Associations of Dialysis Modality and Infectious Mortality in Incident
Dialysis Patients in Australia and New Zealand David W. Johnson, PhD, 1,2 Hannah Dent, BSc(Hons), 1,3 Carmel M. Hawley, MMedSci, 1,2 Stephen P. McDonald, PhD, 1,4 Johan B. Rosman, MD, 1,5 Fiona G. Brown, PhD, 1,6 Kym M. Bannister, MD, 1,7 and Kathryn J. Wiggins, MD 1,8 Background: The aim of the present investigation is to compare rates, types, causes, and timing of infectious death in incident peritoneal dialysis (PD) and hemodialysis (HD) patients in Australia and New Zealand. Study Design: Observational cohort study using the Australian and New Zealand Dialysis and Transplant Registry data. Setting & Participants: The study included all patients starting dialysis therapy between April 1, 1995, and December 31, 2005. Predictor: Dialysis modality. Outcomes &Measurements: Rates of and time to infectious death were compared by using Poisson regression, Kaplan-Meier, and competing risks multivariate Cox proportional hazards model analyses. Results: 21,935 patients started dialysis therapy (rst treatment PD, n 6,020; HD, n 15,915) during the study period, and 1,163 patients (5.1%) died of infectious causes (PD, 529 patients; 7.6% versus HD, 634 patients; 4.2%). Incidence rates of infectious mortality in PD and HD patients were 2.8 and 1.7/100 patient-years, respectively (incidence rate ratio PD versus HD, 1.66; 95% condence interval [CI], 1.47 to 1.86). After performing competing risks multivariate Cox analyses allowing for an interaction between time on study and modality because of identied nonproportionality of hazards, PD consistently was associated with increased hazard of death from infection compared with HD after 6 months of treatment (6 months hazard ratio [HR], 1.08; 95% CI, 0.76 to 1.54; 6 months to 2 years HR, 1.31; 95%CI, 1.09 to 1.59; 2 to 6 years HR, 1.51; 95%CI, 1.26 to 1.80; 6 years HR, 2.76; 95%CI, 1.76 to 4.33). This increased risk of infectious death in PDpatients was largely accounted for by an increased risk of death caused by bacterial or fungal peritonitis. Limitations: Patients were not randomly assigned to their initial dialysis modality. Residual confound- ing and coding bias could not be excluded. Conclusions: Dialysis modality selection signicantly inuences risks, types, causes, and timing of fatal infections experienced by patients with end-stage kidney disease in Australia and New Zealand. Am J Kidney Dis 53:290-297. 2009 by the National Kidney Foundation, Inc. INDEX WORDS: Bacteremia; continuous ambulatory peritoneal dialysis; hemodialysis; peritoneal dialysis; peritonitis; pneumonia; septicemia; bacterial infection; fungal infection; incidence; prevalence; treatment modality; viral infection. I nfection is a major cause of morbidity and the second leading cause of death in dialysis populations. 1,2 The most common types of fatal infections recorded in dialysis registries are sep- ticemia, peritonitis, pneumonia, wound infec- tions, and urosepsis. 1,3 Some studies have ob- served greater rates of septicemia, 4 endocarditis, 5 and pneumonia 1 in patients treated with hemodi- alysis (HD) compared with peritoneal dialysis (PD). However, other studies have not ob- Fromthe 1 Australia and NewZealand Dialysis and Trans- plant Registry, Adelaide; 2 Department of Renal Medicine, University of Queensland at Princess Alexandra Hospital, Brisbane; 3 Discipline of Public Health, University of Ad- elaide; 4 Department of Nephrology and Transplantation Services, University of Adelaide at the Queen Elizabeth Hospital, Adelaide, Australia; 5 Renal Department, Middle- more Hospital, Otahuhu, Auckland, New Zealand; 6 Depart- ment of Nephrology, Monash Medical Center, Clayton, Victo- ria; 7 Department of Nephrology, Royal Adelaide Hospital, Adelaide; and 8 University of Melbourne Department of Ne- phrology, St Vincents Hospital, Fitzroy, Victoria, Australia. Received February 25, 2008. Accepted in revised form July 7, 2008. Originally published online as doi: 10.1053/j.ajkd.2008.06.032 on September 22, 2008. Address correspondence to David W. Johnson, PhD, Department of Nephrology, Level 2, ARTS Bldg, Princess Alexandra Hospital, Ipswich Rd, Woolloongabba, Bris- bane Qld 4102, Australia. E-mail: david_johnson@health. qld.gov.au 2009 by the National Kidney Foundation, Inc. 0272-6386/09/5302-0013$36.00/0 doi:10.1053/j.ajkd.2008.06.032 American Journal of Kidney Diseases, Vol 53, No 2 (February), 2009: pp 290-297 290 served an appreciable impact of dialysis modal- ity on infection type, 6 and the US Renal Data System (USRDS) Registry has reported no signicant difference in septicemia rates be- tween PD and HD patients. 