A population-based approach indicates an overall higher patient mortality with peritoneal dialysis compared to hemodialysis in Korea

Transcription

1 & 2014 International Society of Nephrology clinical investigation see commentary on page 877 A population-based approach indicates an overall higher patient mortality with peritoneal dialysis compared to hemodialysis in Korea Hyunwook Kim 1, Kyoung Hoon Kim 2, Kisoo Park 3, Shin-Wook Kang 4, Tae-Hyun Yoo 4, Song Vogue Ahn 5, Hyeong Sik Ahn 6, Hoo Jae Hann 7, Shina Lee 8, Jung-Hwa Ryu 8, Seung-Jung Kim 8, Duk-Hee Kang 8, Kyu Bok Choi 8 and Dong-Ryeol Ryu 8 1 Department of Internal Medicine, Wonkwang University College of Medicine Sanbon Hospital, Gunpo, Korea; 2 Department of Public Health, Graduate School, Korea University, Seoul, Korea; 3 Department of Preventive Medicine, School of Medicine and Health Science Institute, Gyeongsang National University, Jinju, Korea; 4 Department of Internal Medicine, College of Medicine, Yonsei University, Seoul, Korea; 5 Department of Preventive Medicine, Wonju College of Medicine, Yonsei University, Wonju, Korea; 6 Department of Preventive Medicine, College of Medicine, Korea University, Seoul, Korea; 7 Ewha Medical Research Institute, School of Medicine, Ewha Womans University, Seoul, Korea and 8 Department of Internal Medicine, School of Medicine, Ewha Womans University, Seoul, Korea To date, only a few large-scale studies have measured the effect of dialysis modality on mortality in Asian populations. Here, we sought to compare survival between incident hemodialysis (HD) and peritoneal dialysis (PD) patients using the Korean Health Insurance Review & Assessment Service database. This enabled us to perform a population-based complete survey that included 32,280 incident dialysis patients and followed them for a median of 26.5 months. To reduce biases due to nonrandomization, we first matched 7049 patient pairs with similar propensity scores. Using the log-rank test, we found the mortality rate in PD patients was significantly higher than that in HD patients. Subsequent subgroup analyses indicated that in older patients (55 years and older), with the exception of the subgroup of patients with no comorbidities and the subgroup of patients with malignancy, PD was consistently associated with a higher mortality rate. In younger patients (under 55 years), regardless of the covariates, the survival rate of PD patients was comparable to that of HD patients. Thus, while the overall mortality rate was higher in incident PD patients, mortality rates of some incident PD and HD patients were comparable in Korea. Kidney International (2014) 86, ; doi: /ki ; published online 7 May 2014 KEYWORDS: hemodialysis; mortality; peritoneal dialysis Correspondence: Dong-Ryeol Ryu, Department of Internal Medicine, School of Medicine, Ewha Womans University, Mok-Dong, Seoul, Korea. drryu@ewha.ac.kr Received 1 July 2013; revised 7 March 2014; accepted 13 March 2014; published online 7 May 2014 With rapid increases in the incidence and prevalence rates of end-stage renal disease (ESRD) throughout the world, kidney transplantation (KT) has become the most preferred modality of renal replacement therapy for patients suffering from ESRD. However, its widespread use is hampered due to the shortage of living and deceased donors. Therefore, most of the ESRD patients are forced to start renal replacement therapy by hemodialysis (HD) or peritoneal dialysis (PD). Although a randomized-controlled study is the most reliable and informative method of choice for comparing outcomes of these two treatments, an attempt to perform such a trial has never been successful due to the patient s predetermined preference. 1 Instead, several registry or nonrandomized cohort-based studies, particularly from the United States, Canada, Australia, and the Netherlands, have been used as valuable alternatives to help determine the optimal dialysis modality However, as dialysis outcomes following the use of the chosen modality can be affected by various factors, their interpretation is not always straightforward. In this regard, several reports have indicated that a number of differences exist in the characteristics and outcomes of dialysis patients according to their racial/ethnic groups. An analysis of the United States Renal Data System (USRDS) in 1998 revealed that Asians were younger at the initiation of dialysis therapy and had a lower body mass index than Whites. 12 The primary causes of ESRD also differed between Asians and Whites: Asians had more ESRD from glomerulonephritis and diabetes, but had less ESRD from cystic kidney disease and hypertension. 12 In addition, baseline coronary artery disease, congestive heart failure, hypertension, and peripheral vascular disease were significantly less prevalent in Asians than in Whites. 13 Furthermore, there are significant differences in the overall and cardiovascular mortality in dialysis patients according to the racial/ethnic groups Therefore, Kidney International (2014) 86,

