PREDICTORS OF PERITONITIS AMONG CANADIAN PERITONEAL DIALYSIS PATIENTS

Transcription

1 PREDICTORS OF PERITONITIS AMONG CANADIAN PERITONEAL DIALYSIS PATIENTS By Sharon J. Nessim, MD A thesis submitted in conformity with the requirements for the degree of Master of Science Graduate Department of the Institute of Medical Science University of Toronto Copyright by Sharon J. Nessim (2009)

2 Predictors of Peritonitis among Canadian Peritoneal Dialysis Patients Sharon J. Nessim Master of Science, Institute of Medical Science, University of Toronto, 2009 Abstract Despite the decreasing incidence of peritoneal dialysis (PD) peritonitis over time, its occurrence is still associated with adverse outcomes. This thesis focuses on determining factors associated with PD peritonitis in order to facilitate identification of patients at risk. Using data collected in a multicentre Canadian database between 1996 and 2005, the study population comprised 4,247 incident PD patients, of whom 1,605 had at least one peritonitis episode. Variables independently associated with peritonitis included age [rate ratio (RR) 1.04 per decade increase, 95% CI ], Black race (RR 1.37, 95% CI ) and having transferred from hemodialysis (RR 1.24, 95% CI ). There was an interaction between gender and diabetes (p=0.011), with an increased peritonitis risk only among female diabetics (RR 1.27, 95% CI ). Choice of continuous ambulatory PD vs. automated PD did not influence peritonitis risk. These results contribute to our understanding of peritonitis risk among PD patients. ii

3 Acknowledgements I would like to thank several people, without whom this thesis would not have been possible. First and foremost, I am extremely grateful to my supervisor, Dr. Vanita Jassal, for her invaluable guidance, advice and support throughout my Master s degree. I am also indebted to my thesis committee members, Dr. Joanne Bargman, Dr. Peter Austin and Dr. Jan Hux, who provided insight into the clinical importance of the project, the study design and the intricacies of the data analysis and interpretation. I thank you all for your interest, your time and your patience. I would also like to thank Dr. Ken Story, Dr. Alex Kriukov and Dr. Rosane Nisenbaum for assistance and advice regarding the statistical methods used in this thesis. This work was generously supported by a Kidney Foundation of Canada Fellowship Award, as well as by an Educational Fellowship from Baxter Healthcare. The University of Toronto Clinician Scientist Program provided additional financial support. iii

4 Table of Contents Abstract ii Acknowledgements iii List of tables vii, viii List of figures..ix List of Abbreviations x 1. INTRODUCTION Rationale Research objectives Hypotheses BACKGROUND End stage renal disease and dialysis options Renal replacement therapy: Peritoneal dialysis vs. hemodialysis Peritoneal dialysis submodalities: CAPD and APD Peritonitis in peritoneal dialysis patients Definition Incidence and outcomes Risk factors common to all PD patients Peritonitis prevention strategies Predictors of peritonitis Current knowledge regarding peritonitis risk Age..19 iv

5 2.3.3 Peritoneal dialysis submodality: CAPD vs. APD Statistical methodology used to study occurrence of peritonitis METHODOLOGY Data sources Patient population Model covariates Outcomes Statistical Analyses RESULTS Patient cohort Independent predictors of PD peritonitis Interactions Era effect Comparison of peritonitis modeling strategies Sensitivity analysis for peritonitis relapse/recurrence exclusion criteria DISCUSSION General discussion Impact on nephrology practice Limitations Conclusions Future directions ILLUSTRATIONS Tables...51 v

6 6.2 Figures REFERENCES APPENDIX...85 vi

7 List of Tables Table 1. Table 2. Table 3. Table 4. Table 5. Table 6. Table 7. Table 8. Table 9. Table 10. Table 11. Table 12. Table 13. Table 14. Table 15. Summary of studies testing the association between age and peritonitis 51 Summary of studies on the association between CAPD vs. APD and peritonitis 52 Distribution of peritonitis episodes within the patient cohort..53 Baseline demographic characteristics for the entire patient cohort..54 Comparison of POET cohort vs. CORR data...55 Multivariable negative binomial model for the outcome of peritonitis...56 Multivariable negative binomial model for the outcome of peritonitis in the subgroup of patients with no submodality switch..57 Multivariable Andersen Gill model for the outcome of peritonitis..58 Multivariable Andersen Gill model for the outcome of peritonitis in the subgroup of patients with no submodality switch.59 Interaction between diabetes and peritonitis by gender..60 Testing for interactions between each variable and era..61 Association between age and peritonitis by era 62 Comparison of results of multivariable negative binomial model and Andersen Gill model for peritonitis..63 Multivariable negative binomial model for the outcome of peritonitis (sensitivity analysis using 45 day relapse/recurrence exclusion criteria).64 Multivariable negative binomial model for the outcome of peritonitis in the subgroup of patients with no submodality switch (sensitivity analysis using 45 day relapse/recurrence exclusion criteria) 65 vii

8 Table 16. Table 17. Multivariable Andersen Gill model for the outcome of peritonitis (sensitivity analysis using 45 day relapse/recurrence exclusion criteria) 66 Multivariable Andersen Gill model for the outcome of peritonitis in the subgroup of patients with no submodality switch (sensitivity analysis using 45 day relapse/recurrence exclusion criteria)...67 viii

9 List of Figures Figure 1. Figure 2. Figure 3. Illustration of intraluminal and periluminal entry of organisms into the peritoneal cavity.68 Flow diagram of patient cohort from POET database..69 Distribution of patients in POET database by province...70 ix

10 List of Abbreviations ANZDATA APD CAPD CI CKD CNS CORR ESRD GN HD HR PD POET RCT RR RRF SD USRDS Australia and New Zealand Dialysis and Transplant Registry Automated peritoneal dialysis Continuous ambulatory peritoneal dialysis Confidence interval Chronic kidney disease Coagulase negative staphylococcus Canadian Organ Replacement Register End stage renal disease Glomerulonephritis Hemodialysis Hazard ratio Peritoneal dialysis Peritonitis Organism Exit sites Tunnel infections Randomized controlled trial Rate ratio Residual renal function Standard deviation United States Renal Data System x

11 INTRODUCTION 1.1 Rationale Despite improvements in the management of patients with chronic kidney disease (CKD), there was a 34% increase in the number of Canadian patients reaching endstage renal disease (ESRD) between 1997 and 2006 (1). When renal transplantation is not an immediate option, the only remaining renal replacement therapy option for patients who reach ESRD is dialysis. There are currently two forms of dialysis available: hemodialysis (HD) and peritoneal dialysis (PD). While HD is the more commonly utilized modality, PD has the advantage of being a home based therapy (relative to HD, in which the majority of patients are required to come to hospital for treatments thrice weekly). One of the most concerning complications associated with PD is infection of the peritoneal space known as peritonitis. To date, little is known about how best to study the occurrence of peritonitis and what factors predict its occurrence. 1

12 1.2 Research Objectives The thesis will address the following research questions: (1) Among incident Canadian peritoneal dialysis patients, what patient and dialysisrelated factors are associated with peritonitis? Specifically: (i) Is increasing age associated with an increased risk of peritonitis? (ii) Does choice of PD submodality (continuous ambulatory PD (CAPD) vs. automated PD (APD)) affect peritonitis risk? (2) Are the results of rate and time to event analyses comparable in the study of the occurrence of peritonitis? 1.3 Hypotheses This thesis seeks to identify whether there are baseline demographic characteristics among incident PD patients that predict the occurrence of peritonitis. Two variables for which the current literature is inconsistent include age and PD submodality. It is hypothesized that increasing age is associated with a higher risk of peritonitis, and that the two submodalities of PD, CAPD and APD, are similar with regard to peritonitis risk. 2

13 It is generally believed that it is vital to incorporate the amount of time spent on dialysis into the way in which peritonitis as an outcome variable is defined, such that the most appropriate analyses for the occurrence of peritonitis would involve defining peritonitis as a either a rate or time to event. We hypothesized that both rate and time to event analyses are effective tools to study peritonitis, and would yield similar predictors of peritonitis with similar risk estimates. 3

14 2. BACKGROUND 2.1 End stage renal disease and dialysis options Renal replacement therapy: Peritoneal dialysis vs. hemodialysis Chronic kidney disease is a growing medical problem in the general population. The increased prevalence of renal disease has largely resulted from the rising life expectancy (2), coupled with an increasing prevalence of diabetes mellitus (3). While the greater awareness of the presence of renal disease among general practitioners has led to earlier referral of patients to nephrologists and better preventive care (4, 5), progression to ESRD remains a major problem. In Canada, the number of individuals starting renal replacement therapy has increased from 3,958 in 1997 to 5,321 in 2006 (1).The optimal mode of renal replacement therapy in these patients is renal transplantation (6). However, given the relatively low living kidney donation rate, the long waiting time for a deceased donor kidney and the ineligibility for transplant in some patients, dialysis is frequently the only available treatment option. There are two dialysis modalities, hemodialysis (HD) and peritoneal dialysis (PD). Hemodialysis is a form of renal replacement therapy in which the patient s blood is passed across a filter that allows for removal of accumulated toxins and electrolytes via diffusion, and removal of excess fluid via ultrafiltration. Using this filter (known as the dialyzer), the dialysis machine attempts to reproduce normal kidney function. In 4