6 A recent North American single-center study reported that di- alysis modality was not an independent predic- tor of overall infection rate in a cohort of incident dialysis patients, but was a strong predictor of type of infection that they experi- enced. 7 Specically, HD was associated with a greater risk of bacteremia, whereas PD was associated with a greater risk of peritonitis. Some of the apparent disparity in ndings between these different studies may be caused by their signicant limitations, including single- center design, data coding ambiguity, use of outdated data, dialysis modality selection bias, lack of adjustment for demographic and clini- cal variables, and residual confounding. The aim of the present study is to evaluate the effects of dialysis modality on frequency, types, and causes of fatal infections in a large incident end-stage renal disease (ESRD) population. METHODS Study Population The study included all patients with ESRD in Australia and New Zealand who were older than 18 years and started dialysis therapy between April 1, 1995, and December 31, 2005. Follow-up continued until December 31, 2005. Complete details of the structure and methods of the Australia and New Zealand Dialysis and Transplant (ANZDATA) Registry have been reported elsewhere. 8 Avail- able data for this study consisted of information about the underlying cause of ESRD, demographic details, a limited range of comorbidities (presence of coronary artery disease, peripheral vascular disease, cerebrovascular disease, chronic lung disease, diabetes mellitus, and smoking), body mass index, and whether the patient had started dialysis therapy within 3 months of referral to a nephrologist. The primary outcome measure was infectious death after starting dialysis therapy. Cause of infectious death was reported to the registry by the patients attending nephrologist according to the site of fatal infection (lung, peritoneum, septicemia, wound, stula, urinary tract, liver, central nervous system, or other) and its microbiological cause (viral, bacterial, fungal, protozoal, or not specied/identied). Septicemia denoted bloodstream infection with an organism (ie, bacteremia or fungemia) in which the primary site of infection was not known. No information was collected regarding labora- tory parameters, antibiotic sensitivities, or antibiotics or other treatments used. Statistical Analysis Results are expressed as frequencies and percentages for categorical variables, mean SD for continuous normally distributed variables, and median and interquartile range for continuousnonnormallydistributedvariables. Baselineanaly- ses were carried out by dividing patients into 2 groups according to whether they were receiving PD or HD at rst treatment (day 0). Differences between the 2 groups were analyzed by using 2 test for categorical data, unpaired t-test for continuous parametric data, and Mann-Whitney test for continuous nonparametric data. Comparison of rates of infectious mortality between PD and HD patients was per- formed by using Poisson regression and presented as inci- dence rate ratio (IRR) with 95% condence interval (CI). Time to infectious death was evaluated by using multivariate Cox proportional hazards survival analyses using a compet- ing risks approach. This involved tting the Cox model on an augmented data set, as described by Lunn and McNeil. 