Characteristics and Risk Factors Associated with Early Mortality in Patients with Polymicrobial Bloodstream Infection: A Retrospective Clinical Study

Background and Objective Polymicrobial bloodstream infections (PBSI) in hospitalized patients are associated with increased mortality, while few studies have characterized the clinical features in this population. This study aimed to assess the risk factors and short-term prognosis of PBSI in hospitalized patients. Materials and Methods 4066 patients with culture-positive blood were included between January 1, 2015 and December 31, 2017 in the First Aliated Hospital of Zhejiang University School of Medicine (Hangzhou, China) in our study. 218 patients were diagnosed as PBSI. The patients were divided into two groups according to the outcome after 30-day follow-up. The number of survival group were 129, while the number of non-survival group were 89. The clinical data, identied microorganisms and severity models were compared between the two groups. A cox regression model was used to identify the risk factors of 30-day mortality in PBSI patients. Five prediction models were compared by Z-test to test the value of these models to predict outcome of PBSI. Results in more likely at the time and showed more They were more likely to develop to be and than Inappropriate initial empirical antimicrobial therapy (HR=1.713 95% p=0.027), white blood cell (HR=1.740 95% CI: 1.002-3.020, p=0.049) and platelet (HR=2.940 95% CI: 1.754-4.930, p<0.001) independent risk factors for 30-day mortality in PBSI SOFA (AUROC=0.882, 95% CI=0.832-0.922) scores was a good prognostic scoring system for predicting short-term mortality in PBSI patients. The SOFA score was valuable than the other four in the of PBSI according to the Z-test (p<0.05).


Introduction
Bloodstream infection continues to be a public health problem and it is closely linked with major morbidity and mortality worldwide. As reported, The prevalence rate of BSI is 19 million cases per year worldwide (1). In the USA, BSI is ranked as the 11th leading cause of death according to a 2008 survey (2) and it caused 600 fatalities every day (1). In China, according to a systematic review and metaanalysis of 72 studies in 2010, the weighted BSI in-hospital mortality rate was 28.7% (3).The diagnosis of BSI is relay on blood cultures obtained from a patient with clinical signs of infection while ruling out contamination. Routine blood culture is always too slow to support rapid therapeutic interventions (1).
The guidelines recommend that empirical antibiotics be used within the rst hour after sepsis or septic shock is detected (4). However, there are some challenges that empiric treatment may not cover the correct pathogen or contribute to the evolution of resistant microorganisms (1).
Polymicrobial bloodstream infection (PBSI) was a special subcategory of BSI. The rates of PBSI have been reported range from 5-20% among patients with BSI (5). The underlying medical conditions including malignancy, gastrointestinal and genitourinary disease, the presence of central venous catheters (CVC), are closely associate with occurrence of PBSI. Notably, PBSI may present different clinical presentations, microbiological characteristics and outcomes compared with monomicrobial BSI. Gram-negative organisms were the most frequent causative agents in PBSI (6)(7)(8)(9) which is contrary to BSI (10). The mortality rate of patients with polymicrobial BSIs ranged from 14-43%, approximately two times the mortality rate of those with monomicrobial BSIs (11). It was reported that inappropriate antimicrobial treatment was related to the high mortality (12).
Although high short-term mortality among patients with PBSI were noted, few reports have fully analyzed the clinical features and short-term prognosis of PBSI in hospitalized patients. How to screen for high-risk patients that we need pay special attention is still a problem. To improve the survival rate, we analyzed the risk factors for short-term mortality to optimize strati cation and no one had done it yet. .

Patients and settings
We identi ed blood culture-positive patients from the microbiology database. Total 4066 patients with BSI between January 1, 2015, and December 31, 2017, in the First A liated Hospital of Zhejiang University School of Medicine (Hangzhou, China) were screened in this study. PBSI was de ned as the growth of two or more organisms from blood culture specimens obtained from a patient within a period of < 72 hours (6).

