Pseudomonas aeruginosa Community-Onset Bloodstream Infections: Characterization, Diagnostic Predictors, and Predictive Score Development—Results from the PRO-BAC Cohort

Community-onset bloodstream infections (CO-BSI) caused by gram-negative bacilli are common and associated with significant mortality; those caused by Pseudomonas aeruginosa are associated with worse prognosis and higher rates of inadequateempirical antibiotic treatment. The aims of this study were to describe the characteristics of patients with CO-BSI caused by P. aeruginosa, to identify predictors, and to develop a predictive score for P. aeruginosa CO-BSI. Materials/methods: PROBAC is a prospective cohort including patients >14 years with BSI from 26 Spanish hospitals between October 2016 and May 2017. Patients with monomicrobial P. aeruginosa CO-BSI and monomicrobial Enterobacterales CO-BSI were included. Variables of interest were collected. Independent predictors of Pseudomonas aeruginosa CO-BSI were identified by logistic regression and a prediction score was developed. Results: A total of 78patients with P. aeruginosa CO-BSI and 2572 with Enterobacterales CO-BSI were included. Patients with P. aeruginosa had a median age of 70 years (IQR 60–79), 68.8% were male, median Charlson score was 5 (IQR 3–7), and 30-daymortality was 18.5%. Multivariate analysis identified the following predictors of CO-BSI-PA [adjusted OR (95% CI)]: male gender [1.89 (1.14–3.12)], haematological malignancy [2.45 (1.20–4.99)], obstructive uropathy [2.86 (1.13–3.02)], source of infection other than urinary tract, biliary tract or intra-abdominal [6.69 (4.10–10.92)] and healthcare-associated BSI [1.85 (1.13–3.02)]. Anindex predictive of CO-BSI-PA was developed; scores ≥ 3.5 showed a negative predictive value of 89% and an area under the receiver operator curve (ROC) of 0.66. Conclusions: We did not find a good predictive score of P. aeruginosa CO-BSI due to its relatively low incidence in the overall population. Our model includes variables that are easy to collect in real clinical practice and could be useful to detect patients with very low risk of P. aeruginosa CO-BSI.


Introduction
Bloodstream infections (BSI) are a common consequence of invasive infections and cause significant morbidity and mortality worldwide. Their prognosis depends on several factors, including the aetiological agent, source of infection, appropriate antimicrobial treatment and the underlying conditions of the patient [1,2]. With respect to aetiological agents, invasive Pseudomonas aeruginosa infections are associated with high mortality rates, and prognosis is strongly associated with early active treatment, among other factors [3][4][5].
Since P. aeruginosa is predominantly a nosocomial pathogen, antipseudomonal antibiotics, especially antipseudomonal beta-lactams, in empirical regimens are often reserved for hospital-acquired infections to avoid overuse. Indeed, invasive infections due to P. aeruginosa outside the hospital setting are regarded as unusual [6,7]; in 2016, this microorganism caused 1.4% of community-acquired BSIs and 3.6% of healthcare-related BSIs in Spain [8]. However, inappropriate empirical antimicrobial treatment is associated with worse prognosis and mortality in P. aeruginosa BSI, including those that are communityonset [5]. Consequently, despite the low frequency, it would be useful to identify patients at risk of community-onset BSI due to P. aeruginosa in order to optimize their management.
The objectives of this study were to identify predictors of P. aeruginosa communityonset bloodstream infections (CO-BSI) and to develop a predictive score. In addition, we provide updated information on the epidemiological characteristics of community-onset P. aeruginosa bloodstream infections in Spain.

