Predictive factors of outcome following nonfermenting gram-negative bacilli peritonitis in peritoneal dialysis

Background Peritonitis due to gram-negative bacilli (GNB), particularly nonfermenting GNB (NF-GNB), is a serious complication of peritoneal dialysis (PD) with a low resolution rate. Beyond the patient’s condition, microbiological properties such as antimicrobial resistance, biolms and the production of other virulence factors can explain the bad outcomes. This study aimed to evaluate the inuence of the patient’s condition, the microbiological characteristics, including biolm production, and the treatment of peritonitis on peritonitis resolution. Methods We reviewed the records of 62 new peritonitis episodes caused by NF-GNB that occurred between 1997 and 2015 at a single university center. The inuence of microbiological and clinical factors on resolution chance was analyzed by logistic regression. Results The etiologies were species of Pseudomonas (51.6%), Acinetobacter (32.2%), and others (16.1%). There was a high (72.9%) proportion of biolm producers’ lineages. The in vitro susceptibility rate of Pseudomonas spp. to amikacin, ciprooxacin, and ceftazidime was signicantly greater than that of Acinetobacter spp. and other species; however, there was a similar low resolution rate (<45%) among the episodes attributable to Pseudomonas spp, Acinetobacter spp, and other NF-GNB. Pre-existent exit-site infection was independently associated with nonresolution. No other factor, including biolm production, was associated with the outcome. Conclusions Peritonitis due to NF-GNB in PD is a severe infection with a reduced resolution rate, and pre-existent exit site infection negatively inuences the chance of resolution. The higher in vitro susceptibility of Pseudomonas compared to that of other NF-GNB with a similar resolution rate, suggests bacterial virulence factors beyond biolm and can act in concert, thereby worsening the outcome. agents favored positive outcomes in Pseudomonas-induced peritonitis. vitro susceptibility cefepime, ceftazidime determined minimum inhibitory gradient the test proportion of strains susceptible to each dened on the Clinical Laboratory breakpoints When strains presented intermediate MIC values, them resistant.


Background
Continuous peritoneal dialysis (PD) was introduced in seventies [1,2], and its initial results were compromised by the high incidence of bacterial peritonitis [3,4]. Since then, technological advances, particularly in disconnection systems and antimicrobial prophylaxis, have strongly reduced the incidence of these infections [5,6]. However, peritonitis remains a serious complication of PD and the main cause of PD failure and is associated with a higher risk of death from all causes and cardiovascular causes [7,8].
Gram-positive cocci are the main etiology of PD peritonitis worldwide, while episodes due to gram-negative bacilli (GNB) usually present greater severity and lower resolution rates [9,10]. Among them, the worst outcomes are reported in infections caused by Pseudomonas species and other nonfermenting GNB (NF-GNB) [11][12][13]. The ndings of a large prospective Brazilian cohort showed that Pseudomonas spp. etiology is independently associated with the nonresolution of peritonitis [14].
The reasons for the unfavorable evolution of NF-GNB peritonitis are not fully known. Beyond the patient's clinical and demographic characteristics and antibiotic treatment, factors associated with intrinsic bacterial virulence and antimicrobial resistance are possible determinants of worse outcomes [13,[15][16][17][18][19].
NF-GNB are ubiquitous and opportunistic microorganisms that are present in nature and the healthcare environment, where they cause different types of infections [20,21]. Pseudomonas spp. are the most isolated NF-GNB and of greatest clinical importance. Pseudomonas species' virulence factors enable them to invade tissues, proliferate rapidly, generate bio lms, and quickly develop antibiotic resistance and provide those species with great motility [16][17][18][19]. Acinetobacter species have been an increasing concern in PD due to their alarming rate of antibiotic resistance development; in particular, the Acinetobacter baumannii complex [13] can form bio lms and colonize catheters [22].
In turn, only a few studies have reported the factors in uencing the outcomes of NF-GNB-induced PD-related peritonitis, highlighting a study by Silva et al [11], who reported that the use of two antimicrobial agents favored positive outcomes in Pseudomonas-induced peritonitis.
Jointly analyzing the microbiological properties of the causative organism, patient-related conditions, PD modality, and peritonitis episode characteristics and its treatment can potentially identify the determinants of outcomes in NF-GNB peritonitis, but such an analysis has not been conducted in Brazilian or Latin American cohorts. Therefore, the present study aimed to investigate whether causative bacterial characteristics, including the ability to produce bio lm, as well as those of the patient, PD modality, peritonitis episode, and peritonitis treatment, in uenced the clinical evolution of NF-GNB-induced PD-related peritonitis.