15 order for HD to be effective, patients generally require a minimum of 12 hours per week connected to the machine. This is usually achieved by having patients come to hospital for 4 hour treatment sessions three times per week. Some HD patients can be trained to dialyze themselves at home, but home HD remains an option for only a minority of patients. Access to the patient s bloodstream for these treatments requires some form of vascular access, with options including a tunneled intravenous dialysis catheter or an arteriovenous fistula or graft. Peritoneal dialysis is the other form of dialysis. Rather than passing a patient s blood through an artificial dialyzer as occurs on HD, PD utilizes the patient s own peritoneal membrane as the filter over which diffusion and ultrafiltration occur. A permanent PD catheter is inserted through which an electrolyte balanced, glucose rich dialysis fluid (known as dialysate) is infused into and drained from the peritoneal cavity. Once the dialysate is instilled, there is diffusion of uremic toxins and electrolytes down their concentration gradient from the bloodstream, across the peritoneal membrane and into the fluid. Simultaneously, the high glucose concentration in the dialysate creates an osmotic gradient for the movement of fluid from the bloodstream into the peritoneal cavity. Once sufficient time has passed for diffusion and ultrafiltration to occur, the dialysate is then drained from the peritoneal cavity and fresh dialysate is instilled. The procedure of filling and draining the peritoneal cavity with dialysate is repeated several times per day. 5

16 Survival outcomes for patients on HD vs. PD have been compared in several studies. Some have reported better outcomes with PD (7 9), while others have shown better outcomes with HD (10 12) or no difference between the two modalities (13 16). The inconsistency in the literature may relate to several factors. Firstly, the study populations varied widely, as demographic characteristics of patients differed among Canada, the United States and the European countries in which studies were conducted. Secondly, the time period over which the studies were carried out ranged from as early as 1987 to as recently as 1999, such that advances in dialytic therapies over that time period could have differentially affected outcome. Thirdly, some studies compared incident PD and HD patients, while others focused on prevalent patients. Finally, the follow up time was variable across studies. The latter two considerations are particularly important as it appears that the relationship between dialysis modality and outcome changes over time, such that the survival advantage for PD reported in some studies was only seen in incident cohorts during the first two years after initiation of dialysis (7 9). The biggest limitation in determining the effect of dialysis modality on outcome is the observational nature of the studies used to try to answer this question. As such, they cannot fully account for the fact that patients who go on PD, a home based modality, frequently differ systematically from those who go on in centre HD despite attempts to adjust for potential confounders. In order to answer this question while avoiding systematic bias, an attempt was made to do a large scale randomized controlled trial (RCT) of HD vs. PD (17). Unfortunately, this RCT was unsuccessful because of 6

17 difficulty in recruitment, with only 5% of eligible patients having no preference for PD or HD and agreeing to be randomized to either modality. In the absence of an RCT, physician opinion varies widely. While some physicians have a bias favoring either HD or PD, most physicians believe that HD and PD are equally effective renal replacement therapy options and that the decision should be left to the patient when possible. Hence, the majority of patients who receive pre dialysis care are given information on both modalities, and are offered a choice. In most countries, HD is the most frequently used therapy, although this varies widely by region (3, 18 21). In Canada, 82% of prevalent dialysis patients in 2006 were on HD with the remaining 18% on PD (1). Since PD is a home based therapy, it provides greater independence to patients and allows for less utilization of hospital based resources relative to in centre HD. As a result, in 2005, the Ministry of Health in Ontario made several recommendations in an attempt to increase the prevalence of PD utilization to 30% (22). It is clear that a better understanding of the complications associated with PD would aid in achieving this target Peritoneal dialysis submodalities: CAPD and APD Once a decision is made to initiate PD, a choice must be made between the two forms of PD: continuous ambulatory peritoneal dialysis (CAPD) and automated peritoneal dialysis (APD). 7

18 In CAPD, patients are taught how to manually fill and drain their peritoneal cavity via their PD catheter using aseptic technique. A typical patient performs 4 exchanges daily, usually in the early morning, mid day, evening, and before bedtime. For example, 2 L of dialysate is instilled at 8 am and left to dwell until 12 pm. This fluid is then drained, and another 2 L would be instilled to dwell until 4 pm, and so on. The net result is typically 4 exchanges of fluid, with dialysate in the peritoneal cavity throughout the 24 hour period. In APD, patients use an automated cycler to perform their exchanges during the night. Patients on APD connect their PD catheter to the cycler before going to bed. The cycler, which is programmable to the patient s specifications, then instills dialysate into the peritoneal cavity, and will exchange the specified volume of fluid at preset intervals during the night while the patient sleeps. For example, the cycler may be programmed to provide four 2 L exchanges over a 9 hour period during the night. In the morning, the patient will disconnect from the cycler, and carry on with his or her daily activities. The choice of CAPD vs. APD in some instances may be guided by the transport characteristics of the patient s peritoneal membrane, but is often left to the discretion of the patient based on lifestyle and usual daily activities. In the event that a patient is unable to perform the manual exchanges required for CAPD due to visual or cognitive impairment or impaired manual dexterity, APD is usually the submodality of choice as connections to the cycler can be performed by a family member or visiting nurse. 8

19 2.2 Peritonitis in peritoneal dialysis patients Definition The peritoneal cavity into which dialysate is infused in PD patients is a sterile environment. Infection of the peritoneal space is known as peritonitis. In the general population, peritonitis is an extremely rare occurrence, and usually results from a perforated abdominal viscus with movement of organisms from the bowel lumen into the peritoneal space. In contrast, peritonitis is a well described complication among patients on PD. The ISPD has defined PD Peritonitis as at least 2 of 3 of the following: (i) clinical symptoms or signs suggestive of peritoneal inflammation, (ii) effluent cell count with greater than 100 white blood cells per µl, of which at least 50% are neutrophils, and (iii) a positive effluent Gram stain or culture (23). Entry of organisms into the peritoneal cavity in PD patients can occur via several mechanisms. The two most common mechanisms include introduction of organisms into the lumen of the catheter by touch contamination at the time of catheter connections, and periluminal entry from the exit site along the outside wall of the catheter through the subcutaneous tunnel and into the peritoneal cavity (Figure 1). Peritonitis episodes can also result from transmigration of organisms across the intestinal wall, and rarely from bacteremia with seeding of the peritoneal cavity or transvaginal migration of organisms. 9

20 2.2.2 Incidence and outcomes In the early years after introduction of PD, peritonitis occurred once in every 9 10 patient months (24, 25). Since that time, however, the frequency of peritonitis as a complication has continued to decline (26 30), with current peritonitis rates as low as 1 in every 41 patient months (29). Despite the decreasing peritonitis rates over time, the occurrence of peritonitis remains a concern given its association with adverse outcomes. Specifically, PD peritonitis is associated with increased mortality (31, 32) and hospitalization (33). In addition, PD related infections are the most frequent reason for discontinuation of PD, accounting for 28% of transfers to HD in one study (34). While HD patients do not get peritonitis, those with tunneled dialysis catheters are instead prone to catheter related bacteremia, such that overall infection rates between PD and HD are similar (35). Despite the comparable infection risk, the occurrence of peritonitis and the associated potential for adverse outcomes has led some nephrologists to avoid recommending PD Risk factors common to all PD patients There are several factors common to all PD patients that favor the occurrence of peritonitis. Firstly, all patients have PD catheters that allow for communication 10

21 between the non sterile external environment and the sterile peritoneal cavity. This allows for both intraluminal and periluminal entry of organisms. A second factor common to all PD patients is the use of dialysate containing a high concentration of glucose. Since glucose is an excellent growth medium for bacteria, the introduction of even a small inoculum of organisms may be enough to cause peritonitis. Furthermore, it has been shown that ESRD patients have impaired host immunity (36 39), as well as abnormal peritoneal immune function (40 42). The reason for the impaired peritoneal immunity is thought to be the unphysiologic nature of conventional PD solutions, including their high glucose concentration, hyperosmolarity, acidic ph and the formation of glucose degradation products during the heat sterilization of the dialysate bags. Under normal circumstances, local peritoneal immunity plays an important role in the prevention and clearance of PD peritonitis. When exposed to conventional dialysate, however, there is abnormal leukocyte recruitment in response to inflammatory stimuli (42) and impaired phagocytic function (40) Peritonitis Prevention Strategies With these common risk factors in mind, several modifications to PD practice have been made over time in order to reduce peritonitis risk. The first major advance was the introduction of improved PD connectology. The initial catheter connection method involved conventional spike connection systems. In the 1980s, it was hypothesized that disconnect systems using a Y set would be superior to spike connection 11

