrollment (8). According to those studies, the survival benefit of one modality over the other has not yet been determined; results across studies conflict. This is attributable to the different baseline characteristics, which lead to interactions between dialysis modality and mortality outcome. Therefore, results were analyzed after stratification based on several factors, such as age, diabetes mellitus status, and dialysis duration. In particular, age is an important factor affecting the association between dialysis modality and mortality because elderly patients with ESRD significantly differ from their younger counterparts with respect to the following issues: the presence of several comorbidities; short dialysis durations due to high mortality; and, concomitantly, a low chance of receiving kidney transplantation or switching modalities.

A few studies that have directly focused on the elderly ESRD population to determine which modality is associated with lower mortality risk (9–11). Other researchers have indirectly shown results among elderly patients in a subgroup analysis (12–29). Yet again, the issue of the selection of dialysis modality in elderly patients remains unresolved. Herein, we sought to compare mortality risk between HD and PD in an elderly subset by using both a Korean dataset for the first time and a meta-analysis of observational studies (including the Korean results); both sources of data were used in order to pool the results and to avoid a cohort-specific bias.

Materials and Methods

Korean Participants and Data Collection

The study protocol complies with the Declaration of Helsinki and received full approval from the institutional review board at the Seoul National University Hospital (no. 1405–032–576). Data on elderly Koreans initiating dialysis were obtained from the Korean Health Insurance database. Data from all elderly Koreans could be reviewed because Koreans must register with the national insurance system, which is composed of national health insurance and medical aid. In total, 35,422 patients with ESRD initiated dialysis between January 2005 and December 2008. From this group, elderly patients age 65 years or older (n=13,065 [36.9%]) were included in the present analysis.

Sociodemographic covariates, including age, sex, type of insurance, and scale of the medical center at which patients were seen (i.e., clinic, hospital, or general hospital), were collected after assigning a unique deidentified number to each patient. Comorbidities, such as diabetes mellitus, myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, connective tissue disease, peptic ulcer disease, liver disease, hemiplegia, and cancer, were identified by screening medical history data from the year leading up to the initiation of dialysis therapy based on the International Classification of Diseases, 10th revision (30). Through use of these covariates, the Charlson comorbidity index score for each patient was calculated (31). All patients were divided into two groups (HD and PD). If patients switched from one modality to the other within 1 month of initiating dialysis, those patients were classified and followed as a changed-modality group as long as no patients died in that month. We also performed a comparison between two modality groups, wherein the dialysis modality was determined on the 90th day after initiating dialysis, similar to an intention-to-treat analysis. In this case, patients who died within 90 days of initiating dialysis were excluded (n=1736). There were no missing data in the Korean dataset.

Study patients were followed until kidney transplantation, death, or December 2009. If the dialysis modality changed, the patient was censored. Mortality data were obtained from the Certificate Database (i.e., listing of reasons for changes in eligibility for national health insurance or medical aid, including death or emigration), as well as through the National Health Insurance Claims Database.

Study Search and Selection for Meta-analysis

Online databases were searched independently by two authors (S.S.H. and D.K.K.) using the following search terms: PubMed: dialysis modality AND (mortality OR survival); Cochrane Library: dialysis modality AND (mortality OR survival); Google Scholar: dialysis modality AND (mortality OR survival) AND elderly. Publication date (after January 1, 2000) and language (English only) restrictions were applied. The authors reviewed titles and abstracts of papers identified by the searches. Searches were re-run using additional search terms identified from papers considered relevant to the review.

Observational cohort studies were eligible for inclusion if they reported an association between dialysis modality and mortality. Studies reporting on elderly patients and those including the elderly as a subgroup were eligible for inclusion; however, the age definition of elderly was allowed to differ among studies. In cases where several publications used the same or overlapping datasets, we included the most thorough or representative information on mortality risk associated with dialysis modality. Final study selection was based on a consensus between the reviewers or, in the case of disagreement, arbitration by a third reviewer (Y.S.K.).

Data Extraction

The following study characteristics were extracted: authors, study cohort, enrollment period, sample size, age range (i.e., elderly definition), follow-up duration, result of comparison (PD versus HD), variable for subgroup analysis, and method of ascertainment. We extracted the adjusted hazard ratios (HRs) for reported mortality. Most studies reported results as HRs, but some reported relative risks. However, regardless of the reported statistic, the method used was the same: Cox proportional hazard model. Therefore, we used HR and relative risk together in the meta-analysis. If studies reported follow-ups at different time points, the longest duration was selected for calculating the pooled HR. For subset analysis of studies that provided relevant data, we pooled within-study HR estimates (e.g., patients with and without diabetes or comorbidities).