9 A shared frailty was included to allow for possible center effect. Dialysis modality was included in the model as a time-varying covariate. Adelay of 60 days was allowed after a change in dialysis modality so that deaths occurring within 60 days of a change were attributed to the previous modality. Covariates included in the model were age, sex, racial origin, body mass index, late referral (referral to nephrolo- gist within 3 months of starting renal replacement therapy), smoking status (never/former or current), chronic lung dis- ease, coronary artery disease, cerebrovascular disease, periph- eral vascular disease, diabetes mellitus, country of treatment (Australia or New Zealand), center size (based on quartiles of numbers of patients), and dialysis vintage. Data were censored for kidney transplantation, recovery of kidney function, loss to follow-up, and end of study (December 31, 2005). Proportional hazards assumptions were checked by using Schoenfeld residuals and scaled Schoenfeld residuals, examined by means of formal hypothesis test and graphi- cally. First-order interaction terms between signicant covari- ates were examined for all models. There was no evidence of an interaction between modality and dialysis vintage. Cumu- lative incidence was used to estimate probabilities of spe- cic types of infectious death because this approach takes into account the presence of competing risks with deaths from other types of infection. Nonoverlapping 95% con- dence bands indicate a signicant difference (at the 5% level) in cumulative incidence between the 2 treatment groups. Data were analyzed using the software packages SPSS for Windows, release 12.0 (SPSS Inc, North Sydney, Australia) and Stata/SE 9.2 (Stata Corp, College Station, TX). P less than 0.05 is considered statistically signicant. RESULTS Baseline Characteristics During the study period, 21,935 patients started dialysis therapy in Australia and New Zealand (rst treatment PD, n 6,020; HD, n 15,915). Total study follow-up was 57,065 person-years (PD, 19,100 person-years; HD, 37,965 person- years). Compared with patients starting dialysis Dialysis Modality and Infection 291 on HD therapy, patients starting on PD therapy were more likely to be older, women, diabetic, nonobese, treated in New Zealand, and referred to a nephrologist more than 3 months before starting renal replacement therapy (Table 1). Pro- portions of patients who had changed modality for PD versus HD were 8.5% versus 21.1% at 6 months, 27.9% versus 24.7% at 2 years, and 63.6% versus 26.9% at 6 years. Although PD patients were more likely to change modality at least once, patients starting HD therapy were more likely to change modality within the rst 6 months of treatment. Statistically signicant dif- ferences were observed in comorbid illness bur- den between the 2 groups, although the magni- tudes of these differences were small. The prevalence of automated PD use in PD patients during the study period was 15.7%. Infectious Mortality A total of 1,163 patients (5.1%) died of infec- tious causes during the study period (PD, 529 patients; 7.6% versus HD, 634 patients; 4.2%). Incidence rates of infectious mortality in PD and HD patients were 2.8 and 1.7/100 patient-years, respectively (unadjusted IRR for PD versus HD, 1.66; 95% CI, 1.47 to 1.86). When cumulative incidence functions were calculated for infec- Figure 1. Cumulative incidence with 95% condence bands of deaths in peritoneal dialysis (PD) and hemodialy- sis (HD) patients from infection. Table 1. Baseline Characteristics of the Study Populations Characteristic Peritoneal Dialysis (n 6,020) Hemodialysis (n 15,915) P Age (y) 62.7 (51.0-71.3) 60.4 (47.8-70.8) 0.001 Women 2,828 (47%) 6,240 (39%) 0.001 White 4,400 (73%) 11,912 (75%) 0.008 Body mass index 0.001 Underweight 220 (4%) 709 (5%) Normal 2,492 (42%) 6,415 (41%) Overweight 2,080 (35%) 4,937 (31%) Obese 1,212 (20%) 3,771 (24%) Late referral 986 (17%) 4,447 (28%) 0.001 Current smoker 724 (12%) 2,246 (14%) 0.001 Chronic lung disease 829 (14%) 2,621 (16%) 0.001 Coronary artery disease 2,448 (41%) 6,323 (40%) 0.2 Peripheral vascular disease 1,709 (28%) 4,171 (26%) 0.001 Cerebrovascular disease 973 (16%) 2,347 (14%) 0.009 Diabetes mellitus 2,430 (40%) 5,856 (37%) 0.001 New Zealand residence 1,572 (26%) 2,443 (15%) 0.001 Vintage 0.001 1995-1997 1,393 (23%) 3,183 (20%) 1998-2000 1,649 (27%) 4,374 (27%) 2001-2003 1,886 (31%) 4,959 (31%) 2004-2005 1,092 (18%) 3,399 (21%) Never change modality 4,069 (68%) 12,142 (76%) 0.001 Center size (no. of patients) 0.001 Small (340) 1,196 (20%) 4,376 (28%) Small-medium (340-610) 1,337 (22%) 4,012 (25%) Medium-large (611-740) 1,755 (29%) 3,143 (20%) Large (740) 1,732 (29%) 4,384 (28%) Johnson et al 292 tious deaths, 95% condence bands did not over- lap after 2 years, conrming that PD therapy was associated with an increased risk of death from infection after 2 years of treatment (Fig 1). On