2 clinical investigation H Kim et al.: Dialysis modality and mortality in Korea it seems to be necessary to determine whether the results from previous studies, conducted almost exclusively in a Western population, can be appropriately generalized to other racial/ethnic groups; however, to date, there are only a few studies comparing the outcomes between HD and PD in Asians. 18,19 Hence, we present here the results of analysis using a population-based, large-scale Korean registry data to compare mortality between HD and PD patients, and to evaluate the relationship between mortality and various baseline covariates. RESULTS Baseline characteristics of the subjects A total of 32,280 eligible patients who had started dialysis therapy between 1 January 2005 and 31 December 2008 were analyzed. At the initiation of dialysis treatment, the mean patient age was 57.0±13.8 years, 58.7% of patients were males, and 50.1% of patients had diabetes. The number of the incident HD and PD patients was 24,399 (75.6%) and 7881 (24.4%), respectively. The median follow-up duration was 26.5 months (range, months). A detailed comparison of the baseline characteristics between HD and PD patients is shown in Table 1 (before matching). Comparisons of mortality according to dialysis modalities During the median follow-up period of 26.5 months, 6479 patients undergoing HD (26.6%) and 2156 patients undergoing PD (27.4%) died, which corresponded to a death rate of and deaths per 1000 patient-years, respectively. The 1-, 2-, and 3-year survival rates of HD patients were 87%, 78%, and 72%, respectively, and those of PD patients were 90%, 79%, and 71%, respectively. Survival analysis using Kaplan Meier plots showed that there was no difference in the unadjusted mortality rate between HD and PD patients (Figure 1; log-rank test, P ¼ 0.955) after censoring for KT. Further, to reduce the bias arising from a potential difference in the probability of receiving KT between modalities, we also compared the unadjusted survival rate between HD and PD patients without censoring for KT and obtained a similar result (Supplementary Figure S1 online; log-rank test, P ¼ 0.714). In order to address the concerns that these results were probably influenced by a large heterogeneity in baseline characteristics between HD and PD patients as shown in Table 1 (before matching), we next performed propensity score matching, thereby incorporating and adjusting for all the listed baseline differences between HD and PD patients. 20 After propensity score matching, the two groups (7049 HD patients vs PD patients) were well balanced for baseline characteristics (Table 1, after matching), and this was visually confirmed by an improvement in overlap of the distribution of propensity scores (Supplementary Figure S2 online). For these matched groups, we first performed Kaplan Meier survival analysis after censoring for KT and found that the mortality was significantly higher in PD patients than in HD patients (Figure 2; log-rank test, Po0.001), which corresponded to a significant increase of 20% in mortality rates (hazard ratio (HR), 1.20; 95% confidence interval (CI), ; Po0.001). This survival advantage for HD over PD in the matched groups was not altered even when the patients were not censored for KT (Supplementary Figure S3 online; log-rank test, Po0.001). Subgroup analyses according to baseline covariates Next, we divided the entire study population into subgroups according to various baseline covariates to compare mortality between dialysis modalities within subgroups. Although there were exceptions in subgroups of patients with liver disease and malignancy and those with none of the listed Table 1 Baseline characteristics of the participants before and after propensity score matching Variables All PD patients (N ¼ 7881) Before matching All HD patients (N ¼ 24,399) P-values Matched PD patients (N ¼ 7049) After matching Matched HD patients (N ¼ 7049) Age (years) 54.2± ±13.8 o ± ± years of age 3962 (50.3) 14,729 (60.4) o (50.8) 3571 (50.7) Females (versus males) 3443 (43.7) 9894 (40.6) o (43.2) 3036 (43.1) National Health Insurance (versus Medical Aid) 6898 (87.5) 20,961 (85.9) o (87.3) 6146 (87.2) Diabetes mellitus 3996 (50.7) 12,190 (50.0) (51.1) 3685 (52.3) P-values Comorbidities other than Myocardial infarction 367 (4.7) 797 (3.3) o (3.9) 280 (4.0) Congestive heart failure 1269 (16.1) 3483 (14.3) o (15.5) 1083 (15.4) Peripheral artery disease 424 (5.4) 1484 (6.1) (5.3) 371 (5.3) Cerebrovascular accident 830 (10.5) 3208 (13.1) o (10.5) 765 (10.9) Chronic pulmonary disease 1233 (15.6) 3988 (16.3) (15.6) 1019 (14.5) Peptic ulcer disease 1072 (13.6) 3637 (14.9) (13.6) 957 (13.6) Liver disease 780 (9.9) 2680 (11.0) (9.9) 682 (9.7) Cancer 321 (4.1) 1647 (6.8) o (3.7) 205 (2.9) Abbreviations:, diabetes mellitus; HD, hemodialysis; PD, peritoneal dialysis. Age is expressed as the mean ±s.d., and all other data are expressed as the number (%). 992 Kidney International (2014) 86,