Statistical Analyses

Statistical analyses and any associated references are available in the online Supplemental Materials.

Results

Mortality Comparison between HD and PD Dialysis in Korean Population

Table 1 shows the baseline characteristics of the study participants. The mean age was 72.2 years. The number (proportion) of participants in each age category was as follows: 65–69 years, n=4939 (37.8%); 70–79 years, n=6650 (50.9%); 80–89 years, n=1418 (10.9%); and ≥90 years, n=58 (0.4%). All of the patients were of Asian descent. The study participants were followed for a mean duration of 1.8±1.3 years (maximum, 5 years). Baseline characteristics for additional analyses are shown in the online Supplemental Materials.

Table 1.

Baseline characteristics of Korean participants

Characteristic	Total (n=13,065)	HD (n=10,675)	PD (n=2390)	P Value
Age (yr)	72.2±5.53	72.4±5.60	71.4±5.12	<0.001
Men (%)	54.4	54.8	52.6	0.05
Health security system (%)				0.61
National health insurance	88.8	88.8	89.1
Medical aid	11.2	11.2	10.9
Medical center classification (%)				<0.001
General hospital	93.8	92.6	99.2
Hospital	2.6	3.0	0.6
Clinic	3.6	4.4	0.1
Medical comorbidities (%)
Diabetes mellitus	53.7	53.1	56.5	0.002
Myocardial infarction	6.0	5.2	9.6	<0.001
Congestive heart failure	19.7	18.6	24.4	<0.001
Peripheral vascular disease	8.4	8.5	8.0	0.49
Cerebrovascular disease	19.5	20.0	17.4	0.003
Dementia	3.5	3.7	2.8	0.03
Chronic pulmonary disease	23.5	23.9	21.1	0.004
Connective tissue disease	3.3	3.4	3.2	0.70
Peptic ulcer disease	17.5	17.8	16.4	0.10
Liver disease	9.1	9.4	7.6	<0.01
Hemiplegia	2.3	2.4	1.6	0.02
Cancer	9.6	10.6	4.9	<0.001
Charlson comorbidity index score	2.7±2.01	2.7±2.04	2.6±1.87	<0.01

Statistical differences between dialysis modalities were calculated. Values expressed with a plus/minus sign are the mean±SD. HD, hemodialysis; PD, peritoneal dialysis.

Figure 1 shows Kaplan–Meier survival curves for all patients by dialysis modality. There was a crossover point regarding survival difference between patients receiving HD and those receiving PD. In the first year, the survival rate in the PD group appeared to be commensurate with that in the HD group; after 1 year, however, the survival rate in the PD group seemed to be lower than that in the HD group. After adjustment for all the covariates, the PD group had a higher HR for mortality than the HD group throughout the overall study period (Table 2). The results of other statistical approaches, such as adjustment for propensity score, the one-to-one matching model, and the marginal structural model, were similar to those mentioned above (see the online Supplemental Materials for detail). The survival curve from the multivariate Cox model is shown in Supplemental Figure 1. When the statistical significance and HRs were reviewed, the survival discrepancy between HD and PD was more prominent in the presence of diabetes mellitus than in its absence (Supplemental Materials). Although a total of 465 patients (3.5%) who switched modality were excluded, the HR for mortality was also higher in the PD group than in the HD group (HR, 1.30 [95% confidence interval (95% CI), 1.22 to 1.39]; P<0.001).

Figure 1.

Kaplan–Meier survival curves of tow dialysis groups. Blue and red lines represent HD and PD, respectively. HD, hemodialysis; PD, peritoneal dialysis.

Table 2.