22 systems for the prevention of peritonitis. The flush before fill technique using a Y connection system allowed for drainage of the spent dialysate into an empty drainage bag, followed by flushing of the tubing with some fresh dialysate before infusion of the remainder of the fresh dialysate into the patient. It was hypothesized that flushing the tubing before filling the peritoneal cavity would reduce the risk of infusing organisms introduced during the connection procedure into the peritoneal cavity. As predicted, the use of this Y set resulted in important reductions in the rate of peritonitis in several studies (24, 43 45). Subsequently, a more advanced form of disconnect system known as the double bag system was also shown to be superior to standard spike connection systems (45). While the double bag (or twin bag) system was hypothesized to further reduce peritonitis risk relative to standard Y sets by having one fewer connection, studies comparing these two disconnect systems have not consistently shown a benefit of one over the other (45 48). Based on the available data, the 2005 International Society for Peritoneal Dialysis (ISPD) guidelines suggest to avoid spiking of dialysis bags in CAPD patients, and to instead use a double bag system with the flush before fill technique to reduce the risk of contamination (49). The second major advance was the introduction of antibacterial ointments applied to the PD catheter exit site or nares to reduce bacterial colonization. The risk associated with bacterial colonization was first recognized by Luzar et al who studied S. aureus nasal carriage in patients on CAPD. In this study, it was found that S. aureus nasal carriers had exit site infection rates that were four times higher than non carriers (50). 12

23 The most likely explanation for this is that patients who have S. aureus colonization in their nares are more likely to have S. aureus colonization at their PD catheter exit site. The corollary of this is that eradication of colonization would reduce peritonitis occurring via periluminal migration of bacteria along the catheter tunnel into the peritoneal cavity. Supporting this hypothesis, it has since been shown that application of antibacterial ointments to the nares or catheter exit site not only reduces the risk of exit site infection (30, 51 53) but also of peritonitis (30, 53). The most studied ointment is mupirocin, which is a topical antibacterial agent with excellent activity against Gram positive organisms (54). The use of an agent active against Gram positive bacteria is appropriate since S. aureus is the most common organism causing exit site infection, accounting for 42% of episodes in one study of Canadian patients (26). While the majority of the studies on ointments for peritonitis prophylaxis have involved use of mupirocin, there are more recent data to suggest that gentamicin cream applied to the catheter exit site is an excellent alternative (30). One advantage of the latter agent over mupirocin is the provision of Gram negative coverage, particularly against Pseudomonas species. Based on these data, the 2005 ISPD guidelines for PD related infections suggested using one of the following regimens: (1) exit site mupirocin daily in all patients or only in S. aureus nasal carriers, (2) intranasal mupirocin for 5 7 days each month in nasal carriers, or (3) exit site gentamicin cream daily in all patients (49). Other strategies that have been adopted over time to reduce risk of peritonitis and exit site infection include use of downward directed catheter tunnels (55), and 13

24 administration of prophylactic antibiotics at the time of catheter insertion (56, 57). Prophylactic antibiotics have also been recommended prior to colonoscopies with polypectomy based on several case reports of peritonitis with enteric organisms occurring shortly after such procedures (49, 58 61). In addition to the above strategies, several studies have looked at different PD catheter designs. While modifications to the extraperitoneal segment of the catheter have not led to reductions in peritonitis (62 68), the data on use of single vs. double cuff catheters are conflicting (69 71). As a result, no definitive recommendations have been made as to the optimal PD catheter type for the prevention of peritonitis (49). Given that impaired peritoneal immunity is thought to be due in part to the bioincompatibility of the standard dialysate solutions used, several studies have sought to determine whether use of newer, more biocompatible PD solutions might be associated with a lower peritonitis risk. The improved biocompatibility of these solutions relates to their more neutral ph and their lower concentration of glucose degradation products. In a small randomized crossover study comparing biocompatible solutions with conventional solutions, the use of biocompatible dialysate for 6 months was associated with enhanced phagocytic activity of peritoneal macrophages, reduced constitutive inflammatory stimulation and better preservation of the mesothelial cell integrity (72). Three observational studies to date have reported a lower peritonitis rate with biocompatible solutions as compared with standard solutions (73 75), with another showing no effect (76). The only RCT of 14

25 conventional vs. biocompatible PD fluids that included data on infectious outcomes did not show a difference in peritonitis rates between the groups (77), although peritonitis was a secondary endpoint and the study was therefore not powered for this outcome. Given the cost associated with these biocompatible solutions, further randomized controlled studies are required to clarify whether the improved peritoneal immunity translates into reduced peritonitis risk. The importance of proper patient training in the prevention of PD related infections has also been studied. In one RCT, 620 PD patients were randomly assigned to receive either enhanced training using an adult learning theory based curriculum or a non standardized conventional training program. Those who received the enhanced training had significantly fewer exit site infections and peritonitis episodes (78). While the frequency of peritonitis has been dramatically reduced with the incorporation of these preventive strategies, peritonitis still occurs. And despite the presence of the aforementioned risk factors in all PD patients, it remains unclear why some patients never develop peritonitis while others go on to have multiple episodes. 15

26 2.3 Predictors of peritonitis Current knowledge regarding peritonitis risk Several observational studies have attempted to identify factors that might predict the occurrence of peritonitis in PD patients. Among demographic characteristics, two American studies have identified an association between Black race and peritonitis, with a 26 32% increased risk (55, 79). In addition, a large observational study from Australia and New Zealand reported a 76% increased risk of peritonitis among Aboriginal patients (80). While each of these observational studies adjusted for a wide range of patient and dialysis related factors, neither accounted for socioeconomic characteristics which may have varied widely by racial category. Another risk factor that has been associated with peritonitis is diabetes mellitus, with a 13 to 64% increased peritonitis risk among diabetics (79, 81, 82). This is not surprising as diabetic patients with renal disease have been shown to have an increased risk of infections in general (35, 83). Obesity has also been linked with a higher risk of peritonitis, with one study reporting a hazard ratio (HR) of 1.08 per 5 kg/m 2 increase in body mass index (80), and another reporting a HR of 1.29 in patients with a body mass index 30 (84). This higher risk may relate to the increased risk of dialysate leak among obese PD patients (85), in that leaks may predispose to peritonitis. In addition, an abdominal pannus may overlie the exit site and impair proper exit site care. 16

27 In addition to the above predictors, current or recent use of immunosuppressive agents for previous transplantation or glomerulonephritis (GN) has been associated with an increased risk of PD peritonitis (86). The increased risk among patients with failed transplants has been confirmed in some studies (87, 88) but not in others (55, 89, 90). While immunosuppression would be expected to be associated with an increased infection risk, the observational nature of these studies raises the possibility of residual confounding, as patients who have received a renal transplant are frequently the healthiest patients within a dialysis population. The improved general health of these transplant patients may offset any potential adverse infection risk associated with immunosuppression. Alternatively, one might hypothesize that the absence of an association between immunosuppression and peritonitis in some studies could reflect the relatively low doses of immunomodulatory agents used in patients with failed transplants who have returned to dialysis. Among biochemical parameters that are routinely measured in PD patients, a low serum albumin at the time of initiation of PD has been shown to be associated with a shorter time to first peritonitis, with HRs ranging from per 10 g/l decrease in baseline albumin (81, 91). The relationship between hypoalbuminemia and peritonitis may be either causative or associative. Specifically, one could hypothesize that malnutrition may directly increase risk of infection as a result of weaker immunity. In contrast, it is possible that a low albumin at the time of PD initiation is simply a 17

28 marker of a more inflamed patient with greater comorbidity who is more predisposed to infectious complications. Another biochemical variable of relevance is residual renal function (RRF) (82). In a recent study, each additional 1 ml/min/1.73 m 2 of residual glomerular filtration rate was independently associated with a 21% lower risk of peritonitis. The reduced risk associated with RRF may relate to the better solute clearance achieved (particularly of middle molecules), as it is known that uremia impairs host immunity (36 39). Alternatively, RRF may simply be a marker of a healthier patient with either less comorbidity or a shorter dialysis vintage, and therefore a patient less likely to develop peritonitis. In addition to known demographic and biochemical associations with peritonitis, having a first peritonitis episode is associated with an increased risk of developing a subsequent episode. In one study, a peritonitis episode occurring within the first 6 months after initiation of PD was associated with a shorter time to subsequent peritonitis, with a HR of 2.15 (79). A second study showed a RR of 2.08 for peritonitis in the patients who had a prior episode (55). This increased risk of subsequent peritonitis likely relates in part to patient factors that predispose to the development of peritonitis. However, an additional consideration may be the formation of a biofilm on the PD catheter over time (92 94). This biofilm is thought to consist of bacteria that attach to the PD catheter and become surrounded by an impenetrable glycocalyx matrix coat. It is hypothesized that the presence of a biofilm may predispose to the 18

