From Epidemiology of Community-Onset Bloodstream Infections to the Development of Empirical Antimicrobial Treatment-Decision Algorithm in a Region with High Burden of Antimicrobial Resistance

Antimicrobial-resistant (AMR) infections have increased in community settings. Our objectives were to study the epidemiology of community-onset bloodstream infections (BSIs), identify risk factors for AMR-BSI and mortality-related factors, and develop the empirical antimicrobial treatment-decision algorithm. All adult, positive blood cultures at the emergency room and outpatient clinics were evaluated from 08/2021 to 04/2022. AMR was defined as the resistance of organisms to an antimicrobial to which they were previously sensitive. A total of 1151 positive blood cultures were identified. There were 450 initial episodes of bacterial BSI, and 114 BSIs (25%) were AMR-BSI. Non-susceptibility to ceftriaxone was detected in 40.9% of 195 E. coli isolates and 16.4% among 67 K. pneumoniae isolates. A treatment-decision algorithm was developed using the independent risk factors for AMR-BSI: presence of multidrug-resistant organisms (MDROs) within 90 days (aOR 3.63), prior antimicrobial exposure within 90 days (aOR 1.94), and urinary source (aOR 1.79). The positive and negative predictive values were 53.3% and 83.2%, respectively. The C-statistic was 0.73. Factors significantly associated with 30-day all-cause mortality were Pitt bacteremia score (aHR 1.39), solid malignancy (aHR 2.61), and urinary source (aHR 0.30). In conclusion, one-fourth of community-onset BSI were antimicrobial-resistant, and one-third of Enterobacteriaceae were non-susceptible to ceftriaxone. Treatment-decision algorithms may reduce overly broad antimicrobial treatment.


Introduction
Bloodstream infections (BSIs) are a leading cause of morbidity and mortality globally [1,2].The epidemiology of BSI has evolved and differs considerably between developed and developing countries.Asia is considered a high burden region of antimicrobial resistance [3].Multidrug resistance was presented in 30% of cases of Gram-negative bacteremia in a community hospital in Thailand [4].The mortality rate of BSI varies from 12% for community-onset BSI [5], 22% in a population-based cohort study [6], 30% for patients who have severe comorbidities, and 40-60% for intensive care unit patients [2].Increasing antimicrobial resistance (AMR) is a significant cause of death worldwide.The highest burden is in low-resource settings [7].Mortality in Escherichia coli and Klebsiella BSI strongly depends on resistance to fluoroquinolones or third-generation cephalosporins and on adequate therapy [8].The median turnaround times were 0.80, 1.81, and 2.71 days for Gram stain, identification of organism, and antimicrobial susceptibility test (AST) results, respectively [9].A prompt selection of empirical antimicrobial treatment without knowing whether an infection is antimicrobial-resistant, while balancing the risk of ineffective treatment versus excessively broad antimicrobial therapy is difficult.Understanding the extent of community-onset BSIs would help address the magnitude and impact of AMR and develop a solution.
We aimed to describe the contemporary epidemiology and outcomes of bacteremia in a community setting at emergency room (ER) and outpatient clinics.Our secondary objective was to identify independent factors associated with AMR in patients with BSI and develop the empirical antimicrobial treatment-decision algorithm for patients with suspected community-onset bloodstream infections.
depends on resistance to fluoroquinolones or third-generation cephalosporins and on adequate therapy [8].The median turnaround times were 0.80, 1.81, and 2.71 days for Gram stain, identification of organism, and antimicrobial susceptibility test (AST) results, respectively [9].A prompt selection of empirical antimicrobial treatment without knowing whether an infection is antimicrobial-resistant, while balancing the risk of ineffective treatment versus excessively broad antimicrobial therapy is difficult.Understanding the extent of community-onset BSIs would help address the magnitude and impact of AMR and develop a solution.
We aimed to describe the contemporary epidemiology and outcomes of bacteremia in a community setting at emergency room (ER) and outpatient clinics.Our secondary objective was to identify independent factors associated with AMR in patients with BSI and develop the empirical antimicrobial treatment-decision algorithm for patients with suspected community-onset bloodstream infections.