Hazard ratios for mortality in peritoneal dialysis group compared with hemodialysis group

Period	Adjusted for Covariates		Adjustd for Propensity Score		Marginal Structural Model		1:1 Matching
	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value
<3 mo	0.94 (0.82 to 1.07)	0.35	0.94 (0.82 to 1.07)	0.32	1.03 (0.98 to 1.09)	0.22	1.02 (0.87 to 1.21)	0.80
≥3 to <6 mo	0.98 (0.84 to 1.15)	0.80	0.98 (0.83 to 1.15)	0.76	0.92 (0.87 to 0.99)	0.02	0.98 (0.81 to 1.20)	0.87
≥6 to <9 mo	1.20 (0.99 to 1.46)	0.07	1.19 (0.98 to 1.45)	0.08	0.98 (0.90 to 1.06)	0.56	1.15 (0.90 to 1.48)	0.26
≥9 to <12 mo	1.37 (1.12 to 1.67)	0.003	1.36 (1.11 to 1.67)	0.003	1.17 (1.07 to 1.27)	<0.001	1.42 (1.09 to 1.86)	0.010
≥1 to <2 yr	1.42 (1.24 to 1.63)	<0.001	1.41 (1.23 to 1.62)	<0.001	1.21 (1.14 to 1.28)	<0.001	1.46 (1.23 to 1.74)	<0.001
≥2 to <3 yr	1.67 (1.39 to 2.02)	<0.001	1.65 (1.37 to 1.99)	<0.001	1.51 (1.39 to 1.63)	<0.001	1.63 (1.29 to 2.05)	<0.001
Overall	1.20 (1.13 to 1.28)	<0.001	1.19 (1.12 to 1.27)	<0.001	1.09 (1.06 to 1.12)	<0.001	1.25 (1.15 to 1.35)	<0.001

HR, hazard ratio; 95% CI, 95% confidence interval.

Study Selection for Meta-Analysis

Through use of the search strategy described earlier, 941 unique citations were initially retrieved. Of these, 20 articles were considered of interest, and the full text for these studies was retrieved for detailed evaluation (Table 3). The results were obtained from national or multicenter databases, which primarily included white persons. All studies except one conducted additional analyses with stratification by the presence of diabetes mellitus, dialysis duration, comorbidities, or dialysis start time. The main statistical approach used was the multivariate Cox model. Some studies used multiple methods, as in the present study. In these cases, the results from the multivariate Cox model were used for the meta-analysis. Of these 20 studies, six were excluded for the following reasons: Three did not show the HRs for mortality and three used the same cohort. Certain studies did not show overall HRs for mortality but described HRs in subgroups, such as patients with diabetes mellitus. In these cases, a pooled HR, which was calculated from subgroup data, was used instead. Study selection is summarized in Figure 2.

Table 3.

Studies on the association between dialysis modality and mortality in elderly patients, published after 2000