29 development of peritonitis, and put patients at increased risk of subsequent infection with the same organism due to difficulty in eradicating the organism. This hypothesis was tested in a cohort of 198 patients with multiple peritonitis episodes (95). In this study, 80% of patients had at least one repeat infection with the same organism, and 79% of patients had more than half of their peritonitis episodes caused by the same organism, suggesting that bacterial biofilm formation on the peritoneal catheters may be an additional factor playing a role in the frequency of peritonitis in some patients. Two important variables for which data have been inconsistent in regard to peritonitis risk include age and choice of CAPD vs. APD. These will be discussed below Age The question of whether increasing age is associated with a higher risk of peritonitis is an important one given the increasing number of older patients reaching ESRD (3). Answering this question is particularly relevant as some nephrologists are reluctant to offer PD as a dialysis option to elderly patients. This was evidenced by a recent study in which North American nephrologists from several centers were asked to assess medical eligibility for HD, PD and renal transplantation among patients with advanced CKD (96). In this study, the most frequently cited reason for not offering PD to a patient was older age. 19

30 The relationship between age and peritonitis among PD patients has been investigated in several observational studies. The largest study to have looked at this studied 11,975 American patients on PD between 1994 and With age treated as a categorical variable, age 44 years was associated with an increased risk of peritonitis relative to those aged 65 to 74, with a HR of 1.09 at 9 months (79). In contrast, in a large, multicenter cohort of 3,162 patients from Australia and New Zealand followed from , age 65 was associated with an increased hazard of peritonitis relative to those under 25, 25 44, and years of age (80). Two other studies of Spanish PD patients reported a higher peritonitis rate in older patients (97, 98), while another study demonstrated an increased peritonitis risk among non diabetic patients greater than 70 years of age (99). Several other studies have reported no association between age and peritonitis (82, ). These studies are summarized in Table 1. There are several possible explanations for the inconsistent relationship between age and peritonitis across studies. The first is that age has been variably defined in these studies. For example, some studies used age as a categorical variable (with various arbitrary cutoffs used to define elderly ), while others used age as a continuous variable. Another contributing factor is the varying size of the studies to date, such that some of the smaller studies may have been underpowered to detect an association, should one have been present. Furthermore, the studies that have assessed the association between age and peritonitis span nearly two decades. Since PD technique has changed significantly over time, it is difficult to know whether 20

31 improvements in technique would have had an impact on the relationship between age and peritonitis. For example, elderly patients who have less manual dexterity may be more likely to have breaks in aseptic technique, such that introduction of the flush before fill technique and prophylactic ointments might offer greatest benefit in this population and offset the increased risk. Thus, the best assessment of the relationship between peritonitis and age would involve studying age as a continuous variable in a large, contemporary cohort of PD patients Peritoneal dialysis submodality: CAPD vs. APD With regard to choice of PD submodality, understanding the risk of peritonitis with CAPD vs. APD is important as it has implications for our recommendations to patients when they are choosing between these two options. CAPD was originally proposed to be associated with an increased risk of peritonitis in the early years of PD before advances in connectology. In contrast, the potential association between APD and a higher peritonitis risk has been attributed to the presence of a high comorbidity burden among a subset of APD patients, who may have been preselected for this modality on the basis of an inability to perform CAPD. If CAPD and APD are equivalent in terms of infectious risk, it would reinforce our practice of offering both options to patients initiating PD when feasible. In the absence of large, contemporary RCTs comparing the outcomes of the two submodalities, we are left to rely mostly on observational data to try to answer this question. 21

32 Several studies have addressed the relationship between use of CAPD vs. APD and peritonitis, including one RCT and several observational studies. The clinical trial conducted by de Fijter et al included 97 patients enrolled from 1988 to 1991, and randomized them to CAPD using a Y connector or APD. In this study, there were 54 peritonitis episodes among 25 CAPD patients, as compared with 31 episodes among 19 APD patients. This corresponded to a peritonitis rate that was significantly higher among CAPD patients relative to APD patients (0.94 vs episodes per patientyear) (104). Two other observational studies that have looked at the peritonitis risk associated with CAPD vs. APD have also suggested that CAPD is associated with a higher peritonitis risk, with HRs ranging from 1.72 to 2.08 (97, 100). However, in the largest observational study to have addressed this question, using data from 11,975 American PD patients, CAPD was associated with a 6% lower risk of peritonitis relative to APD (79). Furthermore, another study of 1,205 Scottish PD patients found no difference between CAPD and APD in terms of peritonitis risk (105). These studies are summarized in Table 2. While the only RCT to have addressed this question suggested that CAPD was associated with a higher peritonitis risk than APD, this study was conducted approximately 20 years ago at a time when peritonitis rates and PD practice were significantly different from what they are at present. As a result, the external validity of the study (or generalizeability) may be limited. In addition, some of the apparent inconsistency among these studies may relate in part to the fact that some of the studies in which CAPD was associated with a higher peritonitis rate included patients 22

33 who were on PD before the adoption of the improved connectology systems, which greatly reduced the risk of contamination at the time of an exchange (24, 43 45). The implementation of such systems would have preferentially benefited CAPD patients. Thus, studying a contemporary cohort of patients who initiated dialysis after adoption of these more advanced connectology systems is vital to understanding the current peritonitis risk with CAPD vs. APD. 2.4 Statistical methodology used to study occurrence of peritonitis Some of the variability in the predictors of peritonitis that have been identified may relate to the patient populations studied, the varying sizes of the cohorts studied, the different variables included in the multivariable models in each study and the different eras over which data were collected. However, an additional complicating factor is the type of analysis chosen to assess for variables associated with peritonitis. The choice of the type of multivariable regression model depends on the way in which the outcome variable, peritonitis, is defined. The simplest way to define peritonitis would be as a dichotomous variable, such that a patient either had or did not have a peritonitis episode. It is apparent that this definition of peritonitis is limited by the fact that it does not take into account the amount of time the patient is on PD (which represents the time at risk). As a result, if all patients are followed for a very short time, one would not be able to distinguish between those patients who would 23

34 never have developed peritonitis and those who were destined to get peritonitis after a certain amount of exposure time on PD. It is clear, then, that the definition of peritonitis as an outcome variable should incorporate the amount of time a patient is on PD. The first method of defining peritonitis that incorporates time involves calculating a peritonitis rate for each patient that is, the number of episodes of peritonitis experienced by a patient divided by the follow up time. Models in which one can determine the relationship between a variable and the peritonitis rate include the Poisson model and the negative binomial model. The Poisson model is best used for a stochastic process when events occur independently of one another. One assumption inherent in Poisson modeling is that the mean is equal to the variance. The negative binomial model is a variant of the Poisson model that does not make the assumption that the mean is equal to the variance, and as a result, it provides a better fit if there is overdispersion of the data (106, 107). The second option for defining peritonitis with incorporation of time at risk is to study the time to peritonitis. The most frequently used model is the Cox proportional hazards model (as long as the hazard associated with each variable is proportional over time). This allows for determination of the predictors of a shorter time to peritonitis. One limitation of this type of analysis is that it allows one to look only at predictors of time to first peritonitis, such that one would not be able to utilize all data on subsequent peritonitis episodes in the same patient. Performing a time to event 24

35 analysis incorporating multiple events occurring over time is possible, but to date, this analytic tool has not been used for the study of peritonitis. There are three time to event modeling approaches that have been used to study multiple events within an individual: the Andersen Gill model, the marginal (WLW) model and the conditional (PWP) probability model (108). The Andersen Gill model is a variant of the Cox model that allows for incorporation of data on recurrent events. One condition of this model is the assumption of independence of events within a subject. Based on simulated data, this model gives nearly unbiased estimates of the treatment effect even when an important covariate has been omitted. The marginal model is a model that was first used by Wei, Lin and Weissfeld to analyze bladder cancer data in the context of multiple recurrences per subject. Limitations of this model include potential violation of the proportional hazards assumption, and biased estimates of effect when covariates are omitted. The conditional model, proposed by Prentice, Williams and Peterson, allows for variation in the underlying intensity function from event to event. However, the conditional model is even more susceptible to biased estimates when an important covariate is omitted. When choosing between these three models for the study of PD peritonitis, the absence of important covariates in many large dialysis datasets would favor the use of the Andersen Gill model. The utility of both rate and time to event analyses in the study of peritonitis was identified in the early years of PD therapy. The first insight into the optimal statistical 25