Study population
In this retrospective study, we reviewed all episodes of PD-related peritonitis caused by NF-GNB that occurred between June 1997 and December 2015 in a single Brazilian university center. The exclusion criteria were: episodes with incomplete clinical data, relapse (episode caused by the same species or a negative culture result within 28 days of completion of antibiotic therapy), recurrence (episode caused by other species within 28 days after starting antibiotic therapy), and repeat episode (episode caused by the same species or after 28 days following completion of antibiotic therapy) The diagnosis of peritonitis was made when at least two of the following criteria were present: presence of a cloudy peritoneal e uent; abdominal pain; dialysate containing more than 100 leukocytes per mL (at least 50% polymorphonuclear cells); and positive culture of dialysate [23]. The outcomes were de ned as: resolution (disappearance of signs and symptoms within 5 days after the initiation of antibiotic therapy); relapse, refractory peritonitis (presence of turbid dialysate after 5 days of treatment with appropriate antibiotics); peritonitis-related death (death of a patient with active peritonitis or the death of a patient who had an episode within the previous 4 weeks) [23]; and nonresolution (catheter removal before the 5th day of treatment, refractory peritonitis, relapse, or peritonitis-related death).

Catheter insertion and care and dialysis procedures
The catheter placements were made under supervision of a senior nephrologist with percutaneous blind insertion of a double cuff straight Tenckhoff catheter using the Seldinger technique. Until 2003, no patients used antibiotic cream application at the catheter exit-site; from 2003 to 2006, we prescribed daily mupirocin cream, and from January 2007 daily gentamicin was prescribed to all incident patients. All patients used a semiocclusive dressing with sterile gauze and microporous adhesive tape.
Until 1999, the CAPD connection systems were the Y set type; the twin bag was introduced in 1999. APD was introduced in 1998, and its indication and prescription were based on clinical criteria or the patient's preference. For both PD modalities, we used standard glucose solutions, with low pH and high glucose degradation product (GDP) levels.
The diagnosis of exit-site and tunnel infection followed International Society for Peritoneal Dialysis (ISPD) criteria Exit-site infection (ESI) is the presence of purulent discharge, with or without erythema of the skin at the catheter-epidermal interface. Tunnel infection is the presence of clinical in ammation or ultrasonographic evidence of collection along the catheter tunnel.

Data collection
We recorded the following information for each case: date, preexistent ESI (ESI diagnosed until four weeks before a peritonitis episode), tunnel infection, topical antibiotic use at the catheter exit-site, initial antimicrobial treatment for peritonitis and adjustments, outcome, treatment time before the peritonitis episode (dialysis vintage) , patient's characteristics (age, sex, race [Caucasian or non-Caucasian], underlying kidney disease, previous peritonitis by other bacteria, PD modality (continuous ambulatory PD or automated PD), and characteristics of the causative germ (species, bio lm production capability, and in vitro antibiotic susceptibility).

Culture and storage
After diagnosis, each dialysate sample was processed following the recommendations of the ISPD [23]. Cultures were performed using the Bactec System (Becton Dickinson Company, Sparks, MD) and then seeded onto blood agar if there was signaling the positivity on the cultures bottles..
After isolation, identi cation, and susceptibility testing, strains were stored at -70°C.
For the present study, the stored samples were reisolated on MacConkey agar plates and reidenti ed. For this, isolates were gram-stained to con rm purity and determine each isolate's morphology and speci c color. Afterward, the isolates were identi ed by conventional biochemical testing [25] and by mass spectrometry using MALDI-TOF (matrix-assisted laser desorption ionization time-of-ight) technology (MALDI-ToF VITEK® MS, Brazil) [26].

Microbiological tests
In vitro susceptibility The in vitro susceptibility to amikacin, cipro oxacin, cefepime, imipenem, and ceftazidime was determined by the minimum inhibitory concentration (MIC)based on gradient diffusion, using the E test ((bioMérieux, Inc., Durham, NC) The proportion of strains susceptible to each drug was de ned based on the 2016 Clinical Laboratory Standards Institute breakpoints [27]. When strains presented intermediate MIC values, we considered them resistant.

