Non-PSEUDO-AntiBiotiCs Score—A Predictor of Bacteraemia Caused by Pseudomonas aeruginosa: A Case-control Study


 Background: Immoderate use of broad-spectrum antimicrobials could lead to emergence of resistant bacteria. The purpose of this study was to identify factors associated with Pseudomonas aeruginosa bacteraemia and develop an exclusion scoring system to help clinicians select an appropriate antimicrobial treatment. Methods: This single-centre case-control study recruited inpatients and outpatients (age ≥ 20 years) with P. aeruginosa or Escherichia coli bacteraemia at St. Luke’s International Hospital in Tokyo from April 2005 to March 2020. Bivariate associations were assessed using χ2 test, Fisher's exact test, or Mann–Whitney U test, and the relationship between P. aeruginosa bacteraemia and other variables was determined using multivariable logistic regression analysis. Results: A total of 1562 patients (208 patients with P. aeruginosa bacteraemia and 1354 patients with E. coli bacteraemia) were included. Multivariable analysis revealed 11 variables associated with P. aeruginosa bacteraemia: Nosocomial infections, Pneumonia, Sex (males), Exposure to antibiotics within 90 days, Urinary tract infection, Urinary catheterization, abDOminal infection, Age < 77 years, Body mass index < 19, presence of Central venous catheter, and Central line-associated bloodstream infection/peripheral line-associated bloodstream infection; these variables were used to develop the Non-PSEUDO-AntiBiotiCs score. The area under the receiver operating characteristic curve was 0.89 (95% confidence interval: 0.88–­0.92), and the best cut-off-point was 5; a score of ≥ 5 had a sensitivity of 80% and a specificity of 83%. Conclusion: We developed the Non-PSEUDO-AntiBiotiCs score. This score may allow clinicians to rule out the possibility of P. aeruginosa bacteraemia and prevent the abuse of broad-spectrum antimicrobials.


Background
Pseudomonas aeruginosa bacteraemia (PAB) has a higher mortality rate (29%) [1] than bacteraemia caused by other gram-negative rods, such as Escherichia coli and Klebsiella spp. [2]. P. aeruginosa is an innately antibiotic-resistant microorganism, and its toxicity mechanisms have been shown to be associated with restricted permeability of the cell wall, enhanced activity of the e ux system, and presence of the ampC gene for inducible chromosome beta-lactamase [3]. Empiric antimicrobial treatment failure is known to result in higher mortality rates in both community-acquired and nosocomial infections [4]. Therefore, clinicians often need to determine whether P. aeruginosa is a causative organism based on previously reported risk factors.
Notably, the abuse of broad-spectrum antimicrobials for treating P. aeruginosa could lead to emergence of resistant bacteria and an increase in adverse events [5]. Because inadequate empirical treatment is related to all-cause mortality and infection-related mortality due to PAB [1], appropriate and prompt antimicrobial treatments are essential for the management of PAB. Risk factors for PAB have already been identi ed and include male sex [1,6,7], haemodialysis [1,7], solid organ transplants [1], cancer [1], human immune de ciency virus infection [1], diabetes [8], neutropenia [6,7,9], haematological malignancy [7,9,10], central venous catheter [11,12], healthcare-associated infection [7,11], prior hospital admission [7], and prior antimicrobial use [12,13]. Although these factors are useful for suspecting the presence of PAB, to the best of our knowledge, no scoring systems are available to help clinicians rule out the possibility of PAB. Therefore, the purpose of this study was to identify factors associated with PAB and develop an exclusion score system that will assist in clinical practice.

Study Aim and Design
This study aimed to identify factors associated with Pseudomonas aeruginosa bacteraemia and develop an exclusion scoring system to help clinicians select an appropriate antimicrobial treatment. A retrospective, single-centre, case-control study was performed at St. Luke's International Hospital (a 520bed hospital) in Tokyo, Japan, between April 2005 and March 2020.

Inclusion and Exclusion Criteria
Inpatients and outpatients (age ≥ 20 years) with PAB or E. coli bacteraemia (ECB) were selected from the electronic hospital database based on positive blood culture ndings. Patients with E. coli detected by blood culture during the corresponding period were included in the control group, and those with multiple bacteraemias were excluded. In addition, in patients with multiple episodes of bacteraemia, only the rst episode of bacteraemia was included in the analysis.