36 modeling of peritonitis came in 1981 when Corey studied a cohort of 129 Toronto PD patients and determined that the distribution of peritonitis was random based on the goodness of fit provided by the Poisson regression model (109). A later study proposed a variant of this strategy in the form of a mixed effects Poisson model, which could incorporate not only the fixed effects corresponding to information collected across individuals, but also a random effect due to individuals (110). Another study looked at using a life table analysis to model peritonitis, and showed that the peritonitis probability curve constructed with only the first episode of peritonitis was almost identical to that constructed from all episodes of peritonitis (111). The authors concluded that this finding further supported the random distribution of peritonitis among patients, and suggested that analyzing the time tofirst peritonitis was an accurate means of expressing the probability of developing peritonitis. The majority of studies to date that have looked at predictors of the occurrence of peritonitis have either performed a peritonitis rate analysis or a time to first peritonitis analysis. While readers of the literature tend to interpret these studies interchangeably, there are little comparative data to date to suggest whether peritonitis rate analyses and time to event analyses are equivalent as analytic tools in the study of peritonitis. Two studies have compared modeling strategies. In the first study of a small number of pediatric patients, a tight correlation was demonstrated between a peritonitis rate analysis using a negative binomial model and a time to first peritonitis analysis (112). A second study also showed similar predictors of peritonitis 26

37 and parameter estimates using a negative binomial model and a time to first peritonitis model (80). It is not known whether modeling peritonitis using a rate analysis and an Andersen Gill model for multiple events would yield similar predictors, and whether the estimates of risk would be congruent. Demonstration of such congruency would be relevant to the interpretation of published studies and to the design of future studies. 27

38 3. METHODOLOGY 3.1 Data sources The study included Canadian PD patients for whom data were available through the Peritonitis Organism Exit sites Tunnel infections (POET) database (Baxter Healthcare). The POET Clinical Monitoring System is a software program designed to organize and analyze PD patient data to identify and monitor the causes of infection, catheter complications, and therapy transfers. The POET software was offered to all PD centers in Canada that were using Baxter PD products (representing an estimated 85% of Canadian PD centers). No financial or other incentive was provided to the individual centers for use of the software. Installation of the software was documented for 56 Canadian centers. To create the database, Canadian centers that used POET in a consistent manner to track infectious and non infectious complications were asked to contribute their data. Specifically, centers that reported data for at least 1 year and had cumulative data for at least 20 adult patients in their program were invited to contribute their de identified data. Twenty five centers met these criteria, and were included in the database. These Canadian centers ranged in size from patients per center, and reported on patients between 1990 and Data collection was performed by the PD nurses in the majority of centers. 28

39 Prior to study initiation, research ethics board approval was obtained from the University Health Network. 3.2 Patient population The database included both prospectively collected data on incident PD patients as well as data on prevalent patients that were retrospectively entered into the database when a given center started using the POET software. In order to distinguish prospectively collected data from retrospectively collected data, the 25 Canadian centers contributing to the database were contacted to determine the exact time when their center started using the POET software for data collection. In order to avoid a survivorship bias related to inclusion of prevalent patients and a recall bias related to retrospective data collection, we included only incident patients in whom data were collected prospectively. The time period for prospective data collection was from January 1, 1996 until September 12, Model covariates Demographic data used in the current study included age, gender, race, diabetic status, GN as a cause of ESRD, modality before PD start (new to dialysis, transfer from HD, failed transplant) and PD submodality (CAPD vs. APD). Data available for the latter variable included the submodality at the time of initiation of PD, and the 29

40 submodality as of the most recent data entry. Detailed information on all submodality switches during a patient s time on PD was not available. As a result, in order to reduce confounding by modality switching, a secondary analysis was performed after exclusion of any patient who switched from one PD submodality to another (CAPD to APD or vice versa). While the database included extensive comorbidity data, the majority of these comorbidities (with the exception of diabetes mellitus) appeared to be under reported when compared with CORR data and were therefore excluded from the analysis (5). This is not surprising as comorbidities are known to be under reported in administrative data (113). Given that the prospective cohort included patients who initiated PD over a 10 year period, we defined two eras of patients in order to assess for an era effect: an earlier cohort consisting of those who initiated PD between January 1, 1996 and December 31, 2000, and a more contemporary cohort consisting of those who initiated PD between January 1, 2001 and September 12, Outcomes The primary outcome was the occurrence of peritonitis, which was defined in two ways: (1) as a peritonitis rate, and (2) as a time to peritonitis incorporating all peritonitis episodes. 30

41 In order to focus on independent peritonitis events, relapsing or recurrent episodes were excluded. Standard ISPD definitions were used, with a relapse defined as an episode occurring within 4 weeks of completion of therapy of a prior infection with negative culture or the same organism, and a recurrence defined as an episode occurring within 4 weeks of completion of therapy of a prior infection but with a different organism (49). While we did not have data on the duration of antibiotic therapy, it was standard practice over the time period of the study to treat peritonitis with a minimum of 2 weeks of antibiotics, with some patients being treated with up to 4 weeks of antibiotics. The primary analyses were performed based on the conservative assumption of 4 weeks of antibiotic therapy, such that all peritonitis episodes occurring within 60 days of a previous episode were excluded. In order to ensure that the assumption about the antibiotic duration did not affect the results, we performed a sensitivity analysis, with repetition of the analyses after exclusion of peritonitis episodes occurring within 45 days of a prior episode (assuming 2 weeks of antibiotic therapy). 3.5 Statistical Analyses Continuous variables were reported as mean ± SD. Two models were used to assess the predictors of peritonitis. In the first, potential predictors were tested using a multivariable negative binomial model that modeled the number of peritonitis episodes per patient, using the duration of follow up as an offset variable. In the 31

42 second model, peritonitis outcome was reported as the time to each peritonitis event, and analyzed using an Andersen Gill model for ordered multiple events. This model allowed information on all events to be included with the assumption that each event was independent. A priori selected variables for inclusion as covariates included age, gender, race, diabetic status, GN as a cause of ESRD, modality before PD start (new to dialysis, transfer from HD, failed transplant) and PD modality (CAPD vs. APD). To assess whether the era of PD initiation influenced the relationship between each variable and peritonitis, we tested an interaction term between era and each variable as a method of initial screening. If the interaction was found to be statistically significant, then subsequent analyses were performed to determine the relationship between that variable and peritonitis in each of the two eras. Statistical significance was defined as a p value of <0.05. All statistical analyses were performed using SAS (version 9.1). 32

43 4. RESULTS 4.1 Patient cohort The entire cohort consisted of 6,544 patients, including 4,247 incident patients in whom data were collected prospectively and 2,297 prevalent patients in whom data were retrospectively entered. After exclusion of prevalent patients, the study sample consisted of 4,247 incident PD patients, of whom 1,605 had 3,058 episodes of peritonitis. The remaining 2,642 patients had no peritonitis episodes. Of the 3,058 peritonitis episodes, 503 were excluded as they occurred within 60 days of a prior episode and were assumed to be recurrent or relapsing events. Consequently, the analyses included data for 2,555 peritonitis episodes seen amongst 4,247 patients with a total of 7,319 years of follow up. A flow diagram of the patient cohort included in the analyses is illustrated in Figure 2. The distribution of peritonitis episodes within the cohort is shown in Table 3. Eight provinces contributed data to the POET database, with the largest number of patients coming from Ontario. The distribution of patients by province is shown in Figure 3. When all 3,058 peritonitis episodes were counted, the overall peritonitis rate was 1 episode in 26 patient months on PD. This decreased to 1 episode in 33 patientmonths after exclusion of recurrent or relapsing events. The median time on PD was 1.37 years with an interquartile range of 0.62 to 2.43 years. Of the 4,247 patients included in the study, 1,445 (34.0%) were still being followed at the end of the data 33

44 collection period (median follow up time 2 years), 18.4% of patients in the cohort died after a median time on PD of 1.31 years, 27.2% transferred to HD after a median time on PD of 0.93 years, and 12.2% received a renal transplant after a median time on PD of 1.21 years. Demographic characteristics of the patients are presented in Table 4. A comparison between the Canadian POET cohort and prevalent Canadian PD patients from the 2006 CORR data is shown in Table 5 (1). 4.2 Independent predictors of PD peritonitis For the analysis in which peritonitis was modeled as a count, the negative binomial model was used. In the multivariable negative binomial model, variables independently associated with a higher peritonitis rate included age (rate ratio (RR) 1.04 per decade increase, 95% confidence interval (CI) , p=0.010), Black race (RR 1.37, 95% CI , p=0.05) and transfer from HD to PD (RR 1.24, 95% CI , p<0.001). Predictors of a lower peritonitis rate included having GN as the cause of ESRD (RR 0.87, 95% CI , p=0.05) (Table 6). An interaction between gender and diabetes was identified and is reported in detail below (section 4.3). The relationship between use of CAPD vs. APD and peritonitis was assessed in the subset of 3,180 patients who did not switch submodalities during their time on PD 34