Study (Reference)	Cohort	Enrollment Period (Year)	Elderly Patients (HD: PD) (n)	Definition of Elderly (yr)	Follow-up Duration	HR (95% CI)	Subgroup	Analysis
Heaf et al. (12)a	Danish Terminal Uremia Register	1990–1999	No data	≥56	Maximum, 10 yr	0.85 (0.76 to 0.94)	DM	Adjustment of covariates
Winkelmayer et al. (10)	Medicare and Medicaid participants in NJ	1991–1996	2503 (1966: 537)	≥66	Maximum, 1 yr	1.24 (1.09 to 1.41)	Dialysis duration	Adjustment of propensity score
Termorshuizen et al. (13)	Netherlands Cooperative Study on the Adequacy of Dialysis	1997–no data	626 (466: 160)	≥60	Until September 2002	2.01 (1.09 to 2.92)b	DM, dialysis duration	Adjustment of covariates
Vonesh et al. (14)	Centers for Medicare & Medicaid Services	1995–2000	203,578 (186,934: 166,44)	≥65	Maximum, 3 yr	1.08 (0.92 to 1.24)b	DM, comorbidity	Adjustment of covariates
Jaar et al. (15)	81 facilities in United States (CHOICE)	1995–1998	377 (no data)	≥65	Mean, 2.4 yr	Multivariate: 1.66 (0.93 to 2.97); propensity: 1.67 (0.96 to 2.90)	–	Adjustment of covariates; propensity-matched
Mircescu et al. (16)	14 Romanian facilities	1995–2001	No data	≥65	Until April 2002	0.74 (0.41 to 1.08)b	DM, comorbidity	Adjustment of covariates
Liem et al. (17)	Dutch ESRD registry (RENINE)	1987–2002	No data	≥70	Mean, 2.38 yr	1.27 (1.02 to 1.52)b	DM, dialysis duration	Adjustment of covariates
Couchoud et al. (11)	French REIN registry	2002–2005	3512 (2880: 632)	≥75	Mean, 8.6 mo	1.1 (0.9 to 1.3)	DM, CHF	Adjustment of covariates
Huang et al. (18)	Taiwan Renal Registry	1995–2002	32,255 (30,662: 13,60)	≥56	Maximum, 10 yr	1.65 (0.98 to 2.31)b	DM, comorbidity	Adjustment of covariates
Sanabria et al. (19)c	Dialysis Outcomes in Colombia	2001–2003	No data	≥66	Until December 2005	Similar mortality (only figure)	DM	Adjustment of covariates
McDonald et al. (20)c	ANZDATA registry	1991–2005	No data	≥60	Until December 2005	Difference (<1 yr and others) (only figure)	DM	Adjustment of covariates
Weinhandl et al. (21)c	Centers for Medicare & Medicaid Services	2003	4255 (no data)	≥65	Mean, 2.3 yr	Difference (<1 yr and others) (only figure)	Dialysis duration	Propensity-matched
Mehrotra et al. (22)d	USRDS	1996–2004	332,552 (310,010: 22,542)	≥65	Until September 2007	1.16 (1.09 to 1.24)b	DM, comorbidity, dialysis-starting time	Marginal structural model
Sens et al. (23)a	Patients with CHF from French REIN Registry	2002–2008	2306 (no data)	≥75	Until December 2008	1.46 (1.26 to 1.66)b	DM	Adjustment of covariates
van de Luijtgaarden et al. (24)	ERA-EDTA Registry	1998–2006	5993 (5189: 804)	≥ 70	Mean, 1.6 years	0.87 (0.76 to 0.99)	DM, comorbidity, sex	Adjustment of covariates
Chang et al. (25)	Taiwan National Health Insurance	1997–2006	1309 (680: 629)	≥65	Until December 2006	1.20 (1.05 to 1.41)	Comorbidity, dialysis-starting time	Propensity-matched
Yeates et al. (26)	Canadian Organ Replacement Register	1991–2004	23,411 (17,400: 6011)	≥ 65	Until Dec 2007	1.12 (0.98 to 1.25)b	DM, sex	Adjustment of covariates
Lukowsky et al. (27)a,e	DaVita dialysis clinics	2001–2004	11,445	≥66	2 yr	0.27 (0.12 to 0.61)	Dialysis duration	Marginal structural model
Heaf et al. (28)	Danish Nephrology Registry	1990–2010	5679 (4335: 1344)	≥65	Until January 2011	0.89 (0.83 to 0.95)	DM, dialysis-starting time	Adjustment of covariates
Mircescu et al. (29)	Romanian renal registry	2008–2011	No data	≥61	Until December 2012	1.01 (0.87 to 1.20)b	DM	Adjustment of covariates

Of the above 20 studies, six were excluded from the meta-analysis. HD, hemodialysis; PD, peritoneal dialysis; HR, hazard ratio; 95% CI, 95% confidence interval; DM, diabetes mellitus; CHOICE, Choices for Healthy Outcomes in Caring for ESRD Patients; RENINE, Dutch End-Stage Renal Disease Registry; CHF, congestive heart failure; REIN, Renal Epidemiology and Information Network; ANZDATA, Australia and New Zealand Dialysis and Transplant Registry; USRDS, US Renal Data System; ERA-EDTA, European Renal Association–European Dialysis and Transplant Association.

^aStudy excluded because of duplicate cohort.

^bThe pooled HR is obtained from subgroup results.

^cStudy excluded for not reporting HRs.

^dIn a subgroup analysis, this duplicate study was not used because it did not have a subgroup result.

^eThis study, the duplicate of the study noted in footnote d, was used in a subgroup analysis.

Figure 2.

Flow diagram of study selection for meta-analysis. HD, hemodialysis; PD, peritoneal dialysis; HR, hazard ratio; RR, relative risk.

Main Results of Meta-Analysis

A meta-analysis of the 15 unique studies was performed; these studies reported on >631,421 elderly patients with ESRD. Three studies did not report the number of elderly patients. Figure 3 shows the combined results obtained from all of these cohorts. Of the 15 studies, seven, including the present study, reported the overall HR in elderly patients. However, because the other eight studies described only the subgroup results, pooled HRs were calculated. The HRs for mortality (i.e., PD versus HD) of the former and the latter cohorts were 1.09 (95% CI, 0.94 to 1.24) and 1.12 (95% CI, 1.01 to 1.22), respectively. The pooled HR from all cohorts was 1.10 (95% CI, 1.01 to 1.20). Visual inspection of the funnel plot and tests for its asymmetry indicated no evidence of publication bias among studies of dialysis modality and mortality (P=0.22 by Egger test; P=0.73 by Begg test) (Supplemental Figure 2).