competing risks multivariate Cox proportional hazards model analysis, cause-specic hazard ratios (HRs) of both fatal infection and death from other causes between PD and HD patients were found to be nonconstant over time (ie, nonproportional hazards). The proportional haz- ards assumption was also violated for body mass index, late referral to a nephrologist, and current smoking status. To deal with nonproportional hazards, the follow-up period was divided into survival during the rst 6 months of treatment, between 6 months and 2 years of treatment, between 2 and 6 years of treatment, and after 6 years of treatment. In each of these periods, the proportional hazards assumption was veried and interaction terms between time on study and each of the covariates showing nonproportional hazards were included in the model to obtain HRs for each period. Using this model, the hazard of death from infection in PD patients was not signicantly different from that in HD patients in the rst 6 months of treatment (6 months HR, 1.08; 95% CI, 0.76 to 1.54), but was consistently and signicantly increased in PD patients compared with HD patients after 6 months of treatment (6 months to 2 years HR; 1.31; 95% CI, 1.09 to 1.59; 2 to 6 years HR, 1.51; 95% CI, 1.26 to 1.80; 6 years HR; 2.76, 95% CI, 1.76 to 4.33). In addition to dialysis modality, the other consistent predictors of increased risk of infectious death were Aboriginal and Torres Strait Islander origin, current smoking, chronic lung disease, treated in Australia, and center size. Causes of Infectious Mortality Causes of infectious death are listed in Table 2. Peritonitis was signicantly more common as a cause of death in PD patients compared with HD patients (IRR, 56.9; 95% CI, 23.3 to 139). There were no signicant differences between PD and HD patients in relative incidences of septicemia, pneumonia, wound infections, or other infec- tions. In the Cox analysis (using competing risk cumulative incidence functions), the excess infec- tious mortality associated with PD compared with HD therapy was explained by an augmented risk of peritonitis (Fig 2). Cumulative incidences of deaths from peritonitis in PD and HD patients were 0.012 and 0.001 at 2 years and 0.047 and 0.002 at 6 years, respectively. There were no signicant differences between PD and HD pa- tients in cumulative incidences of septicemia, pneumonia, wound infections, or other infec- tions (data not shown). Of 1,163 infectious deaths registered, patho- genic microorganisms were identied and reported Figure 2. Cumulative incidence with 95% condence bands of deaths from peritonitis in peritoneal dialysis (PD) and hemodialysis (HD) patients. Table 2. Outcomes of Hemodialysis and Peritoneal Dialysis (modality at death or censoring) Patients in Australia and New Zealand, 1995 to 2005 Parameter Peritoneal Dialysis (n 6,943) Hemodialysis (n 14,992) Follow-up (y) 19,100 37,965 Total deaths 3,431 (17.96) 5,571 (14.67) Infectious deaths Septicemia 117 (0.61) 239 (0.63) Peritonitis 220 (1.1) 21 (0.06) Pneumonia 84 (0.44) 163 (0.43) Wound infection 38 (0.20) 67 (0.18) Fistula infection 1 (0.01) 17 (0.04) Urosepsis 4 (0.02) 5 (0.01) Central nervous system 5 (0.03) 9 (0.02) Liver 2 (0.01) 3 (0.01) Other 58 (0.30) 113 (0.29) Total 529 (2.76) 634 (1.67) Modality change 1,931 (32%) 3,748 (24%) Kidney transplantation 1,111 (10%) 2,810 (14%) Lost to follow-up 5 (0.1%) 19 (0.1%) Recovery of kidney function 4 (0.04%) 111 (0.29%) Note: Death outcomes expressed as number (incidence rate/100 patient-years). Dialysis Modality and Infection 293 in 368 PD patients and 480 HD patients. Recorded causes of fatal fungal and bacterial infections are listed in Tables 3 and 4, respectively. Rates of fatal fungal infection were signi- cantly increased in PD patients compared with HD patients (IRR, 5.19; 95% CI, 3.02 to 8.94). This increased risk was largely accounted for by an increased risk of fatal fungal peritonitis in PD patients (IRR, 25.85; 95% CI, 8.22 to 130.74). The cumulative incidence of death from fungal infections (Fig 3) was signicantly greater for PD patients after 2.5 years of treatment. At 2 years, cumulative incidences were 0.003 and 0.001 for PD and HD patients and 0.009 and 0.001 at 6 years, respectively. Rates of fatal bacterial infections were also increased in patients on