45 (Table 7). In this subgroup, CAPD was not associated with a higher peritonitis rate than APD (RR 1.03, 95% CI , p=0.66). Using the multivariable Andersen Gill Cox model, variables associated with a shorter time to peritonitis included age (HR 1.03 per decade increase, 95% CI , p=0.025), Black race (HR 1.47, 95% CI , p=0.002) and transfer from HD to PD (HR 1.24, 95% CI , p<0.001). Having GN as the cause of ESRD was associated with a longer time to peritonitis (HR 0.86, 95% CI , p=0.015) (Table 8). Among the 3,180 patients with no submodality switch, use of CAPD was not associated with a shorter time to peritonitis than use of APD (HR 1.02, 95% CI , p=0.69) (Table 9). 4.3 Interactions In both analyses, a significant interaction between gender and diabetes was seen (p=0.011 in the negative binomial analysis and p = in the Andersen Gill model). When the association between diabetes and peritonitis by gender was assessed, it was found that female diabetics were at increased risk of peritonitis (RR 1.27, 95% CI , p=0.001), while male diabetics were not (RR 0.99, 95% CI , p=0.88) (Table 10). 35

46 4.4 Era effect Initial screening for an era effect for each of the variables revealed that the only significant interaction was for the relationship between age and era (p=0.001) (Table 11). Because of the presence of an interaction, the relationship between age and peritonitis in each era was assessed (Table 12). In this analysis, it was found that the higher peritonitis risk associated with increasing age in the overall analysis was entirely accounted for by those initiating dialysis prior to the year 2001 (RR 1.11 between 1996 and 2000, 95% CI , p<0.001), with no relationship between age and peritonitis thereafter (RR 1.00 between 2001 and 2005, 95% CI , p=0.83) (114). 4.5 Comparison of peritonitis modeling strategies The relationship between peritonitis and each of the variables in the multivariable model was tested in both a negative binomial rate model and an Andersen Gill timeto event model. These results are compared in Table 13. The two models yielded similar predictors of peritonitis, and with comparable estimates of risk. 36

47 4.6 Sensitivity analysis for peritonitis relapse/recurrence exclusion criteria For the main analyses, any peritonitis episode occurring within 60 days of a prior episode was excluded as a recurrent or relapsing episode based on the assumption of a maximum of 4 weeks of antibiotic therapy. Since some patients were likely to have received shorter courses of antibiotics, we repeated the analyses after exclusion of any peritonitis episode occurring within 45 days of a prior episode in order to exclude any bias introduced by this assumption. In these analyses, the results did not change appreciably, with identification of similar predictors of peritonitis and similar risk estimates in both the negative binomial model and the Andersen Gill model (Tables 14 17). 37

48 5. DISCUSSION 5.1 General discussion Using a large multicentre database of Canadian patients initiating PD between 1996 and 2005, the work contained within this thesis highlights several novel and confirmatory findings. Predictors of PD peritonitis identified in this study included Black race, transfer from HD to PD, and being a female diabetic. Increasing age was only associated with an increased risk of peritonitis among those initiating PD before the year In contrast to several prior studies, we found that choice of CAPD vs. APD did not influence the peritonitis risk. Furthermore, these results were similar regardless of modeling strategy, suggesting that both rate analyses and time to event analyses are comparable analytic tools for studying the occurrence of PD peritonitis. Prior to this study using the POET database, the two largest observational studies to have looked at variables associated with peritonitis utilized the USRDS database (79) and the ANZDATA registry (80). The former analysis included 11,975 American patients on PD between 1994 and Covariates included age, gender, race, cause of ESRD, comorbidities, PD submodality, number of entry period hospitalization days, entry period hematocrit and peritonitis during the entry period. Unfortunately, as a result of the method of data collection, patients who did not survive their first 9 months on PD were excluded, as were those with secondary pay Medicare insurance patients or those insured by health maintenance organizations. 38

49 Furthermore, peritonitis episodes occurring in the first 3 months on PD were not captured, nor could the database capture whether a patient had one or more peritonitis episodes during the 6 month entry period. The ANZDATA analysis, which included data on 3,162 patients from Australia and New Zealand commencing PD between 1999 and 2003, was more comprehensive in that it was able to capture data on all new PD starts from the time of PD initiation. Covariates included age, gender, race, comorbidities, body mass index, timing of nephrology referral and peritoneal transport status. Similar to the ANZDATA registry, advantages of the POET database include the multi center nature of the database, the inclusion of a relatively contemporary PD cohort and the availability of data from the first day of initiation of PD. Among the variables that have been linked to peritonitis, the data on age have been conflicting (79, 80, 82, 97, ). As discussed, several factors may be responsible for this variability. Firstly, different results may reflect the varying age cutoffs used to define elderly in the studies. Secondly, many of the studies that have looked at the effect of age on peritonitis have been small, single center studies with limited statistical power. Thirdly, the era in which the patients received dialysis is quite variable, with some studies reporting on patients who were on PD in the late 1980s, and others reporting on more contemporary PD cohorts. The importance of the latter issue relates to the major advances in PD connectology (24, 45 48, 115) and exit site care (30, 51 53) that occurred over this time period. While increasing age was associated with a higher peritonitis rate in our overall analysis, we have identified an 39

50 era effect for age, such that increasing age is only associated with peritonitis among those who initiated PD before the year The lack of association between increasing age and peritonitis in recent years may reflect the fact that the flush before fill technique and the use of topical antibacterial agents provide an added safety net to contamination in elderly patients who may have impaired vision or dexterity. Importantly, there was no era effect for any of the other predictor variables, suggesting that their association with peritonitis is not related to the year in which the patient initiated PD. The finding that Black race is associated with a greater risk of peritonitis is consistent with previous American studies (55, 79). The basis for the increased risk is unclear, but could relate to genetic differences or to socioeconomic factors that are not captured in most large databases. While the higher peritonitis rate among African American patients has contributed to increased technique failure rates (116), the survival of Black patients on PD remains at least as good as that of Caucasian patients (116, 117). The increased peritonitis rate associated with transfer from HD to PD has not been previously reported. We hypothesize that this increased risk may be attributable to two high risk groups. The first group includes those who were crash starts on HD with little pre dialysis care and subsequently chose to transfer to PD. These patients would likely be sicker, with poorer nutritional status, a greater degree of inflammation, more rapid loss of RRF and an increased susceptibility to infection and adverse 40

51 outcomes ( ). The second group includes those who had been on HD for years and exhausted all vascular access options. For the latter group, the lack of RRF at the time of transfer to PD may contribute to their peritonitis risk given that loss of RRF is an independent predictor of peritonitis (82). Since we do not have information on dialysis vintage prior to transfer, we cannot determine with certainty which group of patients accounted for the increased peritonitis risk. Nevertheless, physicians caring for PD patients should be aware of the higher peritonitis rate among those transferring from HD. It is not surprising that diabetes has been previously reported to be associated with a higher peritonitis rate (79, 81, 82) as it is known that diabetic patients with renal disease are at higher risk of infection in general (35, 83). However, in this study, we found for the first time a significant interaction between gender and diabetes, such that the higher peritonitis rate was present only among female diabetics. This is of particular interest as several large US studies have demonstrated a higher incidence of death on PD among women, in particular among female diabetic patients (7, 11, 15). In one study using USRDS data, Bloembergen et al noted a differential effect of gender on PD outcomes, with women at significantly higher risk of death due to infection than men (11, 122). In a subsequent comparison of PD and HD outcomes by Vonesh et al, female diabetic patients were one of the few subgroups in which PD was associated with a higher risk of death than HD (15). Furthermore, Collins et al reported a higher risk of all cause death for female diabetics 55 years of age on PD as compared with HD (7). In cause specific analyses in the latter study, it was found 41