Figure 3.

Meta-analysis of overall mortality in peritoneal dialysis compared with hemodialysis. ID, identification; ES, effect size; 95% CI, 95% confidence interval; HR, hazard ratio.

Subgroup Meta-Analysis

We further conducted subgroup meta-analyses that considered possible confounding factors, such as diabetes mellitus, dialysis duration (≤1 year and >1 year), and dialysis start time (1990s and 2000s), on the basis of previous studies, a majority of which stratified analyses using these variables (Figure 4). Of the 15 studies, subgroup analyses for presence and absence of diabetes mellitus were possible in nine and 11 studies, respectively; the pooled HR for mortality was 1.26 (95% CI, 1.13 to 1.40) in the diabetes group and 1.10 (95% CI, 1.02 to 1.18) in the nondiabetes group. During the first year after initiating dialysis, there was no difference between PD and HD (HR, 0.95; 95% CI, 0.72 to 1.18). After 1 year, however, the PD group had a higher mortality rate than the HD group (HR, 1.41; 95% CI, 1.20 to 1.61). The pooled HRs, which were obtained from cohorts that started dialysis before and after 2000, showed a similar trend in mortality. However, when the studies were limited to the two studies for which data from both the 1990s and 2000s were available, the HR for PD compared with HD decreased over time (Supplemental Figure 3C). The results of the meta-analysis that excluded studies with only subgroup HRs is shown in Supplemental Figure 3. Additional meta-analysis that included patients aged ≥ 70 years is shown in Supplemental Figure 4. The overall trend was similar to the above results.

Figure 4.

Subgroup meta-analysis of mortality in peritoneal dialysis (PD) and hemodialysis (HD). Analyses were stratified by diabetes mellitus (A), dialysis duration (B), and dialysis start time (C). For dialysis duration, the study by Mehrotra et al. did not describe subgroup results. Therefore, the study by Lukowsky et al., which examined the same cohort, was used instead in this subgroup analysis. ID, identification; ES, effect size; 95% CI, 95% confidence interval; DM, diabetes mellitus.

Discussion

It has not yet been determined whether HD or PD provides a survival advantage to elderly patients with ESRD. The present study addresses this issue in a subset of elderly Korean patients with ESRD. On the basis of results of multiple methodologic approaches, patients who started with PD had higher mortality than those who started with HD. The present meta-analysis of observational cohort studies showed a trend toward lower survival in patients undergoing PD than in those receiving HD. When the subgroup analysis was stratified by possible confounding factors, the decreased survival of the PD group was more prominent in the presence of diabetes mellitus and long dialysis durations (or starting dialysis in the 1990s, most likely).

In research among a wider age range, comparisons of mortality rates between HD and PD patients have been documented in several cohort or national studies with varying results. These controversial findings may be attributable to heterogeneity in baseline characteristics across populations (14). Furthermore, even within a cohort, there were different results based on confounding factors, such as dialysis duration or year in which dialysis began (20,25). Above all, the pattern of clinical practice or socioeconomic status may differ between studies, although this difference cannot be easily considered in the analyses. The likelihood of receiving kidney transplantation is an important factor in clinical practice that influences choices regarding dialysis modality or dose and, eventually, outcomes. However, elderly patients have a lower chance of receiving kidney transplantation than their younger counterparts because of the frequent presence of multiple comorbidities. Accordingly, we focused only on elderly patients, who are not expected to be highly influenced by this practice (in the present study, 13 patients [0.1%] received kidney transplantation). This focus also reduced the variability in the study participants’ ages.

The present study revealed a higher mortality rate in the PD group compared with the HD group. This trend was apparent in diabetic patients and those who had undergone longer dialysis durations. Previously, a few studies directly addressed the question of which modality may be most suitable for elderly patients. One meta-analysis of eight studies reported similar mortality results for both modalities (9). However, all of the reviewed studies were published before 2000, and those patients began dialysis between 1983 and 1998. In this regard, the applicability of that result to current clinical practice is tenuous.