Design, Study Sites and Periods
The PRO-BAC study was a multicentre prospective cohort study conducted in 26 Spanish hospitals (18 tertiary and 8 community hospitals) between 1 October 2016 and 31 March 2017, including all episodes of clinically significant BSI in adult patients (>14 years); no exclusion criteria were applied [8]. All methodological details were previously published [8]. Blood cultures were obtained, processed and interpreted in accordance with standard recommendations [9,10]. The study was approved by the Ethics Committees of the participating centres, which waived the need for informed consent due to the observational design, and it was registered in ClinicalTrials.gov (NCT03148769). The STROBE recommendations were followed in this study (Supplementary Table S1) [11].
For this analysis, episodes of monomicrobial CO-BSIin the PRO-BAC cohort caused by P. aeruginosa or any Enterobacterales were eligible. Predictors of P. aeruginosa CO-BSI were studied by comparing exposure to different variables with those of CO-BSI caused by Enterobacterales, the reason for this being that P. aeruginosa is often considered in patients when Enterobacterales are already a reasonable aetiological option.

Variables and Definitions
Variables collected include the type of hospital; admission ward; demographics; type and severity of chronic underlying diseases according to Charlson index [12]; acute severity of underlying disease according to Pitt score [13]; type of acquisition [14]; exposure to invasive procedures and devices in the previous month (including vascular catheter, urinary catheter, parenteral nutrition, mechanical ventilation and major surgery); antimicrobial use in the preceding month; aetiology of BSI; source of infection; presentation with severe sepsis or shock according to standard [15] and Sepsis-3 criteria [16]; antimicrobial treatment; and 30-day mortality. Data were collected according to an electronic case report form specifically designed by local investigators and supervised by Infectious Diseases and Critical Care physicians BSIs were considered clinically significant if accompanied by signs or symptoms of infection, and contamination was ruled out [9,17]. In the case of potential contaminants, such as coagulase-negative staphylococci (CNS) or diphtheroids, only episodes in which the organism was isolated from ≥2 blood draws were considered [9].
Acquisition was considered community-onset if signs or symptoms of infection started <48 h after hospital admission or 48 h after hospital discharge, including strictly communityacquired cases and healthcare-related cases. Healthcare-related BSI was defined according to previously described criteria [14,18]. Source of BSI was based on clinical and laboratory data, using standard CDC criteria for secondary BSI [18]. For the diagnosis of catheterrelated BSI, differential time to positivity was employed at all sites typically used for surgically implanted indwelling or difficult-to-replacecatheters. Subsequent episodes in the same patient caused by the same pathogen were excluded unless they occurred more than >3 months apart.

Statistical Analysis
Data are expressed as proportions for categorical variables, and as either means with standard deviation, or median with interquartile range (IQR) for continuous variables, as appropriate. Chi-square and Student's t-tests were used to compare categorical and continuous variables, respectively. All time-dependent variables were measured with reference to the day when blood cultures were drawn (considered as day 0). All p values were two-tailed, and p values ≤ 0.05 were considered statistically significant. Independent predictors of CO-BSI due to P. aeruginosa bacteraemia were identified by logistic regression; variables with p value < 0.1 in univariate analysis and those considered potential predictors based on previous data were entered into the models, using a manual backward-stepwise selection procedure. Interactions between variables were checked. A predictive score was created using the final multivariate model; each predictive factor was assigned points, calculated by dividing the beta coefficients by the smallest beta coefficient in the model and rounding to the nearest unit. Positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity of the score were calculated. The predictive ability of the risk score for the observed data was checked by calculating the area under the receiver operating curve (AUROC). Data were analysed using SPSS version 23.0 (Chicago, IL, USA).
Specific microorganisms isolated in the group of 2572 Enterobacterales CO-BSIs were E. coli

Demographics, Comorbidities and Clinical Characteristics
Patient characteristics, sources of infection, clinical presentation, and CO-BSI mortality due to P. aeruginosa and Enterobacterales are shown in Table 1. Overall, patients with P. aeruginosa CO-BSI were younger, more frequently male, and had cancer, immunosuppressive conditions, healthcare-related BSI, and exposure to invasive procedures; the sources of BSI were also different (respiratory tract, unknown source and vascular catheter were more frequent among P. aeruginosa infections, while the urinary and biliary tract, and intraabdominal infections were less frequent). No statistically significant differences in severity were found at clinical presentation or ICU admission, although30-day mortality was higher among episodes caused by P. aeruginosa than than by Enterobacterales.
Specific antibiotic exposures in the 30 days prior to BSI are shown in Supplementary  Table S2. Patients with P. aeruginosa had been more frequently exposed to antibiotics, and specifically to antipseudomonal beta-lactams.