Data collection
All the data of each patient were collected at the diagnosis of polymicrobial bloodstream infection from inpatient records; the collected data included sex; age; alcohol abuse; smoking; co-morbidity such as hypertension, diabetes, hematological malignancy and malignant parenchymal tumor; history of transplant; previous chemotherapy or corticosteroid therapy in the last 3 months; other sites of infection; laboratory examination, clinical presentation, microbiological data, treatments, ICU admission, and prognosis. For those patients who experienced multiple episodes of PBSI during the study period, only the rst PBSI was included in the analysis. We de ned inappropriate initial empirical antibiotic therapy as follows: At least one of the bacteria isolated was not sensitive to the antibiotics administered (13) or no antibiotic use before the positive blood culture (14). Nosocomial infections were con rmed after 48 hours of admission. Five prognostic models, the sequential organ failure assessment (SOFA), acute physiology and chronic health evaluation (APACHE II), systemic in ammatory response (SIRS), the quick sequential organ failure assessment (qSOFA) and simpli ed acute physiology score II (SAPS II) were used to predict early mortality. The SOFA score is widely used to diagnose organ failure in general intensive care units by evaluating liver, kidney, brain, coagulatory, circulatory or respiratory failure (15). The APACHE II score indicates the severity of disease through a disease classi cation system that generally uses a point score based on the initial values of 12 routine physiologic measurements, age and previous health status (16).
SIRS is used to describe the complex pathophysiologic response when at least 2 of the following four symptoms appeared: tachypnea, tachycardia, leukopenia or leukocytosis and fever or hypothermia (17).
The qSOFA was calculated by following criteria: systolic arterial blood pressure ≤ 100 mmHg; respiratory rate > 21 breaths/min; or altered mental status (17). The SAPS II includes 12 physiology variables, age, type of admission and three underlying disease variables (18). The APACHE II score and SAPS II were mainly evaluated in the rst 24 hours in the ICU but were calculated at the time of diagnosis of PBSI in this study.

Microbiologic studies
Blood specimens were collected with one or two tubes from both sides of each patient who presented with fever ≥ 38.5℃ or when BSI was suspected based on clinical signs within 5 min. The identi cation of blood isolates was carried out according to routine methods by the staff of the microbiology laboratory in our hospital.

Statistical analysis
The statistical analyses were performed by using SPSS software version 20 (IBM Inc., Chicago, IL, USA). The results are shown as mean ± standard deviations (SDs) for continuous variables or numbers and percentages for categorical variables. Student's t-tests were was used to analyze the differences between the continuous variables, and χ2-tests were used for the categorical variables. The variables that were statistically signi cant in the univariate analysis were included in a multiple cox regression model to analyze the potential predictors associated with mortality. The survival dates were analyzed with Kaplan-Meier survival curves. The different prognostic scoring systems were compared by receiver operating curves (ROCs).

Characteristics of patients
After excluding the contaminated specimens, 218 of 4066 patients with BSI from January 1, 2015, to December 31, 2017, were screened in this study which is shown in Fig. 1. The clinical characteristics, laboratory examination results, treatments, severity scores and outcomes of patients with PBSI were depicted in Table 1. According to the survival days after the PBSI diagnosis, they were divided into a survival group (more than 30 days) and a nonsurvival group (no more than 30 days). Among the 129 patients in the survival group, the mean age was 59.4 ± 18.2 years, and 89 patients (69.0%) were male.
Among the 89 patients in the nonsurvival group, the mean age was 60.1 ± 17.8 years, and 56 patients (62.9%) were male. The two groups did not show differences in their baseline characteristics, including age, sex, smoking and alcohol abuse. For comorbidities, such as hypertension, diabetes, cancer and history of transplantation, the two groups showed no differences. Moreover, the Charlson comorbidity index (CCI) was not different among the groups. Regarding treatment, patients in the nonsurvival group showed a higher frequency of receiving inappropriate initial empirical antibiotic therapy before the diagnosis than patients in the survival group (p = 0.011). There were no differences in the rates of catheter-related infection, previous chemotherapy, or previous corticosteroid therapy. Regarding other sites of infection, patients in the nonsurvival groups showed a higher frequency of pulmonary infection than patients in the survival group (survival vs. nonsurvival, p = 0.001). For other coinfections, such as urinary tract infections, cerebral infections, and skin and soft tissue infections, there were no differences. The patients in the nonsurvival group showed more serious in ammation responses than the patients in the survival group. C-reactive protein (CRP) levels (p < 0.001) were higher and platelet counts (p < 0.001) were lower in the nonsurvival group than in the survival group. In addition, the patients in the nonsurvival group had lower hemoglobin (p = 0.003) and hematocrit levels (p = 0.001) than the patients in the survival group, meaning that the patients in the nonsurvival group were more likely to have anemia. In addition, the patients in the nonsurvival group had longer prothrombin times (p = 0.002), lower albumin (p < 0.001), and higher total bilirubin (p = 0.011) than the patients in the survival group; these factors were related to worse liver function, and high blood urea nitrogen (p = 0.002) and lactic acid levels (p < 0.001).