After 2000, two studies were conducted in elderly cohorts. In the New Jersey Medicare and Medicaid cohort, overall mortality was high in PD patients, but this trend was only significant when patients had both diabetes mellitus and a long dialysis duration (i.e., at least 6 months) (10). Another study compared the mortality rates between patients undergoing PD and HD using the French Renal Epidemiology and Information Network registry; the HD group was further separated into planned and unplanned HD groups. As a result, high mortality with PD was prominent in both groups with diabetes mellitus (versus planned HD) and without diabetes mellitus (versus all HD groups) (11). Several other studies compared mortality rates between HD and PD patients without age limitations but included elderly patients in their subgroup analysis. There were also controversial results regarding survival benefits; some studies definitely preferred PD (24,28), but others, including the present study, favored HD (10,25). These heterogeneities across studies were reflected in high I² statistics, which reached 94.3% in the meta-analysis. Certain studies did not describe overall HRs but only the HRs of subgroups, and, thus, combined HRs in those studies could be identified after calculations based on the subgroup results. Despite these factors, the pooled HR was 1.10 (95% CI, 1.01 to 1.20), which indicates that elderly patients receiving HD had a lower overall mortality risk than those receiving PD.

According to the baseline characteristics of previous studies, the presence or proportion of diabetes mellitus is expected to affect the correlations, even when analyses were adjusted for the presence of diabetes mellitus. Studies favoring PD included low proportions of patients with diabetes mellitus (<30%). In contrast, studies preferring HD included high proportions of diabetes mellitus patients, which reached >50%. Furthermore, there was a time-dependent change or crossover in mortality rates of the two modalities, as shown in the present study and most other studies. Dialysis start time should also be considered because certain important clinical practices, such as glucose dialysate, have been changed over time and could affect the mortality rate of PD patients (32). To avoid possible interactions with mortality, we conducted analyses stratified by these factors. When elderly Korean patients had diabetes mellitus or had been undergoing dialysis longer, HD was further preferred to PD with respect to survival. In a meta-analysis, the pooled HR was larger in the diabetes and ≥1 year on dialysis subgroups than in the non-diabetes and <1 year subgroups. The results of this comparison appeared to hold up regardless of the year in which dialysis began when pooling the studies. However, when the analysis was limited to two studies that had cohorts from both the 1990s and 2000s, the HRs of PD compared with HD decreased after the year 2000.

Current observational results cannot provide any clues regarding the mortality discrepancy between HD and PD patients. However, subgroup results may suggest some possible mechanisms. Among patients with ESRD, infectious diseases are one of the main causes of death, and they are common in both HD and PD patients (33). Previous observational studies showed a higher risk of mortality and morbidity from infection among PD patients than HD patients (34,35). Death due to infection may be more influential in elderly patients, who have especially weak immune systems. In addition, the discrepancy between the two modalities may be larger in high-risk situations, such as diabetes mellitus, longer dialysis duration, or the use of high-glucose dialysate. Other factors, such as insufficient home-based therapy, self-care, or education regarding PD, may have an additional effect on the high mortality in PD compared with HD (36,37). Particularly, home-based therapy is a crucial issue regarding elderly patients, because all of them could not be admitted to elderly facilities because of the high costs associated with care.

Although the present results are informative, this study has some limitations. The observational study design limits full applicability of our conclusions to current clinical practice. However, randomized trials that control for all possible confounders are not feasible (8); thus, the present results may be helpful in raising awareness of the trends regarding modalities or in directing the conduct of future randomized controlled studies. The Korean dataset did not provide several covariates, such as biochemical and socioeconomic measures. We could not analyze cause-specific mortality. Potential biases of meta-analysis (i.e., heterogeneity of results; unavailability of information, such as socioeconomic parameters; and performance bias due to study design) could not be excluded. These limitations suggest methodologic and understandable weakness; however, multiple approaches, including subgroup analysis and a careful interpretation of meta-analyses, are believed to be useful for further understanding of individual studies; this was the primary objective of our study.

The choice of dialysis modality in elderly patients is an essential issue. The Korean data presented here and the pooled results from the meta-analysis suggest that a tailored selection of modality is needed in elderly patients. Although randomized controlled studies are needed to draw a firm conclusion, the present study may be helpful in guiding the selection of dialysis modality in clinical practice.

Disclosures

None.