Data Collection
Patient data were extracted from inpatients' and outpatients' electronic medical records. Study variables included patient demographics, comorbidities, probable causes of bacteraemia, presence of foreign bodies, use of immunosuppressive drugs, chemotherapy, antibiotics within 90 days before bacteraemia, hospitalization within 1 year, surgery within 1 year, vital signs, laboratory results and microbiological ndings, length of stay, admission to the intensive care unit, and mortality.
Variable De nitions PAB and ECB were de ned by the laboratory ndings of at least one set of positive blood cultures collected using standard sterile techniques [11]. Neutropenia was de ned as a neutrophil count of less than 500/mm 3 [8]. The source of bacteraemia was comprehensively determined through a review of the electronic discharge summary and the daily chart record of patient symptoms, physical examinations, laboratory results, and radiological ndings. A central venous catheter, such as a central venous port catheter, peripherally inserted central catheter, or temporary or permanent catheter for haemodialysis, was de ned as an intravascular access device or a catheter whose end is placed in or near the heart or in one of the main veins [8, 14,15]. Malignancies included both solid tumours and hematologic malignancies that had been treated with chemotherapy within the last 90 days. Central line-associated bloodstream infection (CLABSI) was classi ed as an infection of central venous catheters and peripherally inserted central venous catheters using the Centres for Disease Control and Prevention de nitions [16], and peripheral line-associated bloodstream infection (PLABSI) was de ned as bacteraemia with at least one of the following criteria: existence of phlebitis and/or disappearance of clinical symptoms after peripheral line withdrawal, and/or careful exclusion of bacteraemia as another focus [17,18]. Intraabdominal infections included cholangitis, cholecystitis, intestinal obstruction, oesophageal varices, appendicitis, diverticulitis, peritonitis, liver abscess, intra-abdominal abscess, and acute pancreatitis. Skin and soft tissue infections included erysipelas, cellulitis, subcutaneous abscess, necrotizing fasciitis, and wound infections. Respiratory diseases included chronic obstructive pulmonary disease (COPD), asthma and COPD overlap syndrome, bronchiectasis, interstitial pneumonia, diffuse panbronchiolitis, and bronchial asthma. Nosocomial infections (healthcare-associated infections) were diagnosed when patients had a positive blood culture after 48 hours of hospitalization. Community-acquired bacteraemia was diagnosed when it occurred in the outpatient setting or within 48 hours of hospitalization; however, patients who ful lled one or more of the following criteria were categorized as having healthcareassociated infections [11,19]: 1) underwent intravenous therapy at home or in an outpatient setting within the past 30 days; 2) attended a hospital, haemodialysis clinic, or received intravenous chemotherapy within the past 30 days; 3) had more than two days of hospitalization in the past 90 days; and 4) spent two or more days in a nursing home or in an assisted living facility in the past 90 days.

Microbiological Tests
We used the BacT/ALERT 3D system (bioMérieux, Inc.) for blood cultures and the MicroScan WalkAway 96PLUS (Beckman Coulter, Tokyo, Japan) for susceptibility testing.

Statistical Analyses
In bivariate associations, we used the chi-square or Fisher's exact test for categorical variables and the Mann-Whitney U test for continuous variables. Variables signi cantly associated with PAB in univariate analysis were applied to the CART model as a part of the multivariable analysis after removing highly correlated explanatory variables in terms of the variance in ation factor for multicollinearity. For continuous variables, those that were statistically signi cant in univariate analysis were converted into binary categorical variables using the CART model. As a result, variables that appeared at higher branch levels in the tree of the CART model and those that were clinically important among variables that were statistically signi cant in univariate analysis were included as explanatory variables in the multivariate logistic regression analysis. The nal explanatory variables of the model were selected by the forward stepwise approach. The receiver operating characteristic (ROC) curve was plotted with the model, and the values of the cut-off-point and area under the ROC curve were calculated from this ROC curve.
The statistical signi cance threshold was set at P < 0.05. All statistical analyses were carried out using the statistical software program SPSS 19.0J (IBM Japan, Tokyo, Japan) and EZR (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is a graphical user interface for R (The R Foundation for Statistical Computing, Vienna, Austria).

Study Participants
Of the 1592 observations, 14 observations (7 patients with both PAB and ECB) were excluded, and the rst episodes of bacteraemia of 16 patients with multiple episodes of bacteraemia were included. Ultimately, 1562 patients (208 PAB and 1354 ECB) were included in the study (Fig. 1). Comparisons of demographic characteristics, comorbidities, and site of infection between the PAB and ECB groups are shown in Table 1   To develop a prediction system for PAB, we focused on the beta coe cients corresponding to the variables that were statistically signi cant in the multivariable analysis. We calculated the weight scores by dividing each beta coe cient by the smallest value among the absolute beta coe cients and rounding the divided beta coe cients to develop the understandable rule. The obtained weight scores In the nal analysis, the area under the ROC curve was 0.89 (95% CI: 0.88-0.92) (Fig. 2), and the best cutoff point was 5. A score of 5 or higher had a sensitivity of 80%, a speci city of 83%, a positive predictive value of 42%, and a negative predictive value of 96%.