3 H Kim et al.: Dialysis modality and mortality in Korea clinical investigation Cumulative survival P = HD PD Follow-up (years) Number at risk (events) HD PD 24,399 23,436 16, (0) (3597) (1575) (828) (397) (84) (0) (999) (633) (329) (168) (27) Figure 1 Crude Kaplan Meier survival curves for HD patients (gray line) and PD patients (black line) censored for kidney transplantation. The full prematch group of 32,280 patients was included in this analysis. HD, hemodialysis; PD, peritoneal dialysis Cumulative survival P HD PD Follow-up (years) Number at risk (events) HD PD (407) (0) (817) (423) (219) (109) (25) (0) (892) (571) (290) (149) (21) Figure 2 Crude Kaplan Meier survival curves for HD patients (gray line) and PD patients (black line) censored for kidney transplantation. The propensity score matched groups (7049 patients for each dialysis modality) were included in this analysis. HD, hemodialysis; PD, peritoneal dialysis. comorbidities, we found that mortality risk was consistently lower in HD patients compared with PD patients in all other subgroups (Supplementary Figure S4 online), and a similar trend was also seen in the propensity score matched groups (Supplementary Figure S5 online). Therefore, we next tested various combinations of interactions between baseline covariates and dialysis modality for mortality in the propensity score matched groups, and we found that, in agreement with Kidney International (2014) 86,

4 clinical investigation H Kim et al.: Dialysis modality and mortality in Korea the results of previous reports, 2,6 7,11 the age at the initiation of dialysis therapy significantly interacted with dialysis modality for mortality, not only as a categorical variable (X55 years vs. o55 years, P for interactiono0.001) but also as a continuous variable (P for interaction ¼ 0.034). Therefore, we performed subsequent analyses whereby mortality rates were assessed separately for the older age group (X55 years) and the younger age group (o55 years) in the propensity score matched groups. Within each age group, the patients were further stratified into subgroups according to other baseline covariates. These subgroup analyses indicated that in the older age group (X55 years), the mortality rates were consistently higher in PD patients than in HD patients, irrespective of the baseline characteristics, with the exception of patients with malignancy (HR, 0.94; 95% CI, , P ¼ 0.739) or patients with none of the listed comorbidities (HR, 1.00; 95% CI, , P ¼ 0.988; Figure 3), whereas in the younger age group (o55 years), regardless of comorbidities, there were no significant differences in mortality rates between HD and PD patients (Figure 4). These trends were also observed in the entire study population (Supplementary Figure S6 and S7 online). As these results suggest that the age at the start of the dialysis treatment is a key determinant of mortality according to the dialysis modality, we divided the patients into 10-year age groups and then examined the mortality rates on the basis of the presence or absence of diabetes in the propensity score matched groups. As shown in Figure 5, we found that there were no significant differences in mortality rates between HD and PD patients who started dialysis at an age of less than 50 years, regardless of the presence or absence of diabetes, whereas mortality rates were consistently higher in PD patients compared with HD patients who started dialysis at an age of 60 years or above, regardless of the presence or absence of diabetes. In contrast, only in patients aged between 50 and 59 years were the mortality rates affected by the presence or absence of diabetes. In this age group, the overall mortality rates were significantly higher in patients on PD compared with patients on HD (HR, 1.26; 95% CI, , P ¼ 0.001), which can be mainly attributed to significantly higher mortality rates in diabetic patients on PD than in diabetic patients on HD (HR, 1.32; 95% CI, ; P ¼ 0.001), whereas there was no significant difference in the mortality rates between nondiabetic HD Subgroup. of patients (%) PD better HD better HR (95% CI) P- values All patients aged 55 years 7151 (100) 1.30 ( ) Sex 3840 (53.7) 3311 (46.3) Male Female 1.28 ( ) 1.33 ( ) Diabetes mellitus 4467 (62.5) 2684 (37.5) 1.35 ( ) 1.23 ( ) Myocardial infarction 477 (6.7) 6674 (93.3) 1.31 ( ) 1.30 ( ) Congestive heart failure 1464 (20.5) 5687 (79.5) 1.45 ( ) 1.26 ( ) Cerebrovascular accident 1079 (15.1) 6072 (84.9) 1.42 ( ) 1.28 ( ) Peripheral artery disease 498 (7.0) 6653 (93.0) 1.74 ( ) 1.27 ( ) Chronic pulmonary disease 1361 (19.0) 5790 (81.0) 1.39 ( ) 1.28 ( ) Liver disease 573 (8.0) 6578 (92.0) 1.39 ( ) 1.28 ( ) Peptic ulcer disease 1077 (15.1) 6074 (84.9) 1.26 ( ) 1.31 ( ) Malignancy 268 (3.7) 6883 (96.3) 0.94 ( ) 1.32 ( ) Comorbidity 6040 (84.5) 1111 (15.5) 1.35 ( ) 1.00 ( ) Figure 3 Subgroup analyses comparing hazard ratios (HR) for mortality between HD patients and PD patients aged more than 55 years in the propensity score matched groups. HD, hemodialysis; PD, peritoneal dialysis. 994 Kidney International (2014) 86,