PD therapy at the time of death (IRR, 1.52; 95% CI, 1.33 to 1.74). Again, this increased risk was largely accounted for by an increased risk of fatal bacterial peritonitis in PDpatients (IRR, 20.38; 95%CI, 12.21 to 36.50). The cumulative incidence of death from bacterial infections (Fig 4) was signicantly greater in PD patients after 3 years of treatment. Cumulative incidences for PD and HD patients were 0.023 and 0.019 at 2 years and 0.081 and 0.052 at 6 years, respectively. There were no signicant differences ob- served in cumulative incidences of deaths from Figure 4. Cumulative incidence with 95% condence bands of deaths from bacterial infections in peritoneal dialysis (PD) and hemodialysis (HD) patients. Table 3. Fatal Fungal Infections in Hemodialysis and Peritoneal Dialysis Patients in Australia and New Zealand, 1995 to 2005 Site of Fatal Fungal Infection Peritoneal Dialysis (19,100 patient- years) Hemodialysis (37,965 patient- years) Fungemia 4 (0.02) 1 (0.003) Peritonitis 39 (0.20) 3 (0.01) Pneumonia 2 (0.01) 9 (0.02) Wound infection 1 (0.01) 1 (0.003) Fistula infection 0 (0) 0 (0) Urosepsis 0 (0) 1 (0.003) Central nervous system 1 (0.01) 2 (0.01) Liver 0 (0) 0 (0) Other 0 (0) 1 (0.003) Total 47 (0.25) 18 (0.05) Note: Deaths expressed as number (incidence rate/100 patient-years). Table 4. Fatal Bacterial Infections in Hemodialysis and Peritoneal Dialysis (modality at death) Patients in Australia and New Zealand, 1995 to 2005 Site of Fatal Bacterial Infection Peritoneal Dialysis (19,100 patient- years) Hemodialysis (37,965 patient- years) Septicemia 72 (0.38) 180 (0.47) Peritonitis 165 (0.86) 16 (0.04) Pneumonia 44 (0.23) 100 (0.26) Wound infection 31 (0.16) 64 (0.17) Fistula infection 1 (0.01) 15 (0.04) Urosepsis 4 (0.02) 4 (0.01) Central nervous system 2 (0.01) 3 (0.01) Liver 1 (0.01) 2 (0.01) Other 48 (0.25) 96 (0.25) Total 368 (1.92) 480 (1.26) Note: Deaths expressed as number (incidence rate/100 patient-years). Figure 3. Cumulative incidence with 95% condence bands of deaths from fungal infections in peritoneal dialy- sis (PD) and hemodialysis (HD) patients. Johnson et al 294 viral infections or deaths caused by an organism that was not identied or specied (data not shown). DISCUSSION The present study shows that PD treatment was associated with an increased risk of death from infection compared with HD. This excess risk was accounted for by an increased occur- rence of fatal peritonitis. No signicant differ- ences were observed between PD and HD pa- tients with respect to cumulative incidences of fatal pneumonia, septicemia, or other infections. These results differ from those of a recent single-center North American study of 119 HD and 62 PD patients in which dialysis modality was signicantly associated with type of infec- tion, but was not predictive of overall infection rates. 7 Specically, peritonitis occurred exclu- sively in PD patients and bacteremia occurred exclusively in HD patients. There was a trend toward increased overall infection risk with PD (relative risk, 1.30; 95% CI, 0.93 to 1.8; P 0.1), such that the negative ndings compared with those of our much larger study may simply reect a lack of statistical power in the former. Our studys ndings were within the broad CIs from this study. In keeping with the ndings of Aslam et al, 7 the risk of fatal peritonitis in our study was greatly increased in PD patients compared with HD patients. A novel nding was that the risk was not proportional over time, but increased with increasing duration of dialysis therapy. The explanation for this nding was not able to be determined from our study, but may be related to decreasing residual kidney function in PD pa- tients, cumulative peritoneal exposure to bioin- compatible uids resulting in impaired host de- fense against infection, establishment of biolms in PD catheters acting as a nidus for infection, and a progressive decrease in use of HD cath- eters with increasing time on HD therapy be- cause of the creation of permanent vascular ac- cess. Moreover, the excess risk of infectious death in PD patients was explained primarily by the occurrence of bacterial and fungal peritonitis. Although there is limited evidence supporting oral nystatin coadministration with antibiotic courses in PD patients to prevent fungal peritoni- tis 10,11 and topical antimicrobial prophylaxis (eg, mupirocin and gentamicin) 11 to prevent bacterial peritonitis, the extent to which Australasian units had adopted this practice is not known. A recent Australasian survey suggested that nearly 70% of units do not routinely use recommended infec- tion control measures in PD patients (D.W.J., unpublished data). Moreover, treatment proto- cols for patients with peritonitis in Australian and New Zealand renal units are unknown, al- though delayed catheter removal has been identi- ed as a key contributing factor to peritonitis- related deaths. 