Isolated microorganisms
The isolates from all the PBSI cases by group were shown in Table 4. There were 461 microorganisms isolated from the blood cultures of 218 patients with PBSI.    Gram-positive bacteria were identi ed in 69.2% of the PBSI samples, and gram-negative bacteria were identi ed in 71.6%. There were both gram-positive and gram-negative bacteria in 45.9% of the PBSI samples. Among the patients in our study, 20.6% were infected with three or more microorganisms. However, the survival days were not signi cantly different between the two-pathogen group and the three or more-pathogen group (p = 0.37). The Kaplan-Meier diagrams showed that PBSI patients with fungal infection had signi cantly higher overall 30-day mortality than those without fungal infection (p = 0.024). However, the survival analysis identi ed no difference between the groups with or without gram-positive bacteria and the groups with or without gram-negative bacteria (Fig. 2).
Value of the prognostic models in predicting 30-day morality in PBSI patients Five models, including the SOFA, APACHE II, SIRS, qSOFA, SAPS II were tested for the prediction of 30-day mortality in patients with PBSI (   (19). We identi ed that the 30-day mortality rate of PBSI was 40.8% (89/218) in our study. A previous study showed that the average mortality due to PBSI was 47% (11). However, its clinical presentation and the factor associated with prognosis were rarely reported. In our study, we observed that patients with poor prognosis showed more severe systemic in ammatory response, including elevated WBC and Creactive protein, reduced platelet, and more frequency to be septic shock (2). Inappropriate initial empirical antimicrobial therapy, WBC > 11.2*10 9 and platelet ≤ 54*10 9 were exhibited to be independent risk factors for PBSI-related 30-day mortality (3). Gram-negative organisms accounted for most pathogens, and Klebsiella pneumoniae and Acinetobacter spp. ranked the rst and second (4). It worth noting that patients co-infected with fungi was associated with worsen prognosis (5) SOFA scores was more accurate in predicting the prognosis of patients with PBSI.
To help direct our selection of treatment methods and improve patients' early survival rate, we discussed the risk factors for 30-day mortality. Inappropriate initial empirical antimicrobial therapy was a main prognostic factor for early mortality in this study. Effective initial empirical treatment before receiving the blood culture results is necessary. The choice of empirical antibiotics is often a challenge for physicians.
Some reports have demonstrated that appropriate antibiotic treatment can reduce mortality and improve clinical outcomes in patients with BSI (20). In our study, the rate of inappropriate initial empirical antibiotic therapy in the nonsurvival group was 30.3%, which was 16.3% (p < 0.05) higher than that in the survival group. In another report, the rate was 53.6% for PBSI patients in the emergency department. Among the patients who received inappropriate initial empirical antibiotic therapy, only 17.4% of the patients received no empirical antibiotic treatment. A delay in the application of effective initial empirical antibiotics has been reported to lead to poor outcomes in patients with BSI, especially critical patients (21). Therefore, increasing attention needs to be paid to initial empirical antibiotic therapy in patients with suspected infections. However, 41.3% of patients required combination therapy that was not received; for example, some patients had two pathogens that could not be treated by only one antibiotic. It is more di cult to administer adequate antibiotic therapy in patients with PBSI than in patients with monomicrobial BSI.
One antibiotic is not enough to treat multiple pathogens, even if it is a broad-spectrum antibiotic (14). To provide adequate empirical coverage, it is vital to measure clinical characteristics and evaluate risk factors for acquiring PBSI. We also discovered that WBC > 11.2 × 10 9 and platelet ≤ 54 × 10 9 were optimal cut-off points as another two prognostic factors. Leukocytosis and thrombocytopenia were common hematologic ndings in BSI, which means widespread systemic in ammation (1). It is an exaggerated defense response triggered by some pathogens. Additionally, Systemic in ammatory response syndrome (SIRS) will lead to dysregulated cytokine storm and even massive in ammatory cascade. These may resulted in organ dysfunction and death (22).