Discussion
To the best of our knowledge, this study is the rst to develop a clinically predictive (exclusion) score for PAB, "Non-PSEUDO-AntiBiotiCs score", using 11 elements that were signi cantly different in multivariable analysis. Males, age < 77 years, BMI < 19, placement of the urinary catheter, indwelling central venous catheter, nosocomial infection, use of antibiotics within 90 days, pneumonia, and CRBSI as a source of bacteraemia were independent associated factors for PAB, and the probability of PAB tended to be low if the score was less than 5. The current study differs from previous studies in that it included larger sample size and provided a simple and useful prediction score in the clinical setting. This scoring system would help clinicians select empirical antibacterial therapy.
It is often di cult for clinicians to decide whether P. aeruginosa should be considered in the setting of suspected community-acquired and nosocomial infections, such as pneumonia, urinary tract infection, and intra-abdominal infections. While the failure of empirical treatment in patients with PAB has already been shown to increase the mortality rate [20], abuse of broad-spectrum antimicrobial agents could lead to multidrug-resistant bacteria [5] and increase adverse events, such as Clostridium di cile infections [21], thereby resulting in longer hospital stays and additional costs [22]. Therefore, it is important to select appropriate empirical antimicrobials based on patients' risk factors and clinical ndings.
Two previous large studies determined the risk factors for PAB using ECB as a control group and found that an indwelling central venous catheter (CVC) [11], neutropenia [11], presentation with septic shock [11], healthcare-associated infection [11], and respiratory tract infection [8] were independent risk factors. Similarly, the present study also showed that the presence of CVC and respiratory infection were independent factors for PAB [8,13]. In addition, urinary tract infection was found to be one of the negative factors for PAB; this nding is consistent with that in a previous study [8].
In comparison to ndings in previous studies, one of the new independent associated factors identi ed in this study was CLABSI/PLABSI. This result could be explained by the fact that P. aeruginosa is the most common gram-negative rod causing CLABSI/PLABSI, while E. coli is relatively rare [23,24]. A previous study showed the above trend, but the nding was not statistically signi cant [11]. The larger sample size might have contributed to the signi cant ndings in this present study. The current study also showed that BMI < 19 was an associated factor for PAB. Because this study was a cross-sectional study, it is unclear whether low BMI is a cause or a consequence of PAB; however, P. aeruginosa infection could be a cause of emaciation because chronic infection with P. aeruginosa has been reported to produce pyocyanin, which leads to weight loss [25]. In addition, patients with chronic underlying diseases, such as bronchiectasis and cancer, may develop malnutrition [26,27]. These patients may have contracted P. aeruginosa infection during their medical treatment in hospitals. Moreover, the ndings of other signi cant associated factors, including age, male sex [1,6,7], exposure to antibiotics [12,13], and urinary catheterization [12], are consistent with those in previous studies. However, it is possible that some hidden confounding factors could not be identi ed in the present retrospective study since it addressed only cases of recent hospitalization for a period of less than one year, and this could be an issue for future studies.
Since delays in treatment of PAB are known to be associated with increased mortality [1,20], early prediction of PAB may contribute to patients' clinical outcomes. Our scoring system may help clinicians determine whether P. aeruginosa should be considered a causative pathogen of infection. Moreover, the use of this scoring system may reduce unnecessary broad-spectrum antimicrobial use, thereby preventing the emergence of multidrug-resistant bacterial pathogens.
This study has several limitations. First, this study was performed at a single institution, and the sample size was relatively small. Second, this was an observational study; thus, it is possible that unknown risk factors were distributed unequally between the two groups. Third, this study included only patients with positive blood cultures. Therefore, this scoring system cannot be applied to all patients with negative blood cultures. Since it is often unclear whether bacteraemia is present in the early onset of the disease, this scoring system should be interpreted expansively and judged carefully together with clinical information. Fourth, this scoring system cannot be used if the source of infections remains unknown at the time of diagnosis because clinicians need to use the source of infection to calculate the score.

Conclusion
We developed the Non-PSEUDO-AntiBiotiCs score. This scoring system may allow clinicians to rule out the possibility of P. aeruginosa bacteraemia and optimize the use of broad-spectrum antimicrobials.   The receiver operating characteristic curve of the Non-PSEUDO-AntiBiotiCs score In the nal analysis, the area under the receiver operating characteristic curve was 0.89 (95% CI: 0.88-0.92), and the best cut-off point was 5. A score of 5 or higher had a sensitivity of 80%, a speci city of 83%, a positive predictive value of 42%, and a negative predictive value of 96%.