12 The overall peritonitis rates observed in Australia (1 episode/20.3 patient- months) 3 and New Zealand (1 episode/17.0 pa- tient-months) 3 were more frequent than those reported in Canada (1 episode/27.6 patient- months), 13 the United States (1 episode/32.7 patient-months), 13 and many Asian countries. 14 However, the reported death rate from peritonitis in Australia and New Zealand (5%; K.M.B., unpublished data) was similar to that noted in other countries (3% to 6%). 13 Results of our study therefore are likely to still be generalizable to other countries. The inuence of dialysis modality on septice- mia in patients with ESRD is more contentious than for peritonitis, with some studies reporting a clearly increased risk in HDpatients, 5,7,15 whereas others describe similar rates of bacteremia be- tween the 2 dialysis modalities. 6 Abbott and Agodoa 5,15 identied HD (versus PD) as a strong independent risk factor for both septicemia and endocarditis in 327,993 incident dialysis pa- tients. Conversely, Powe et al 6 conducted a longi- tudinal cohort study of 4,918 incident patients with ESRD and found that although the rate of septicemia was greater for HD patients (11.7%) than PD patients (9.4%) during the 7 years of follow-up, the difference was not statistically signicant for any period. Some of the apparent disparity may be explained by data-coding ambi- guity, given that sepsis and septicemia may be confused with peritonitis by coding physi- cians. 7,16 Such ambiguity was avoided in the present study by the use of more specic terms (bacteremia and peritonitis) in the ANZDATA Registry. It also is likely that the differential effect of dialysis modality on bacteremia rates is inuenced by the prevalence of dialysis catheter use in HD populations. 17 Arecent analysis of the Dialysis Morbidity and Mortality Wave 2 Study Dialysis Modality and Infection 295 (DMMS Wave 2) 18 identied initial dialysis ac- cess as the main antecedent of bacteremia or septicemia. The HR for PD catheters was not signicantly different from those for arterio- venous grafts or native arteriovenous stulae, but was substantially less than those for perma- nent or temporary HD catheters. Other studies have observed that the risk of bacteremia progres- sively increases from native stulae to grafts to dialysis catheters. 19-21 HD access type was not specically able to be examined in the present study, although the overall prevalence of HD catheter use in Australia (4%) 22 and New Zea- land (24%) 23 is similar to or less than that re- ported in other countries (18% to 33%). 24,25 In contrast to the USRDS data, 1 our study did not observe a lower risk of fatal pneumonia in PD compared with HD patients on univariate analysis. Reasons for this nding are uncertain, but may reect patient selection because PD patients in Australia and New Zealand were signicantly less likely to have preexisting chronic lung disease, although the absolute differ- ence in frequencies was small (14% versus 16%; P 0.001). The strengths of this study include its very large sample size and inclusiveness and robust- ness of ndings across different statistical meth- ods. We included all patients who began dialysis therapy in Australia or New Zealand during the period; thus, a variety of centers were included with varying approaches to the use of dialysis modalities and varying rates of transplantation. This greatly enhanced the external validity of our ndings. Moreover, our cohort consisted solely of incident patients, thereby avoiding the poten- tially confounding factor of survivor bias associ- ated with prevalent population studies. The time span of the cohort was contemporaneous, which was of particular relevance because selection criteria and practices