5 H Kim et al.: Dialysis modality and mortality in Korea clinical investigation Subgroup. of patients (%) PD better HD better HR (95% CI) P- values All patients aged < 55 years 6947 (100) 0.99 ( ) Sex 4117 (60.1) 2770 (39.9) Male Female 1.05 ( ) 0.90 ( ) Diabetes mellitus 2820 (40.6) 4127 (59.4) 0.97 ( ) 0.99 ( ) Myocardial infarction 79 (1.1) 6868 (98.9) 0.86 ( ) 0.99 ( ) Congestive heart failure 713 (10.5) 6234 (89.7) 0.97 ( ) 0.98 ( ) Cerebrovascular accident 426 (6.1) 6521 (93.9) 1.29 ( ) 0.95 ( ) Peripheral artery disease 247 (3.6) 6700 (96.4) 0.92 ( ) 0.99 ( ) Chronic pulmonary disease 756 (10.9) 6191 (89.1) 0.93 ( ) 0.99 ( ) Liver disease 810 (11.7) 6137 (88.3) 0.82 ( ) 0.02 ( ) Peptic ulcer disease 837 (12.0) 6110 (88.0) 1.25 ( ) 0.95 ( ) Malignancy 197 (2.8) 6750 (97.2) 0.98 ( ) 0.97 ( ) Comorbidity 4434 (63.8) 2513 (36.2) ( ) 0.75 ( ) Figure 4 Subgroup analyses comparing hazard ratios (HR) for mortality between HD patients and PD patients aged 55 years or less in the propensity score matched groups. HD, hemodialysis; PD, peritoneal dialysis patients and nondiabetic PD patients (HR, 1.18; 95% CI, ; P ¼ 0.224). Similar trends were also seen in the entire study population (Supplementary Figure S8 online). Comparison of mortality rates according to the dialysis modality on the basis of the time interval In the propensity score matched groups, the proportional hazards of dialysis modality were visually confirmed. However, in contrast with our results, a number of previous studies have shown that the mortality risk associated with dialysis modality changed over time. 4,7,21,22 To address this issue, we calculated dialysis modality associated HRs for mortality separately at each 6-month interval in the matched groups. As shown in Figure 6, after the first 6 months of dialysis treatment, the mortality risks were consistently higher in PD patients compared with HD patients up to 36 months of follow-up, regardless of whether the patients receiving KT were censored. Next, in the entire study population of 32,280 patients, we estimated baseline covariateadjusted HRs for mortality at each 6-month interval and obtained similar results (Supplementary Figure S9 online). DISCUSSION As in the other industrialized countries, the prevalence of ESRD is increasing at an estimated rate of 12% per year in Korea. 23 Therefore, the importance of choosing the optimal dialysis modality has been highlighted. In this regard, a number of previous studies have provided several guides to help choose the modality. 4,5,7,21,22,24 26 However, a large-scale study has rarely been performed to determine whether these guidances can be generalized to Asians, particularly to Koreans. Therefore, in this study, we present the results of the analysis of population-based registry data for almost ethnically homogenous Koreans, with two distinguishing features that helped to make our results more robust with a minimal bias toward modality selection. First, in most of the major medical centers in Korea, selections of the dialysis modality are made according to the patient s preference and the physician s choice based on medical factors without any restrictions due to reimbursement problems as in other countries. 27 In relation to this aspect, penetration of PD in Korea is not low and its prevalence rate in 2010 was estimated as 15.6%, which is much higher than that in the United Kidney International (2014) 86,

6 clinical investigation H Kim et al.: Dialysis modality and mortality in Korea Subgroup (% Of all). of patients (% In subgroup) PD better HD better HR (95% CI) P- values 520 (100) 1.70 ( ) < 30 years (3.7) 80 (15.4) 0.92 ( ) (84.6) n ( ) (100) 1.06 ( ) ~ 39 years (11.1) 384 (24.6) 1178 (75.4) n ( ) 1.00 ( ) (100) 0.90 ( ) ~ 49 years (21.8) 1328 (43.2) 1743 (56.8) n ( ) 0.83 ( ) (100) 1.26 ( ) ~ 59 years (25.7) 2200 (60.8) 1417 (39.2) n ( ) 1.18 ( ) ~ 69 years (25.0) 3531 (100) 2314 (65.5) 1217 (34.5) n ( ) 1.37 ( ) 1.29 ( ) (100) 1.27 ( ) > 70 years (12.7) 981 (54.6) 816 (45.4) n ( ) 1.25 ( ) Figure 5 Subgroup analyses comparing hazard ratios (HR) for mortality between HD and PD patients divided into 10-year age groups stratified by the presence or absence of diabetes in the propensity score matched groups., diabetes mellitus; HD, hemodialysis; PD, peritoneal dialysis. States, and it corresponds approximately to the average rate of PD in the countries that are registered in the USRDS 2012 (B15.0%). 28 Second, we recruited a large number of incident dialysis patients; a total of 32,280 patients with a median follow-up of 26.5 months, which allowed further generalization of the results of our study. When comparing the results obtained in this study with those from the previous studies conducted in Western populations, we observed several notable findings. First, dialysis patients in this study had a significantly lower death rate of per 1000 patient-years than that mentioned in the USRDS registry 2012 (193 per 1000 patient-years) or that mentioned in registry data of other Western countries (the United Kingdom, Sweden, Spain, rway, Netherlands, Denmark, Belgium, Germany, France, and so on). 17 This difference may be due to the relatively younger age and a lower prevalence of preexisting cardiovascular diseases in our subjects at the initiation of dialysis therapy compared with those in previous studies conducted in Western populations. 13,21,26 In addition, genetic and environmental factors may also have contributed to this difference. 17 However, the overall death rate of Korean dialysis patients in this study was substantially higher than that in the previously published data in This difference may be explained by a recent rapid shift in the peak age of incident dialysis patients from a younger age toward an older age. 23 Second, the overall mortality rate was higher in PD patients than in HD patients in our study. Considering that this study included the participants who had started dialysis relatively recently ( ), the results of this study are in contrast with those of some recent cohort studies. A study using USRDS data revealed that there was no significant difference in the mortality risk between HD and PD patients in the cohorts. 11 Another study analyzing the Center for Medicare and Medicaid (CMS) cohort, which consisted of adult patients who initiated dialysis in the 996 Kidney International (2014) 86,