Bio lm production
The bacterial samples were grown in Tryptic Soy Broth (TSB) (BD™, Le Pont de Claix, France) at 37°C for 18 hours. To assess the bacterial ability to adhere to abiotic surfaces, we used 96-well polystyrene plates and added 200 µl of TSB and 10 µ of the bacterial suspension (approximately 108 CFU/mLµ) to each well, except one well that was inoculated only with culture medium to be used as a reading standard (blank). The plates were incubated at 37°C for 48 hours and then washed with phosphate-buffered saline 4 times to remove non-adherent bacteria. Bacteria that adhered to the abiotic surface were then xed with formalin (2%), and after 20 minutes, the formalin was removed, and the preparations were washed 4 more times with water. Then, the preparations were stained with a crystal violet solution (1%) for 20 minutes, after which they were washed 3 times with water to remove excess dye. After drying, the dye was solubilized with methanol for 10 minutes, and the optical density, measured at 540 nm, was determined [28]. Then, we classi ed the bio lm production into one of four categories as previously published [28]: no producer, weak producer, moderate producer, and strong producer. In our study, we opted for a 48 hours method, instead of faster methods, to obtain a more reliable result, as it allows the strains to have enough time for the production of bio lm because some NF-GNB species show slow growth.

Clinical-microbiological associations
Each patient's characteristics, pre-existent ESI, topical antibiotic use at the catheter exit-site, initial treatment for peritonitis, treatment adjustment with two antipseudomonal antibiotics, previous peritonitis by other bacteria, and microbiological characteristics were analyzed regarding their association with the outcome. For this purpose, we classi ed the outcomes into two mutually exclusive results: resolution or nonresolution.

Statistical analysis
For comparison between frequencies, we used the chi-squared test or Fisher's exact test. Binary logistic regression with a backward stepwise procedure was used to determine the independent predictors of outcomes. For this purpose, we rst performed a univariate logistic regression analysis to select the variables that would enter the nal model, with p> 0.20 as the elimination criterion. Collinearity among variables was tested, and if statistically signi cant interactions occurred, one of the variables was excluded. A p value < 0.05 was considered signi cant.

Results
Between June 1997 and December 2015, there were 726 episodes of bacterial peritonitis in 542 PD patients in our center. Of these, 194 (26.7%) were caused by GNB, 70 of which were caused by NF-GNB. Based on the exclusion criteria, we studied 62 index cases of peritonitis caused by NF-GNB from 62 patients (Figure 1). Pre-existent ESI was diagnosed in 16 cases (25.8%): nine were caused by Pseudomonas aeruginosa, three by Burkholderia cepacia, one by Stenotrophomona maltophilia, one by Acinetobacter baumanii, and two by Corynebacterium spp. These same organisms were seen in peritoneal e uent culture in 12 patients. For all ESI episodes, we prescribed oral cipro oxacin as the initial treatment. No tunnel infection was diagnosed by clinical criteria, although we did not perform ultrasonographic evaluation of the subcutaneous catheter tunnel.
Previous peritonitis caused by other bacteria was reported in 23 cases (37.1%). The clinical and demographic data of the 62 patients at the time of their rst episode of peritonitis are shown in Table 1.  (2 mg/kg daily) . When the results of peritoneal e uent culture and in vitro susceptibility tests were available, we adjusted the treatment. The duration of antibiotic therapy was at least 21 days [23]. Cefazolin plus amikacin was used in 20 cases, cefazolin plus ceftazidime was used in nine cases, and vancomycin plus amikacin was used in 33 cases. After the results of the in vitro susceptibility tests, we adjusted the initial treatment in 18 cases, (two antipseudomonal agents in 16, imipenem in one, and cefepime in one patient).
Of the 62 episodes, there was resolution in 20 (32.2%), relapse in six (9.6%), refractory peritonitis in 19 (30.6%), removal of the peritoneal catheter before the 5th day of treatment in 14 (22.6%), and peritonitis-related death in three (4.8%). Regarding the 16 episodes, in which we made adjustment of the initial treatment to two antipseudomonal agents, six (37,7%) progressed to resolution, while both patients treated with imipenem or cefepime, after prescription adjustment evolved to resolution. We did not observed allergic reactions attributable to antibiotics regimens.
The descriptions of the etiological agents are shown in Table 2. Of the total episodes of peritonitis included in the study, microbiological tests were carried out in 48 episodes, as it was not possible to recover the other strains. The results of in vitro susceptibility are described in the Table 3. Pseudomonas species were more susceptible than Acinetobacter species to all of the tested antimicrobials, except imipenem. Pseudomonas species were also more susceptible than Achromobacter species to amikacin, cipro oxacin, and cefepime. Isolates of Burkholderia cepacia and Stenotrophomonas maltophilia were tested only for ceftazidime, and all were susceptible. Table 3. Non-fermenting Gram-negative bacilli-causing peritoneal dialysis-related peritonitis and their in vitro susceptibility rates 1=p<0.05 vs Acinetobacter spp, 2=p<0.05 vs Achormobacter Regarding bio lm production, of the 48 samples, 35 (72.9%) produced bio lm. There were 18 strong producers, seven medium producers, and 10 weak producers. The bio lm producers were 22 of the 24 Pseudomonas isolates, 11 of the 18 Acinetobacter isolates, and one of the three Achromobacter isolates. Both Burkholderia cepacia and Burkholderia gladioli isolates were not producers, while the only isolate of Stenotrophomonas maltophilia was a producer. Among the episodes caused by bio lm, there were nine cases with ESI, and among those by nonproducers, there were four (p=0.72).
Factors associated with peritonitis outcome