may have changed over time in both HD and PD therapy. Appropriate analyses were performed to cater for nonpropor- tional hazards, and competing risks analyses were undertaken. Dialysis modality was handled as a time-varying covariate. A key difference of our registry cohort from that of others, such as the USRDS, is the high penetration of PDtherapy (41%) and the higher risk prole of the PD group, which helped minimize error bias related to nonrandom dialysis modality selection. Nevertheless, the study has a number of limita- tions. The strength of registry analyses with respect to their extent of coverage must be bal- anced against their main weakness, which is limited depth of coverage. ANZDATA does not collect detailed information for PD or HD pre- scription, patient compliance, nonfatal infection data, HD catheter use, hospitalizations, indi- vidual unit management protocols, laboratory values, residual urine output, or severity of comor- bidities; therefore, unidentied associations could not be entirely excluded. Although we adjusted for a large number of patient characteristics, the possibility of residual confounding also cannot be excluded. In common with other registries, ANZDATAis a voluntary registry and there is no external audit of data accuracy (including coding of cause of death). Consequently, the possibility of coding/classication bias cannot be excluded. 26 Although we included all patients who started dialysis therapy as their rst renal replacement therapy modality, overall treated ESRD inci- dence rates in Australia currently are around 100/million/year, substantially less than in the United States and slightly less than in many European countries, although broadly similar to rates in other countries, such as Sweden, The Netherlands, and Poland. 1 Differences in patient characteristics might explain the differing re- sults; alternatively, it might be that there are systematic differences in treatment outcomes be- tween countries. Although there are no published data that allow comparisons of PD mortality rates, there is substantial variation in HD mortal- ity rates that is not explained by measured comor- bidities. 27 In conclusion, the present study shows that dialysis modality is an independent predictor of rates, types, causes, and times of occurrence of infectious death in incident patients with ESRD. This excess risk is caused primarily by a height- ened risk of peritonitis, which increases with increasing time on dialysis therapy. Although the increased risk of infection-related mortality in PD patients in our study should not be inter- preted as an endorsement of HDover PDtherapy, these study ndings should form part of the basis for informing shared decision making about di- alysis modality selection in conjunction with patient preference, individual circumstances, lo- cal infection rate data, unit practice patterns Johnson et al 296 (including HD catheter use rates), and the other well-characterized pros and cons of selecting PD versus HD therapy. Further attention to infection control management in PD patients, especially antimicrobial prophylaxis measures, would also appear warranted. ACKNOWLEDGEMENTS The authors gratefully acknowledge the substantial contri- butions of the entire Australian and New Zealand nephrol- ogy community (physicians, surgeons, database managers, nurses, renal operators, and patients) in providing informa- tion for and maintaining the ANZDATARegistry databas. Support: None. Financial Disclosure: Professor Johnson is a consultant for Baxter Healthcare Pty Ltd and has previously received research funds from this company. He has also received speakers honoraria and research grants from Fresenius Medical Care. Dr Bannister is a consultant for Baxter Healthcare Pty Ltd. Dr McDonald has received speaking honoraria from AMGEN Australia, Fresenius Australia, and Solvay Pharmaceuticals and travel grants from AMGEN Australia, Genzyme Australia, and Jansen-Cilag. REFERENCES 1. US Renal Data System: USRDS 2006 Annual Data Report: The National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, 2006 2. Johnson DW, McDonald SP, Excell L, Livingston B, Shtangey V: Peritoneal dialysis, in McDonald SP, Excell L (eds): ANZDATA Registry Report 2005. Adelaide, South Australia, ANZDATA, 2006, pp 84-100 3. Johnson DW, Chang