52 that these patients had a significantly higher risk of infectious death on PD. A smaller single center study subsequently reported that infection was the second leading cause of death among older diabetic women on PD (123). Our finding that female diabetics have the highest peritonitis rates therefore suggests that the higher risk of infection related death in this group may be mediated in part through a higher risk of PD peritonitis. While the basis for the increased peritonitis risk among female diabetics requires further study, loss of RRF may play a role, as it has been shown that diabetes is associated with a greater decline in RRF ( ). Furthermore, there is one study demonstrating a more rapid loss of residual kidney function among female dialysis patients (124), although the data on gender and RRF are conflicting ( ). With regard to the lower peritonitis rate among patients with GN as the cause of ESRD, the results are not surprising. While use of immunosuppressive agents in this subgroup of patients may increase infection risk (86), these patients tend to be younger and have fewer comorbidities. Since we were not able to adjust for comorbidities other than diabetes, the reduced peritonitis risk among patients with glomerulonephritis is likely due to residual confounding. While the majority of patients initiate PD either as their initial modality or after transfer from HD, some patients start PD after failure of their renal transplant. There are data to suggest that those who return to dialysis after graft loss are at increased risk of adverse outcomes, including death (87, 128) and peritonitis (86 88). Other studies 42

53 have refuted these findings (89, 90). The theoretical basis for an increased peritonitis risk includes the use of immunosuppressive therapy, and the long renal disease vintage in the majority of these patients. While having a failed transplant was not associated with an increased peritonitis risk in our analysis, this group accounted only for 3% of PD starts so that this dataset was likely underpowered to detect a difference, should it have been present. Several previous studies have addressed the issue of whether the use of CAPD vs. APD has an effect on peritonitis risk. The majority of studies have found that CAPD is associated with a higher risk of peritonitis (97, 100, 104), including the only RCT to have studied the relationship between modality and peritonitis risk (104). However, there are also observational data to suggest a lower risk of peritonitis on CAPD relative to APD (79), or no difference between the two submodalities in terms of peritonitis risk (105). The apparent inconsistency among these studies may relate in part to the fact that some of the studies in which CAPD was associated with a higher peritonitis rate included patients who were on PD before the adoption of the improved PD connectology systems, which greatly reduced the risk of contamination at the time of an exchange. In our study, which included a larger and more contemporary cohort of patients than in most of the previous studies, there was no association between peritonitis and the use of CAPD vs. APD. These data support the generally accepted practice of having the choice between CAPD and APD guided by patient preference if the patient is capable of performing both modalities. 43

54 With regard to the optimal modeling approach to study the occurrence of peritonitis, there are little comparative data. Most studies have reported either peritonitis rates or time to first peritonitis. Two studies have compared modeling using a negative binomial rate model and a time to first peritonitis analysis, with both demonstrating similar predictors (80, 112). However, one of the limitations of the time to peritonitis analyses reported to date is that all studies using this type of modeling have only incorporated time from initiation of dialysis until the first peritonitis episode. In our analyses, we have used an Andersen Gill model which allows for modeling of time to peritonitis with the incorporation of all events occurring in each patient. Using this type of modeling, information on all peritonitis episodes can be included. Based on the similar results between the rate and time to event analyses in our study, we conclude that both are appropriate analytic methods in the assessment of factors related to peritonitis. 5.2 Impact on nephrology practice While peritonitis rates among PD patients have declined over time, the occurrence of peritonitis remains a major concern for both patients considering PD as well as the nephrologists looking after them. The inconsistent data on the relationship between increasing age and peritonitis may be a contributing factor in regard to the concern over offering PD to older patients (96). The data presented in this thesis suggest that among a contemporary cohort of PD patients, peritonitis risk does not increase with 44

55 increasing age and should therefore not be a limiting factor in the selection of PD as a dialysis modality. Furthermore, we have shown in this observational study that peritonitis risk is similar between CAPD and APD. In the absence of strong evidence suggesting a difference in occurrence of peritonitis between CAPD and APD, peritonitis risk should not contribute to the decision making when selecting between these submodalities. Finally, we have for the first time identified diabetes among women and transfer from HD to PD as predictors of peritonitis. While these do not represent modifiable risk factors, an awareness of the increased risk in these patients should heighten the vigilance among the members of treating team. 5.3 Limitations Our study has several limitations. As with most large datasets, the data have not been validated against patient charts. Since this was a clinical database, the data entry was performed by the PD nurses and not trained data collectors. As such, the accuracy and reproducibility of the data entry cannot be verified. The variables most likely to be entered accurately include basic demographics such as age, race and gender, as well as easy to identify comorbidities such as diabetes, and the welldefined outcome of peritonitis. Some comorbidity data such as the presence of cardiovascular disease or lung disease are more difficult to define, and hence may be 45

56 less accurately reported or under reported. As a result, for the purpose of our analyses, we elected to study only those variables that were most likely to have complete and accurate data entry. In order to reduce some of the bias inherent in observational data, we studied only a sub population of the entire cohort in the database. Specifically, we chose to include only those patients in whom data were prospectively entered in order to avoid bias related to retrospective data entry. We also chose to exclude prevalent patients as inclusion of these patients in the analysis might have introduced a survivor bias. With regard to the data source, we cannot exclude the possibility of a selection bias in that the 25 centers included in the database had to have met the following criteria: (i) were among the 85% of Canadian centers using Baxter PD products, (ii) agreed to use the POET software for data collection, and (iii) had consistent data entry for a minimum of 20 patients. Despite this limitation, the database included centers of varying sizes in both academic and community hospitals with good representation from almost all Canadian provinces. While the multivariable regression models incorporated several potentially important variables, as with many database studies, the covariates were limited to those for which data were available. We acknowledge that there are important variables for which we did not have information. Firstly, we did not have data on biochemical parameters such as serum albumin and RRF both of which have been linked with 46

57 peritonitis risk. In addition, we did not have data on dialysis vintage. This would have been particularly relevant among patients who transferred from HD in order to distinguish whether the increased peritonitis risk was largely attributed to crash starts on dialysis or long term HD patients. We also did not have information on S. aureus nasal carriage, or on use of prophylactic ointments that are known to reduce peritonitis risk. Other important variables that could not be adjusted for included those that pertain to socioeconomic status, as these might have influenced peritonitis risk. Finally, while the POET database provided information on race, there was no separate category for Aboriginal race. This would have been of interest, as this is a highly prevalent population in several Canadian provinces, with significant differences in demographics and comorbidities, and increased frequency of technique failure (129). While large multicenter datasets have the advantage of greater power to detect clinically meaningful associations, their limited ability to adjust for all potentially important variables leads to the inevitable possibility of residual confounding. Since we did not have detailed information on all switches between CAPD and APD during a patient s time on PD, we tested the association between the PD modality and peritonitis by performing the analysis in a subgroup of patients who did not switch between CAPD and APD during their time on PD. Despite this, the number of patients in this subgroup was still larger that the majority of studies that have tested this association. While these data are reassuring, it is important to note that there are many factors that influence the choice of CAPD vs. APD. While we adjusted for basic patient demographics and diabetic status, we did not adjust for other comorbidities 47

58 that may have differed between the patient groups, nor could we adjust for nonmedical factors contributing to modality selection. As a result, we cannot exclude the possibility of residual confounding due to variables that were not included in our model. 5.4 Conclusions In conclusion, our study has, for the first time, identified transfer from HD to PD as an independent risk factor for PD peritonitis. In addition, there was an interaction between gender and diabetes such that a higher peritonitis risk was only seen among female diabetics. In contrast to previous studies, the choice of CAPD vs. APD did not affect the risk of peritonitis. We have also found that age is not a risk factor for peritonitis in a contemporary cohort of PD patients. Finally, we have demonstrated that rate analyses and time to event analyses are both appropriate analytic tools to study the occurrence of PD peritonitis. 5.5 Future directions In light of the newly identified predictors of peritonitis, future studies can be directed at trying to understand the basis for the increased risk and potential strategies to reduce risk in these patients. For example, the finding that patients transferring from HD to PD are at increased risk of peritonitis should be explored further in order to 48

59 determine whether those at increased risk are crash starts on HD or those with a long dialysis vintage, and whether more extensive training among these patients might be warranted. With respect to the consistent association between female diabetics and adverse outcomes, including increased peritonitis risk, future studies are needed to explore the basis for this increased risk, including whether hormonal mediators, residual renal function or other factors are responsible for this important finding. Given the increasing incidence of ESRD among Canadian Aboriginal patients and their higher risk of PD technique failure, further study is warranted to examine the risk of PD related infectious complications in this population. Finally, the POET database can also be used to further explore other aspects of PD peritonitis risk, including the relationship between PD catheter type and peritonitis, and the microbiology and outcomes of infectious complications. 49