Risk Factors and Predictive Score Development
A multivariate analysis using a logistic regression model showed that the independent risk factors for a P. aeruginosa aetiology in CO-BSI were male gender [1.89 (1.14-3.  Supplementary Table S2 were statistically significant risk factors after adjustment for other variables. A predictive score was developed using the risk factors for P. aeruginosa CO-BSI identified in the multivariate analysis. The scores assigned to each significant variable according to their regression coefficient are listed in Table 2.The score had an area under the ROC curve of 0.66 (0.57-0.74), indicating a moderate predictive ability. The NPV, PPV, sensitivity and specificity for the different values of the score are shown in Table 3.  We also explored a model excluding patients with oncohaematological malignancies and neutropenia, as these patients would probably be candidates for antipseudomonal drugs anyway. The AUROC curve for the resulting model was 0.67 (0.55-0.74), showing a similar predictive ability.

Stratified Analysis by Source of Infection
To explorethe risk factors associated with P. aeruginosa CO-BSI versus Enterobacterales CO-BSI indifferent types of infection, a stratified analysis by source of bacteraemia was performed. Among

Discussion
P. aeruginosa is an aetiology of concern in invasive infections as it is often associated with frail or immunocompromised patients and is associated with high mortality rates. Importantly, therapeutic options are more limited than for other common bacteria, and resistance is increasing worldwide [6,[19][20][21][22]; in fact, the World Health Organization has listed carbapenem-resistant P. aeruginosa as a critical priority pathogen [23]. P. aeruginosa is considered an infrequent pathogen in non-nosocomial infections, despite the increasingly blurred boundaries between the hospital environment and the community. Furthermore, previous reports have shown that the epidemiological differences between community-acquired and healthcare-related BSI in all-cause bacteraemia were not well-define din the case of P. aeruginosa alone [24]. Treatment of P. aeruginosa is limited to a small number of antibiotics. Due to the well-known ability of P. aeruginosa to develop resistance, even during antibiotic treatment [20,21,25], empirical antipseudomonal antibiotics are mostly reserved for nosocomial infections. At the same time, delay in the administration of appropriate treatment is associated with increased mortality in P. aeruginosa BSI [3][4][5]26], which is crucial, especially when resistant or multi-drug-resistant P. aeruginosa is involved or suspected, due to the limited number of potentially useful antibiotics for these infections. However, the association between phenotypic resistance, virulence factors and biofilm-forming ability remains controversial [27].Thus, although the prevalence of P. aeruginosa as a cause of CO-BSI is low [7,8], it is necessary to identify in advance which patients would benefit from the use of antipseudomonal agents.
In this prospective multicentre bacteraemia cohort, we identified a substantial number of cases of P. aeruginosa CO-BSI and compared them with cases presenting with Enterobacterales CO-BSI. In our study, as reported previously [5], male sex, oncological disease, chronic pulmonary disease, haematological malignancy, immunosuppressive treatment and healthcare-related acquisition were more frequent in the P. aeruginosa CO-BSI group. In contrast toother studies [26,28], we observed that there were no significant differences in age or Charlson comorbidity index between P. aeruginosa CO-BSI and Enterobacterales CO-BSI after adjusting forsex, comorbidities and invasive procedures. Obstructive uropathy was also found to be an independent risk factor for P. aeruginosa CO-BSI. In the same vein, Esparcia et al. [29] developed a study only on community-onset urinary sepsis in the elderly, in which male sex, urinary tract neoplasia, recurrent urinary tract infection, indwelling urinary catheter and healthcare-related acquisition were found to be independent risk factors for community-onset P. aeruginosa.