Univariate analysis
The univariate logistic regression analysis revealed that pre-existent ESI, age, resistance to ceftazidime, and initial treatment with cefazolin plus ceftazidime were associated with a higher risk for nonresolution of peritonitis at a p value <0.20 ( Table 4). The other variables (underlying kidney disease, gender, race, PD modality, dialysis vintage, topical antibiotic use, previous peritonitis due to other bacteria, bio lm production, resistance to amikacin, and treatment adjustment with two antipseudomonal agents) did not reach p < 0.20 and were not included in the multivariate model.
There was collinearity between resistance to ceftazidime and initial treatment; therefore, it was not possible to include them together in the same regression model.

Multivariate analysis
This analysis showed that only pre-existent ESI was an independent predictor of nonresolution (4). We constructed a second model (model 2), including the covariate treatment and removing bacterial resistance, in which we observed a tendency for the protocol of cefazolin plus ceftazidime (using other initial treatments as a reference) to be associated with nonresolution, retaining pre-existent ESI as a predictor of nonresolution (Table   5).

Discussion
In human infections, the clinical course and outcome are strongly dependent on the characteristics of the infecting microorganism and the patient's condition. In the case of bacteria, despite the indisputable role of bacterial resistance, this does not seem to be the only property in uencing the outcomes; this holds true for PD-related peritonitis. Previous publications by our group showed a high number of virulence factors among Staphylococcus aureus lineages, some of which are associated with worse PD-related peritonitis outcome, despite their low resistance rate to methicillin [29,30]. In a previous report on PD-related Escherichia coli peritonitis episodes, we did not nd any resistant strain to amikacin or ceftazidime used as initial treatment; in contrast, only 48.1% of the episodes progressed to resolution [31]. In the same study, approximately 50% of the isolates were medium or strong bio lm producers, which tended to be associated with nonresolution, but this did not reach statistical signi cance (p=0.09) [31].
NF-GNB present both a high antimicrobial resistance rate and high production of virulence factors, such as bio lm, as con rmed in the present series, which potentially could explain the low observed resolution rate. In addition, important virulence factors are present in NF-GNB, particularly Pseudomonas species, which induce bacterial adhesion, destruction of cell membranes, and inhibition of the macrophage response, in addition to other actions [16-19, 32,33].
Interestingly, we did not nd a high resistance rate among episodes caused by Pseudomonas species, over 80% of which were susceptible to frequently used antimicrobials such as amikacin and ceftazidime. Even so, the resolution rate of these episodes was just over 30%, such as that observed with peritonitis caused by other NF-GNB, which suggests the in uence of other factors on the outcome. Moreover, the adjustment of the initial therapy for antibiotic regimens with two antipseudomonal drugs did not have association with the outcome. On the other hand, over 90% of the Pseudomonas isolates were bio lm producers. We emphasize that antibiotic susceptibility is based on the MIC of the drug for planktonic cells, which are more sensitive to antimicrobials than cells wrapped in bio lms [34].
The aggressive character of Pseudomonas spp. can explain, at least partially, the ndings of this and of the two largest studies, which previously described peritonitis caused by these bacteria as the most frequent etiology of peritonitis among NF-GNB. Silva et al. studied 191 episodes of col peritonitis that occurred in Australian patients who reported high rates of catheter removal (44%), permanent hemodialysis transfer (35%), hospitalization (96%), and change to a second antibiotic (66%). Lu et al. [12] reviewed 153 episodes of peritonitis caused by Pseudomonas species in Hong Kong, reporting an overall primary response rate of 53.6% and complete cure rate of 42.4%. Interestingly, that study showed a decrease in the incidence of germs resistant to ceftazidime and gentamicin over time.