60 6. ILLUSTRATIONS 6.1 Tables 6.2 Figures 50

61 Table 1. Summary of studies testing the association between age and peritonitis AUTHOR (reference) DATA SOURCE # OF PTS ERA Definition of age Oo (79) USA 11, (USRDS) Lim (80) Australia 3, and New Zealand (ANZDATA) 65 Rodriguez Spain Continuous Carmona (97) variable Han (82) Korea Continuous variable Huang (100) Taiwan Continuous variable Statistical RESULTS analysis Cox PHM 0 44: HR 1.09 ( ) 45 64: HR 1.01 ( ) 65 74: reference group 75: 1.07 ( ) NBM 0 24: RR 0.9 ( ) Cox PHM 25 44: RR 0.83 ( ) 45 64: RR 0.88 ( ) 65: reference group Rate model Increased risk of episodes (not specified) per patient year per year older Cox PHM No effect of age (HR 0.99, 95% CI ) Cox PHM No effect of age (HR 1.00, p=0.94) De Vecchi (99) Italy vs 40 60t test 0.52 vs episodes/patientyear (p<0.002) Kadambi (101) USA < ANOVA No difference (p=ns) Li (103) Hong Kong vs <65 Life table analysis No difference (p=0.75) Holley (102) USA vs 18 49Poisson No difference (p=ns) model Perez Contreras (98) Spain vs <65 t test 0.72 vs episodes/patientyear (p=0.01) PHM = proportional hazards model, NBM = negative binomial model, ANOVA = analysis of variance 51

62 Table 2. Summary of studies testing the association between CAPD vs. APD and peritonitis AUTHOR Country # OF PTS ERA STATISTICAL ANALYSIS RESULTS De Fijter (104) Holland t test Life table analysis APD better: RR 0.54 (p=0.03) 11 vs. 18 months to first peritonitis (p=0.06) Oo (79) USA (USRDS)11, Cox PHM CAPD better: HR 0.94 (p=0.008) Kavanagh (105) Scotland (Scottish Renal Registry) Rodriguez Spain Rate model (not Carmona (97) specified) 1, Poisson model No difference (p=0.21) APD better: CAPD associated with excess of 0.2 episodes/patient year (p=0.008) Huang (100) Taiwan Cox PHM APD better: HR 0.58, p=

63 Table 3. Distribution of peritonitis episodes within the patient cohort Number of peritonitis episodes Number of patients (total n=4,247) Percentage of patients % % % % % % % % % % 53

64 Table 4. Baseline demographic characteristics for the entire patient cohort n=4,247 Age (mean, years) 59 ± 16 Gender (% male) 55% Race (%) Caucasian Black Asian Other Modality (% on CAPD) Initial Most recent Modality before PD start (%): New to dialysis Transfer from HD Failed transplant Other/unknown 82% 2% 6% 10% 74% 52% 58% 24% 3% 15% Cause of ESRD: Diabetes Mellitus 35% Hypertension 17% Glomerulonephritis 15% Cystic kidney disease 5% Other 27% Diabetic 40% CAPD = continuous ambulatory peritoneal dialysis, PD = peritoneal dialysis, HD = hemodialysis, ESRD = end stage renal disease 54

65 Table 5. Comparison of patient demographics between POET and CORR POET database CORR data Study patients Age (mean, years) % 19% 38% 24% 17% 1% 15% 39% 25% 20% Gender (% male) 55% 56% Cause of ESRD: Diabetes Mellitus Hypertension Glomerulonephritis Cystic kidney disease Other 35% 17% 15% 5% 27% Prevalent patients % 18% 19% 6% 26% CORR = Canadian organ replacement register, ESRD = end stage renal disease 55

66 Table 6. peritonitis Multivariable negative binomial model for the outcome of n = 4,247 patients NEGATIVE BINOMIAL MODEL Rate ratio 95% CI p value Age (per decade increase) Black Asian Diabetes female male Glomerulonephritis Transfer from HD <0.001 Failed transplant HD = hemodialysis, CI = confidence interval 56

67 Table 7. Multivariable negative binomial model for the outcome of peritonitis in the subgroup of patients with no submodality switch n = 3,180 patients NEGATIVE BINOMIAL MODEL Rate ratio 95% CI p value Age (per decade increase) Black Asian Diabetes female male Glomerulonephritis Transfer from HD <0.001 Failed transplant CAPD vs. APD CAPD = continuous ambulatory peritoneal dialysis, APD = automated peritoneal dialysis, HD = hemodialysis, CI = confidence interval 57

68 Table 8. Multivariable Andersen Gill model for the outcome of peritonitis n = 4,247 patients ANDERSEN GILL MODEL Hazard ratio 95% CI p value Age (per decade) Black Asian Diabetes female male < Glomerulonephritis Transfer from HD <0.001 Failed transplant HD = hemodialysis, CI = confidence interval 58

69 Table 9. Multivariable Andersen Gill model for the outcome of peritonitis in the subgroup of patients with no submodality switch n = 3,180 patients ANDERSEN GILL MODEL Hazard ratio 95% CI p value Age (per decade) Black Asian Diabetes female male Glomerulonephritis < Transfer from HD <0.001 Failed transplant CAPD vs. APD CAPD = continuous ambulatory peritoneal dialysis, APD = automated peritoneal dialysis, HD = hemodialysis, CI = confidence interval 59

70 Table 10. Association between diabetes and peritonitis by gender n = 4,247 patients NEGATIVE BINOMIAL MODEL ANDERSEN GILL MODEL Diabetes female male Rate ratio 95% CI p value Hazard ratio % CI p value < CI = confidence interval 60

71 Table 11. Interaction between each variable and era in the multivariable negative binomial model INTERACTION TERM WITH ERA p value Age Black 0.28 Asian 0.85 Diabetes x Gender 0.58 Glomerulonephritis 0.87 Transfer from HD 0.41 Failed transplant 0.33 *CAPD vs. APD 0.24 *subgroup of 3,180 patients who did not switch between CAPD and APD during their time on PD HD = hemodialysis, CAPD = continuous ambulatory peritoneal dialysis, APD = automated peritoneal dialysis 61

72 Table 12. Association between age and peritonitis by era. N = 4,247 patients NEGATIVE BINOMIAL MODEL ANDERSEN GILL MODEL Rate ratio 95% CI p value Hazard ratio 95% CI p value OVERALL (n=4,247 patients) (n=1,494 patients) < < (n=2,753 patients) CI = confidence interval 62

73 Table 13. Comparison of results of multivariable negative binomial model and Andersen Gill model for peritonitis N = 4,247 patients NEGATIVE BINOMIAL MODEL ANDERSEN GILL MODEL Rate ratio 95% CI p value Hazard ratio 95% CI p value Age (per decade) Black Asian Diabetes female male < Glomerulonephritis Transfer from HD < <0.001 Failed transplant *CAPD vs. APD *subgroup of 3,180 patients who did not switch between CAPD and APD during their time on PD HD = hemodialysis, CAPD = continuous ambulatory peritoneal dialysis, APD = automated peritoneal dialysis, CI = confidence interval 63

74 Table 14. Multivariable negative binomial model for the outcome of peritonitis (sensitivity analysis using 45 day relapse/recurrence exclusion criteria) n = 4,247 patients NEGATIVE BINOMIAL MODEL Rate ratio 95% CI p value Age (per decade increase) Black Asian Diabetes female male Glomerulonephritis < Transfer from HD <0.001 Failed transplant HD = hemodialysis, CI = confidence interval 64

75 Table 15. Multivariable negative binomial model for the outcome of peritonitis in the subgroup of patients with no submodality switch (sensitivity analysis using 45 day relapse/recurrence exclusion criteria) n = 3,180 patients NEGATIVE BINOMIAL MODEL Rate ratio 95% CI p value Age (per decade increase) Black Asian Diabetes female male Glomerulonephritis Transfer from HD <0.001 Failed transplant CAPD vs. APD HD = hemodialysis, CAPD = continuous ambulatory peritoneal dialysis, APD = automated peritoneal dialysis, CI = confidence interval 65

76 Table 16. Multivariable Andersen Gill model for the outcome of peritonitis (sensitivity analysis using 45 day relapse/recurrence exclusion criteria) n = 4,247 patients ANDERSEN GILL MODEL Hazard ratio 95% CI p value Age (per decade) Black Asian Diabetes female male < Glomerulonephritis Transfer from HD <0.001 Failed transplant HD = hemodialysis, CI = confidence interval 66

77 Table 17. Multivariable Andersen Gill model for the outcome of peritonitis in the subgroup of patients with no submodality switch (sensitivity analysis using 45 day relapse/recurrence exclusion criteria) n = 3,180 patients ANDERSEN GILL MODEL Hazard ratio 95% CI p value Age (per decade) Black Asian Diabetes female male Glomerulonephritis < Transfer from HD <0.001 Failed transplant CAPD vs. APD HD = hemodialysis, CAPD = continuous ambulatory peritoneal dialysis, APD = automated peritoneal dialysis, CI = confidence interval 67

78 6.2 Figures Figure 1. Illustration of intraluminal (dashed arrow) and periluminal (solid arrow) entry of organisms into the peritoneal cavity Peritoneum Peritoneal cavity PD catheter 68