Clinical Characteristics of Patients with Antimicrobial-Resistant BSIs (AMR-BSIs) and Non-Antimicrobial-Resistant BSIs (NAMR-BSIs)
A cohort of 450 unique adult patients with bacterial BSIs with AST were identified during the study period and included in the comparative analysis for risk factors associated with AMR.Of these, 114 BSIs (25%) were antimicrobial-resistant.The baseline char-  A cohort of 450 unique adult patients with bacterial BSIs with AST were identified during the study period and included in the comparative analysis for risk factors associated with AMR.Of these, 114 BSIs (25%) were antimicrobial-resistant.The baseline characteristics of AMR-BSI compared to NAMR-BSI patients are shown in Table 1.The majority of bacterial BSIs were detected at ER (70%).The median age in the AMR-BSI group and NAMR-BSI group was 74 years (interquartile range [IQR] 57-83 years) and 71 years (IQR 59-80 years), respectively.The severity of the acute illness index was not different between the two groups.Both groups had a Pitt bacteremia score of 1 (IQR 0-2).Some differences between the two groups were observed.Patients with AMR-BSI were more likely to have a neurological disease, connective tissue disease prior admission, colonization or infection with multidrug-resistant organisms (MDROs), previous antimicrobial exposure within 90 days, and a higher proportion of urinary sources.

Appropriateness of Antimicrobial Use
There were 441 unique adult patients with bacterial BSI receiving at least one dose of an antimicrobial.Of 328 patients with NAMR-BSI, 190 (57.9%) received empirical treatment with broad spectrum active coverage.In total, 34 (30.1%)patients in 113 AMR-BSI patients were empirically treated with inactive spectrum coverage.Appropriate definitive treatment was not significantly different between AMR-BSI and NAMR-BSI patients.Optimal drug and duration in the AMR-BSI and NAMR-BSI groups were 33.0% and 36.3%, respectively.The median duration of treatment was 13 days in both groups.The possibility of shortening treatment duration in the AMR-BSI group and NAMR-BSI group were 57.8% and 50.2%, respectively (Table 4).

Proposed Empirical Antimicrobial Treatment Algorithm for Patients with Suspected Community-Onset Bloodstream Infections
Important risk factors of AMR from the analysis were integrated into treatmentdecision algorithms.The following triage steps were created: identifying patients with clinical symptoms and signs suspecting bacterial BSI and stratifying patients by risk of AMR.Patients with the highest risk were defined as those who had MDROs during the preceding 90 days.These patients would be treated with broad spectrum antimicrobials.Patients not meeting this definition would be reviewed for prior antimicrobial exposure within 90 days; those with no exposure would be treated with narrower spectrum antimicrobials.Patients previously exposed to antimicrobials within 90 days would be evaluated for the suspected source of infection; those with suspected urinary source would be treated with broad-spectrum antimicrobials, and those with suspected non-urinary source would be treated with narrower spectrum antimicrobials.The development of a treatment-decision algorithm is depicted in Figure 2. The sensitivity and specificity of the algorithm for predicting AMR-BSI were 49.1% (95% CI 36.9-58.7%)and 85.4% (95% CI 81.2-89.0%),respectively.The PPV and NPV were 53.3% (95% CI 45.4-61.1%)and 83.2% (95% CI 80.4-85.6%),respectively.A receiver operating characteristic curve (ROC) curve derived from a logistic regression comprising the three most important variables yielded a C-statistic of 0.73.treated with narrower spectrum antimicrobials.The development of a treatment-decision algorithm is depicted in Figure 2. The sensitivity and specificity of the algorithm for predicting AMR-BSI were 49.1% (95% CI 36.9-58.7%)and 85.4% (95% CI 81.2-89.0%),respectively.The PPV and NPV were 53.3% (95% CI 45.4-61.1%)and 83.2% (95% CI 80.4-85.6%),respectively.A receiver operating characteristic curve (ROC) curve derived from a logistic regression comprising the three most important variables yielded a C-statistic of 0.73.Empirical treatment following the algorithm resulted in 14.9% of patients with NAMR-BSI receiving broad spectrum active coverage, and 17.7% of patients with AMR-BSI receiving inactive spectrum coverage.
Sensitivity analyses were performed on the subset of 298 patients who had Enterobacteriaceae BSIs.The three significant risk factors for AMR remained similar to the entire dataset.The sensitivity and specificity of the algorithm for predicting ceftriaxone-resistant BSI were 57.5% (95% CI 46.8-67.6%)and 81.9% (95% CI 75.9-86.9%),respectively.The PPV and NPV were 59.3% (95% CI 50.7-67.2%)and 80.7% (95% CI 76.6-84.2%),respectively.The C-statistic was 0.76.The internal validation of 227 bacterial BSIs revealed a C-statistic of 0.77 (Table 5).Empirical treatment following the algorithm resulted in 14.9% of patients with NAMR-BSI receiving broad spectrum active coverage, and 17.7% of patients with AMR-BSI receiving inactive spectrum coverage.