7 H Kim et al.: Dialysis modality and mortality in Korea clinical investigation PD better HD better HR (95% CI) P- values 0 to 6 Months 1.03 ( ) ( ) to 12 Months 1.20 ( ) 1.20 ( ) to 18 Months 1.36 ( ) 1.27 ( ) 18 to 24 Months 1.30 ( ) 1.30 ( ) to 30 Months 1.25 ( ) 1.26 ( ) to 36 Months 1.38 ( ) 1.39 ( ) Months ( ) 1.25 ( ) Figure 6 Hazard ratios (HR) by different starting points at a 6-month interval in the propensity score matched groups (&: HR not censored for KT, : HR censored for KT). HD, hemodialysis; KT, kidney transplantation; PD, peritoneal dialysis United States in 2003, showed that the overall survival was similar for HD and PD patients who survived for the first 90 days. 10 Furthermore, the most recent large-scale European study showed that the overall adjusted survival benefit was significantly higher in PD patients compared with HD patients. 29 However, it should be noted that, although the relative survival rate of PD patients seems to be lesser than that of HD patients, the survival rate of PD patients in this study was not low in absolute terms when compared with that in previous studies. 6,9 11,30 Third, in this study, we explored the effects of interactions between dialysis modality and various baseline covariates for mortality, and found that age was the most important determinant of dialysis modality related mortality; within the groups stratified by age (X55 years vs. o55 years), other baseline covariates had a marginal effect on the mortality. Among these other covariates, notably, there were no significant interactions between diabetes, myocardial infarction, or congestive heart failure and dialysis modality for mortality (P for interaction ¼ 0.540, 0.578, and 0.293, respectively), which is in contrast to the results of previous studies. 5 6,21,26 Furthermore, in contrast to the results of a recent European study by van de Luijtgaarden et al. 29 there was no interaction between sex and dialysis modality (P for interaction ¼ 0.645). Interestingly, there were only two exceptions to this result. First, in the patients with no additional comorbidity, the survival of PD patients was similar to that of HD patients irrespective of the age group. This result is consistent with those of previous studies, in which McDonald et al. 9 found that the benefit of PD in the first 12 months was particularly significant in theo60 years group without comorbidities. Mehrotra et al. 11 also documented from the USRDS records that, in younger (o65 years) nondiabetic patients with no additional comorbidity, PD was associated with a significantly lower mortality risk compared with HD from 1999 onwards; whereas van de Luijtgaarden et al. and Vonesh et al. 6,29 reported that, irrespective of age, HD was associated with a significantly higher mortality Kidney International (2014) 86,

8 clinical investigation H Kim et al.: Dialysis modality and mortality in Korea compared with PD in patients without a reported baseline comorbidity. Second, in the patients with malignancy, who might be considered to have a too-high mortality risk to be affected by the dialysis modality, there was no difference in mortality between dialysis modalities. Only a few studies have examined the interactions between various other baseline comorbidities and the dialysis modality for mortality in detail. A recent European study by van de Luijtgaarden et al. 29 has explored these relationships, and showed that the absence of diabetes, ischemic heart disease, peripheral vascular disease, cerebrovascular disease, and malignancy was associated with better survival in PD patients than in HD patients. In contrast to this result, our models showed that the presence or absence of such comorbidities had no or marginal effects on mortality. Finally, another interesting finding of our study was that the lower mortality risk in HD patients compared with PD patients was relatively constant over the duration of dialysis therapy (Figure 6 and Supplementary Figure S9 online). This result is in contrast with those of other previous studies, in which survival was better in the early years (usually in the first 12 to 24 months after initiation of dialysis therapy) in PD patients, and thereafter the outcomes varied among the different study participants. 3,5,8,9,24 The reason for this difference is not clear, but it can be speculated that some previously established factors related to the early benefits of PD provided relatively smaller benefits to the participants in this study, such as better preservation of residual renal function in PD, or lower incidences of late referral related and vascular access related complications in HD patients. 