A signi cant number of NF-GNB-induced peritonitis cases involved Acinetobacter species. Of these, the majority were due to Acinetobacter baumannii, similar to previous reports [13,35]. The importance of these bacteria has increased in recent years due to its great capacity to acquire mechanisms of resistance to different classes of antibiotics, great ability to survive and adapt to adverse conditions, and ability to adhere to different surfaces by the formation of bio lms [36]. This series con rms that Acinetobacter baumannii is resistant to several antimicrobials, except for imipenem. As expected, over 50% of the strains were bio lm producers. Despite their greater bacterial resistance compared to Pseudomonas spp., the resolution rate was similar between them. In this way, we can speculate that virulence factors in uenced in the outcome.
Other identi ed germs, such as Achromobacter species, lineages of the Burkholderia cepacia complex, and Burkholderia gladioli, have rarely been described as etiologies of PD-related peritonitis [37,38]. The precise identi cation of NF-GNB is a challenge for conventional microbiology due to the phenotypic similarity and taxonomic complexity of these agents. Phenotypic tests based on morphology and biochemical characteristics often provide erroneous identi cation of these species [39]. In our study, such limitations were minimized with the identi cation of the isolates by the MALDI-TOF technique, which is used in clinical microbiology to identify bacterial species based on microorganisms' protein pro les. This identi cation technique was mentioned in the most recent ISPD guideline on PD-related peritonitis, although at the time of its publication there was insu cient evidence for its recommendation [23].
Our study showed that pre-existent ESI was the only independent predictor of nonresolution. The association between ESI and subsequent peritonitis is widely recognized [40]. Staphylococcus aureus and Pseudomonas aeruginosa are the most common microorganisms causing ESI, and they can to tunnel along the subcutaneous pathway and lead to peritonitis [40]. On the other hand, only few publications have focused the in uence of ESI on peritonitis treatment response [41][42][43], reporting the presence of this infection is associated with bad outcomes. According to Gupta et al [41] in such cases, antibiotics do not resolve the peritoneal infection, although transient clearing of the e uent may occur.
Of note, in this study the bio lm production did not in uence the outcome; however, this result does not rule out the possibility that bio lm , in concert with other virulence factors, may in uence the peritonitis evolution.
Our study has several limitations, the most important being the small sample size, aggravated by the impossibility of recovering approximately 20% of the isolates. In addition, we do not have data at this time on the production of virulence factors by bacteria, other than bio lm. Lastly , there is a lack of information on patients' nutritional status and any comorbidity scores. However, this is a study of NF-GNB-induced peritonitis as a whole, and therefore allows comparisons between peritonitis episodes due to Pseudomonas species and those due to other NF-GNB. In addition, to our knowledge, this is the rst study to address the role of bio lm production in the outcomes of NF-GNB-induced PD-related peritonitis. In addition, it revealed novel information about pathogens that cause peritonitis, including those of the genus Achromobacter, but suggests that there is some bene t to using new techniques, e.g., MALDI-TOF, to identify bacteria in peritonitis.
Finally, the prevalence of PD-related NF-GNB in our center was similar to that of the Brazilian PD cohort, the largest Latin American cohort of incident PD patients, and again highlighted the severity of these infections.

Conclusions
NF-GNB-induced PD-related peritonitis is a serious infection with a reduced resolution rate. Bacterial resistance the concomitant presence of ESI negatively in uence the chance of resolution. Bio lm production was not signi cantly associated with the outcome, which does not rule out the possibility that it can act in concert with other virulence factors and thus impair the response to antimicrobial therapy. The presence of uncommon etiologies of peritonitis in PD, such as Achromobacter species, highlights the need for future studies regarding the clinical behavior of these infections. The considerably high prevalence of multi-resistant Acinetobacter species causing PD-related peritonitis raises an alert about care for the prevention and management of these infections.