Gram-negative organisms were the most frequent pathogens in our study, similar to other reports (6,14). Among the gram-negative organisms, Klebsiella pneumoniae and Acinetobacter spp. were the most frequent causative agents. With the wider use of carbapenems, the carbapenem resistance rates of Klebsiella spp. and Acinetobacter spp. have increased to 37% and 69%, and few treatment options are available for these pathogens currently (21). Among the gram-positive organisms, Enterococcus spp. and coagulase-negative staphylococci were the most frequent. A report on PBSI in patients with cancer obtained the same conclusion (6). In contrast with our study, another report on PBSI in patients in the emergency department excluded patients infected with coagulase-negative staphylococci from their study and found that streptococci were the most frequent gram-positive bacteria (14). One study analyzed the isolated microorganisms in community-acquired BSIs and identi ed Escherichia coli, Staphylococcus aureus, and Streptococcus pneumoniae as the most common organisms (23). Considering the small sample sizes in the existing studies on PBSI, relatively large sample sizes are required to help us fully understand the microbiology of PBSIs, which will provide valuable information for empirical antimicrobial treatment.
We found that PBSI patients infected with fungi had poor outcomes. Candida spp. was the most frequent fungal species in our study (77.5%). A report comparing Candida with other isolated pathogens found that BSI patients infected with Candida had increased ICU mortality (21). Traditional initial empirical treatment does not always cover fungi. Waiting to administer antifungal therapy until culture results are returned may be detrimental to some patients. Only 62.5% of patients infected with fungi received antifungal therapy during the period of PBSI in our study. Some patients died before the culture results were returned. Therefore, we believe that timely antifungal treatment may improve the prognosis of patients with PBSI, if the patients are highly suspect co-infection with fungi.
Besides, we also evaluated some prognostic systems, which have been reported link with mortality of patients with BSI(24, 25). We found that the SOFA ,APACHE II and SAPS II all had good predictive accuracy in our study. The SOFA exhibited a higher AUROC than the other two, and the Z-test showed that the probability p value was less than 0.05. Therefore, we believe that SOFA was the better one. The SOFA score is used to evaluate the trend of organ dysfunction in patients(26). Multiorgan dysfunction is common in patients with BSI. It was reported that the SOFA performed well in predicting organ dysfunction in some patients with BSI (24). However, no previous study has considered the SOFA score in patients with PBSI. In some large-scale validation studies on the Sepsis-3 criteria for BSI patients, the median SOFA scores were 6 (IQR 3-9) in the US(27) and 5 (IQR 3-8) in Australia and New Zealand(28). The patients in our study had a median SOFA score of 9 (IQR 5-14). Patients with PBSI had a higher SOFA score than patients with BSI, suggesting severe clinical manifestations.
This study had some major limitations. First, it was a single-center study, which limited its generalizability. Second, the long-term prognosis of these patients was not analyzed in this study.

Conclusion
In summary, hospitalized patients with PBSI are at high risk for short-term mortality. Our data showed that inappropriate initial empirical antimicrobial therapy, white blood cell and platelet were independent risk factors for 30-day mortality in patients with PBSI. SOFA scores could be used to predict the shortterm prognosis of PBSI.

Availability of data and materials
Data analyzed during the current study are available from the corresponding author on reasonable request.
Software: Tiantian Ge, Chao Chen.  Figure 1 The ow diagram of patients selection Kaplan-Meier diagrams for 30-day mortality Figure 3 ROC curves of the ve prognostic scoring systems for predicting 30-day mortality in patients with PBSI.