8,31,32 As in other registry-based studies, this study also has some inherent limitations. Although two different approaches were used to minimize the effects of selection bias, it was not likely that nonrandom assignment of patients to the dialysis modality would be free from indication bias based on prognosis of the ESRD patients, which is difficult to measure, or could provide causality between dialysis modality and mortality. Other dialysis-related factors that might potentially influence mortality, such as data of residual renal function, anemia status, critical laboratory biomarkers of inflammation and nutrition, delivered dialysis doses, parameters of quality of life, or the proportion of the patients having planned vascular access, and the causes of the deaths were also unavailable. Despite these limitations, our study has provided several clinically relevant findings. Owing to a large sample size comprising nearly the entire population of incident dialysis patients with a relatively long follow-up and a high proportion of PD patients, we could successfully match one of the largest propensity score based cohorts of patients initiating dialysis. In addition, we collected the data of patients covered by both the National Health Insurance and Medical Aid, which are distinguished by differences in socioeconomic status, and adjusted for these differences while analyzing the data; the results of this study are not likely to be biased toward a specific socioeconomic background. Furthermore, although we could not incorporate some essential parameters listed above that potentially affect mortality into the analyses, a number of reports have indicated that these parameters are not completely independent of each other but mutually interconnected and influenced by factors such as gender, diabetic status, or cardiovascular morbidities, which we have measured. 31,33 35 Therefore, we can cautiously speculate that the effects of these unmeasured but critical parameters were, at least partly, reflected in the existing results of our study. Finally, although it is not clear which cause of death is dominant in our study, it should be noted that another study based on other large-scale Korean registry data reported that cardiovascular death was the most common cause of death in both HD patients and PD patients in Korea without statistically significant differences in proportions, accounting for 45% and 43% of all deaths in HD patients and PD patients, respectively. 23 Furthermore, in the same study, there was no statistical difference in the mean arterial pressure between HD patients and PD patients. Taken together, it can be cautiously extrapolated that cardiovascular death, independent of the degree of blood pressure control, was more prevalent in PD patients than in HD patients. In conclusion, the results of our study can help to provide some relevant insights on choosing the dialysis modality particularly for Korean patients or for those who are similar to Koreans in terms of both ethnic and medico-social aspects. However, there is no doubt that a definite answer to the dialysis modality choice related questions can only be obtained from a successfully conducted, randomizedcontrolled study, which is difficult to perform in a real clinical environment. 1,36 MATERIALS AND METHODS Data source and study population We used the Korean Health Insurance Review and Assessment Service database. The organization of this database and its use for analysis has been briefly described under Supplementary Methods. The comorbidities of the participants were identified by reviewing their medical history during the last 1 year before the initiation of dialysis therapy. The list of analyzed comorbidities was determined on the basis of the suggestions by Charlson et al., 37 and ICD-10 codes were used according to the proposed algorithms by Quan et al. 38 For analysis of the differences in mortality between the dialysis modalities, we initially identified all of the incident dialysis patients who had started HD or PD therapy between 1 January 2005 and 31 December 2008 in Korea (n ¼ 35,422). Among them, we first excluded the patients who survived for less than 90 days from the date of dialysis initiation. There are mainly three reasons for excluding these patients. First, in most cases, acutely ill patients who needed urgent initiation of dialysis were treated preferentially with HD. Therefore, if analyzed on the basis of the true initial dialysis modality, there is a high probability that the results could be biased against HD. Second, mortality occurring in the first 90 days is considered to be affected more by the preexisting comorbidity than by the dialysis modality per se. Third, even the patients in whom the final dialysis modality is PD are usually treated initially by temporary HD. Therefore, we only included the patients who survived for 90 days and remained on 998 Kidney International (2014) 86,