To our knowledge, two studies reporting predictive scores for P. aeruginosa BSI have been published: the first one for all P. aeruginosa BSI cases, and the second for communityonset cases alone. Gransden et al. [30] presented a predictive score for all P. aeruginosa BSI. Eight variables were included; positive predictors were male gender, neutrophil count ≤1 × 10 9 cells/L, previous/current antibiotic treatment, corticosteroid or cytotoxic treatment, hospital acquisition, intensive care unit patient, high-risk source of bacteraemia, and low-risk source of bacteraemia as a negative predictor. With a score > 3, the probability of BSI due to P. aeruginosa was 19.3%. Schechner et al. [7] described a prediction score for community-onset cases of P. aeruginosa BSI, which included the following variables: age > 90 years, recent antimicrobial use, presence of a urinary device and presence of a central venous catheter. When≥ 2 predictive factors were present, the probability of community-onset P. aeruginosa BSI was nearly 33%, and the area under the receiving operator curve of the multivariate model was0.726. Consistent with both prediction scores, our model has a low positive predictive value and a high negative predictive value for community-onset P. aeruginosa BSI, and a diagnostic accuracy comparable to that reported by Schechner et al. The main strengths of our model are that it includes variables that are easy to collect in real clinical practice and at the patient's first contact with healthcare, allowing us to detect patients at very low risk of P. aeruginosa BSI.
Some limitations should be kept in mind. First, we did not collect data on the blood culture utilization rate, and although the participating sites collected the majority of blood cultures from their catchment population, we were unable to accurately calculate population-based incidence rates. Second, given the observational nature of our study, causality between the variables included in the model and community-onset P. aeruginosa BSI cannot be fully guaranteed. Nonetheless, similar variables have been previously explored for P. aeruginosa infections [5,7,30]. Furthermore, the low positive probability for community-onset P. aeruginosa BSI in our model, as in previously published models, implies the involvement of other unmeasured variables in the development of this disease. Future research should focus on identifying these as-yet undetected variables. Third, despite the fact that a large national prospective cohort was used, P. aeruginosa CO-BSI was uncommon and the frequency of cases was low. The limited predictive ability of the model is a likely corollary of this. Moreover, it has not been validated in an external cohort. Finally, and perhaps most importantly, our study inferred risk factors for pseudomonal aetiology in CO-BSI compared to Enterobacterales, limiting the population to Gram-negative bacteraemia. Because of this, the clinical utility of the model may be limited to the infections most frequently caused by gram negatives, such as those of biliary tract, intra-abdominal or urinary tract origin. The strengths of the study are prospective data collection and the high number of total cases recordedat26 hospitals indifferent regions of Spain with universal public healthcare coverage of the population, which provides a representative sample of P. aeruginosa CO-BSI in our country and makes it more easily generalizable.

Conclusions
Pseudomonas aeruginosa CO-BSI is a difficult entity to predict due to its relatively low incidence in the overall population. However, inappropriate empirical antimicrobial therapy is associated with a worse prognosis and higher mortality in P. aeruginosa CO-BSI. Our model could be useful to detect patients at very low risk of P. aeruginosa CO-BSI in cases of suspected Gram-negative bacteraemia, which would help to avoid overuse of antipseudomonal agents and thus reduce the selective pressure on this and other microorganisms.
Institutional Review Board Statement: The Ethics Committee for research from Hospital Universitario Virgen Macarena and Hospital Universitario Virgen del Rocío, approved the study for Andalucía (Código Promotor: FIS-AMO-2016-01). Approval was also obtained at each participating centre outside Andalucía, according to the local ethic committees for research.

Informed Consent Statement:
The need for informed written consent was waived because of the observational nature of the study. Data Availability Statement: Not applicable.