S, Excell L, Livingston B, Bannis- ter K, McDonald SP: Peritoneal dialysis, in McDonald SP, Excell L (eds): ANZDATA Registry Report 2006. Adelaide, South Australia, ANZDATA, 2007, pp 87-103 4. Foley RN, Guo H, Snyder JJ, Gilbertson DT, Collins AJ: Septicemia in the United States dialysis population, 1991 to 1999. J Am Soc Nephrol 15:1038-1045, 2004 5. Abbott KC, Agodoa LY: Hospitalizations for bacterial endocarditis after initiation of chronic dialysis in the United States. Nephron 91:203-209, 2002 6. Powe NR, Jaar B, Furth SL, Hermann J, Briggs W: Septicemia in dialysis patients: Incidence, risk factors, and prognosis. Kidney Int 55:1081-1090, 1999 7. Aslam N, Bernardini J, Fried L, Burr R, Piraino B: Comparison of infectious complications between incident hemodialysis and peritoneal dialysis patients. Clin J Am Soc Nephrol 1:1226-1233, 2006 8. McDonald SP, Chang S, Excell L: ANZDATARegistry Report 2007. Adelaide, Australia, Australian and New Zea- land Dialysis and Transplant Registry, 2007 9. Lunn M, McNeil D: Applying Cox regression to com- peting risks. Biometrics 51:524-532, 1995 10. Strippoli GF, Tong A, Johnson D, Schena FP, Craig JC: Antimicrobial agents for preventing peritonitis in perito- neal dialysis patients. Cochrane Database Syst Rev CD004679, 2004 11. Strippoli GF, Tong A, Johnson D, Schena FP, Craig JC: Antimicrobial agents to prevent peritonitis in peritoneal dialysis: A systematic review of randomized controlled tri- als. Am J Kidney Dis 44:591-603, 2004 12. Digenis GE, Abraham G, Savin E, et al: Peritonitis- related deaths in continuous ambulatory peritoneal dialysis (CAPD) patients. Perit Dial Int 10:45-47, 1990 13. Mujais S: Microbiology and outcomes of peritonitis in North America. Kidney Int Suppl 103:S55-S62, 2006 14. Blake PG: Peritoneal dialysis in Asia: An external perspective. Perit Dial Int 22:258-264, 2002 15. Abbott KC, Agodoa LY: Etiology of bacterial septice- mia in chronic dialysis patients in the United States. Clin Nephrol 56:124-131, 2001 16. Aslam N, Bernardini J, Fried L, Piraino B: Large body mass index does not predict short-term survival in peritoneal dialysis patients. Perit Dial Int 22:191-196, 2002 17. Polkinghorne KR, McDonald SP, Atkins RC, Kerr PG: Vascular access and all-cause mortality: A propensity score analysis. J Am Soc Nephrol 15:477-486, 2004 18. Ishani A, Collins AJ, Herzog CA, Foley RN: Septice- mia, access and cardiovascular disease in dialysis patients: The USRDS Wave 2 Study. Kidney Int 68:311-318, 2005 19. Taylor G, Gravel D, Johnston L, Embil J, Holton D, Paton S: Incidence of bloodstream infection in multicenter inception cohorts of hemodialysis patients. Am J Infect Control 32:155-160, 2004 20. Taylor G, Gravel D, Johnston L, Embil J, Holton D, Paton S: Prospective surveillance for primary bloodstream infections occurring in Canadian hemodialysis units. Infect Control Hosp Epidemiol 23:716-720, 2002 21. Colville LA, Lee AH: Retrospective analysis of cath- eter-related infections in a hemodialysis unit. Infect Control Hosp Epidemiol 27:969-973, 2006 22. Polkinghorne KR, McDonald SP, Atkins RC, Kerr PG: Epidemiology of vascular access in the Australian hemodialysis population. Kidney Int 64:1893-1902, 2003 23. Polkinghorne KR, McDonald SP, Marshall MR, At- kins RC, Kerr PG: Vascular access practice patterns in the New Zealand hemodialysis population. Am J Kidney Dis 43:696-704, 2004 24. Di BenedettoA, Basci A, Cesare S, Marcelli D, Ponce P, Richards N: Increased use of catheters as vascular access: Is it justied by patients clinical conditions? J Vasc Access 8:21-27, 2007 25. Mendelssohn DC, Ethier J, Elder SJ, Saran R, Port FK, Pisoni RL: Haemodialysis vascular access problems in Canada: Results from the Dialysis Outcomes and Practice Patterns Study (DOPPS II). Nephrol Dial Transplant 21:721- 728, 2006 26. Li SQ, Cass A, Cunningham J: Cause of death in patients with end-stage renal disease: Assessing concor- dance of death certicates with registry reports. Aust N Z J Public Health 27:419-424, 2003 27. Goodkin DA, Bragg-Gresham JL, Koenig KG, et al: Association of comorbid conditions and mortality in hemo- dialysis patients in Europe, Japan, and the United States: The Dialysis Outcomes and Practice Patterns Study (DOPPS). J Am Soc Nephrol 14:3270-3277, 2003 Dialysis Modality and Infection 297