9 H Kim et al.: Dialysis modality and mortality in Korea clinical investigation chronic dialysis. Next, the patients who were younger than 18 years of age were excluded; the remaining eligible 32,280 patients were included in the final analyses. Statistical analysis Baseline characteristics of HD and PD patients were compared using the independent t-test for continuous variables and the Pearson w 2 -test for categorical variables, and the propensity scores were estimated using multiple logistic regression analysis adjusted for the patient s age, sex, type of insurance (National Health Insurance versus Medical Aid), and the presence or absence of a variety of clinical and coexisting conditions (diabetes mellitus, myocardial infarction, congestive heart failure, cerebrovascular accident, peripheral artery disease, chronic pulmonary disease, liver disease, peptic ulcer disease, and cancer). 39 Model discrimination was assessed with C-statistics, and calibration was assessed with Hosmer-Lemeshow statistics. After creating the propensity scores, we matched the HD and PD patients with similar propensity scores in a 1:1 ratio using the 4-digit casecontrol match algorithm, 40 and assessed the balance in baseline covariates between the two groups with the t-test for continuous variables and the McNemar s test for categorical variables. For assessing survival, we adopted an intention-to-treat analysis and considered the dialysis modality at day 90 to be the initial dialysis modality for the reasons mentioned above and used day 90 as the starting point (day 0). Hence, the patients analyzed in this study were left-censored for the immortal first 90 days after the initiation of dialysis, and they were right-censored at the time of KT or 31 December The end point of this study was time to death. We used the Kaplan Meier survival curves with the log-rank test to compare the differences in survival between HD and PD patients and Cox proportional hazard regression models to estimate the HRs. By Cox proportional hazard models, we evaluated the HRs using two different approaches. First, we compared the mortality rates between the propensity score matched groups. Second, we conducted baseline covariate-adjusted subgroup analyses stratified by age (decade), baseline covariates, and 6-month interval of dialysis therapy. We tested the assumption of proportional hazards by a visual examination of both the log-minus-log plots and Schoenfeld residual plots. All statistical tests were evaluated using a two-tailed 95% CI, and Po0.05 was considered statistically significant. All statistical analyses were conducted using the statistical package SAS 9.1 (SAS Institute, Cary, NC). DISCLOSURE All the authors declared no competing interests. ACKNOWLEDGMENTS This study was supported by a 2011 Grant from the Korean Academy of Medical Sciences. SUPPLEMENTARY MATERIAL Figure S1. Crude Kaplan-Meier survival curves for HD patients (grey line) and PD patients (black line) not censored for kidney transplantation. Figure S2. Distribution of propensity scores before (A) and after (B) matching. Figure S3. Crude Kaplan-Meier survival curves for HD patients (grey line) and PD patients (black line) not censored for kidney transplantation. Figure S4. Subgroup analyses comparing all baseline covariateadjusted hazard ratios for mortality between HD patients and PD patients in the entire population. Figure S5. Subgroup analyses comparing hazard ratios for mortality between HD patients and PD patients in the propensity scorematched groups. Figure S6. Subgroup analyses comparing all baseline covariateadjusted hazard ratios for mortality between HD patients and PD patients aged more than 55 years in the entire population. Figure S7. Subgroup analyses comparing all baseline covariateadjusted hazard ratios for mortality between HD patients and PD patients aged 55 years or less in the entire population. Figure S8. Subgroup analyses comparing all baseline covariateadjusted hazard ratios for mortality between HD and PD patients divided into 10-year age groups stratified by the presence or absence of diabetes in the entire population. Figure S9. All baseline covariate-adjusted hazard ratios by different starting points at a 6-month interval in the entire population after adjusting for the baseline covariates (&: hazard ratios not censored for KT, : hazard ratios censored for KT). Supplementary material is linked to the online version of the paper at REFERENCES 1. Korevaar JC, Feith GW, Dekker FW et al. Effect of starting with hemodialysis compared with peritoneal dialysis in patients new on dialysis treatment: a randomized controlled trial. Kidney Int 2003; 64: Bloembergen WE, Port FK, Mauger EA et al. A comparison of mortality between patients treated with hemodialysis and peritoneal dialysis. JAm Soc Nephrol 1995; 6: Vonesh EF, Moran J. Mortality in end-stage renal disease: a reassessment of differences between patients treated with hemodialysis and peritoneal dialysis. J Am Soc Nephrol 1999; 10: Fenton SS, Schaubel DE, Desmeules M et al. Hemodialysis vs. peritoneal dialysis: a comparison of adjusted mortality rates. Am J Kidney Dis 1997; 30: Collins AJ, Hao W, Xia H et al. Mortality risks of peritoneal dialysis and hemodialysis. Am J Kidney Dis 1999; 34: Vonesh EF, Snyder JJ, Foley RN et al. The differential impact of risk factors on mortality in hemodialysis and peritoneal dialysis. Kidney Int 2004; 66: Liem YS, Wong JB, Hunink MG et al. Comparison of hemodialysis and peritoneal dialysis survival in The Netherlands. Kidney Int 2007; 71: Jaar BG, Coresh J, Plantinga LC et al. Comparing the risk for death with peritoneal dialysis and hemodialysis in a national cohort of patients with chronic kidney disease. Ann Intern Med 2005; 143: McDonald SP, Marshall MR, Johnson DW et al. Relationship between dialysis modality and mortality. J Am Soc Nephrol 2009; 20: Weinhandl ED, Foley RN, Gilbertson DT et al. Propensity-matched mortality comparison of incident hemodialysis and peritoneal dialysis patients. J Am Soc Nephrol 2010; 21: Mehrotra R, Chiu YW, Kalantar-Zadeh K et al. Similar outcomes with hemodialysis and peritoneal dialysis in patients with end-stage renal disease. Arch Intern Med 2011; 171: Wong JS, Port FK, Hulbert-Shearon TE et al. Survival advantage in Asian American end-stage renal disease patients. Kidney Int 1999; 55: Goodkin DA, Bragg-Gresham JL, Koenig KG et al. Association of comorbid conditions and mortality in hemodialysis patients in Europe, Japan, and the United States: the Dialysis Outcomes and Practice Patterns Study (DOPPS). J Am Soc Nephrol 2003; 14: Held PJ, Brunner F, Odaka M et al. Five-year survival for end-stage renal disease patients in the United States, Europe, and Japan, 1982 to Am J Kidney Dis 1990; 15: Uhlig K, Levey AS, Sarnak MJ. Traditional cardiac risk factors in individuals with chronic kidney disease. Semin Dial 2003; 16: Goodkin DA, Mapes DL, Held PJ. The dialysis outcomes and practice patterns study (DOPPS): how can we improve the care of hemodialysis patients? Semin Dial 2001; 14: Kidney International (2014) 86,

10 clinical investigation H Kim et al.: Dialysis modality and mortality in Korea 17. Yoshino M, Kuhlmann MK, Kotanko P et al. International differences in dialysis mortality reflect background general population atherosclerotic cardiovascular mortality. J Am Soc Nephrol 2006; 17: Huang C-C, Cheng K-F, Wu H-DI.. Survival analysis: comparing peritoneal dialysis and hemodialysis in Taiwan. Perit Dial Int 2008; 28: S15 S Chang Y-K, Hsu C-C, Hwang S-J et al. A comparative assessment of survival between propensity score-matched patients with peritoneal dialysis and hemodialysis in Taiwan. Medicine (Baltimore) 2012; 91: Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983; 70: Stack AG, Molony DA, Rahman NS et al. Impact of dialysis modality on survival of new ESRD patients with congestive heart failure in the United States. Kidney Int 2003; 64: Termorshuizen F, Korevaar JC, Dekker FW et al. Hemodialysis and peritoneal dialysis: comparison of adjusted mortality rates according to the duration of dialysis: analysis of The Netherlands Cooperative Study on the Adequacy of Dialysis 2. J Am Soc Nephrol 2003; 14: Jin DC, Ha IS, Kim NH et al. Brief report: renal replacement therapy in Korea, Kidney Res Clin Pract 2012; 31: Heaf JG, Lokkegaard H, Madsen M. Initial survival advantage of peritoneal dialysis relative to haemodialysis. Nephrol Dial Transplant 2002; 17: Schaubel DE, Morrison HI, Fenton SS. Comparing mortality rates on CAPD/CCPD and hemodialysis. The Canadian experience: fact or fiction? Perit Dial Int 1998; 18: Ganesh SK, Hulbert-Shearon T, Port FK et al. Mortality differences by dialysis modality among incident ESRD patients with and without coronary artery disease. J Am Soc Nephrol 2003; 14: Just PM, de Charro FT, Tschosik EA et al. Reimbursement and economic factors influencing dialysis modality choice around the world. Nephrol Dial Transplant 2008; 23: U.S. Renal Data System, USRDS 2012 Annual Data Report. Atlas of Chronic Kidney Disease and End-Stage Renal Disease in the United States. National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases: Bethesda, MD van de Luijtgaarden MW, ordzij M, Stel VS et al. Effects of comorbid and demographic factors on dialysis modality choice and related patient survival in Europe. Nephrol Dial Transplant 2011; 26: Sanabria M, Munoz J, Trillos C et al. Dialysis outcomes in Colombia (DOC) study: a comparison of patient survival on peritoneal dialysis vs. hemodialysis in Colombia. Kidney Int Suppl 2008; 73: S165 S Moist LM, Port FK, Orzol SM et al. Predictors of loss of residual renal function among new dialysis patients. J Am Soc Nephrol 2000; 11: Dhingra RK, Young EW, Hulbert-Shearon TE et al. Type of vascular access and mortality in U.S. hemodialysis patients. Kidney Int 2001; 60: McClellan W, Aronoff SL, Bolton WK et al. The prevalence of anemia in patients with chronic kidney disease. Curr Med Res Opin 2004; 20: Stenvinkel P, Heimbürger O, Paultre F et al. Strong association between malnutrition, inflammation, and atherosclerosis in chronic renal failure. Kidney Int 1999; 55: Valderrabano F, Jofre R, Lopez-Gomez JM. Quality of life in end-stage renal disease patients. Am J Kidney Dis 2001; 38: Twardowski ZJ, Misra M, Singh AK. Con: Randomized controlled trials (RCT) have failed in the study of dialysis methods. Nephrol Dial Transplant 2013; 28: Charlson ME, Pompei P, Ales KL et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987; 40: Quan H, Sundararajan V, Halfon P et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005; 43: D Agostino RB Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med 1998; 17: Rubin DB, Thomas N. Combining propensity score matching with additional adjustments for prognostic covariates. J Am Stat Assoc 2000; 95: Kidney International (2014) 86,