Discussion
The most common pathogens of community-onset BSI at our institution included E. coli, K. pneumoniae, and S. aureus.This is consistent with previous reports of communityacquired bacteremia in Thailand and other countries [10][11][12].These top three pathogens accounted for 61.4% of all bacterial isolates.Other leading pathogens were beta-hemolytic streptococcus and Salmonella spp., which were different from previous studies [10][11][12].A surveillance of community-acquired bacteremia in Thailand during 2016-2017 found that E. coli and K. pneumoniae had susceptibility rates to ceftriaxone of 73% and 98%, respectively [10].In contrast, our study revealed lower susceptibility rates for these bacteria with susceptibility rates to ceftriaxone of 60.1% and 83.6%, respectively.The incidence of antimicrobial resistance continues to rise with a change driven by an increase in communityonset cases.The incidence of ceftriaxone-resistant Enterobacteriaceae infection increased by 53% from 2012 to 2017 according to the US Centers for Disease Control and Prevention [13].
In the present study, the independent risk factors for antimicrobial-resistant BSI included the presence of MDROs within 90 days, prior antimicrobial exposure within 90 days, and urinary sources in our study.These factors were similar to those identified in previous studies [3,14].Antimicrobial selective pressure was linked to bacterial resistance [15].High rates of community-onset antimicrobial-resistant infection have been occurring worldwide, predominantly in urinary tract infection with E. coli [16].Ceftriaxone-resistant uropathogens were isolated in 21.3% of patients with acute cystitis in Thai general practice clinics from 2014 to 2016 [17].Similarly, a study at a tertiary care hospital in Thailand reported that 22.3% of E. coli causing community-acquired UTI were ceftriaxone-resistant in 2017 [18].
The Pitt bacteremia score and solid malignancies were associated with an increase in the overall 30-day mortality in our study.The severity of illness has been well established in predicting mortality [19].All patients with solid tumors who died during the 30-day follow-up period in our study had advanced-stage malignancy.Despite no survival advantage, antibiotics were administered in 82% of patients with terminal cancer within three days of death at an academic hospital in a retrospective study conducted in Korea [20].Appropriately directed palliative care can reduce aggressive antimicrobial use near the end of life.It would benefit individual patients' quality of life and decrease selection pressure that can lead to MDROs.The majority of inactive empirical antimicrobials in our study were against ceftriaxone-resistant Gram-negative BSI.Studies have shown mixed results regarding the association between the effectiveness of empirical antimicrobial treatment and ceftriaxone-resistant Gram-negative bacteremia [14,21,22].The finding that inadequate empirical antibiotic treatment does not significantly impact the mortality in our study is consistent with previous studies [14,22].The mortality in our study was below 10%, which is comparable to previous studies [14,22].The patients in our study were not severely ill (median Pitt bacteremia score of 1).Relatively low mortality in community-onset bacteremia could be primarily driven by underlying conditions and disease severity [22].The impact of empirical antimicrobial choice on mortality may be limited in this scenario.Urinary source was significantly associated with lower mortality in our study.A multi-center study in English acute hospitals found that patients with urinary tract-related bacteremia were less acutely unwell [22].Piperacillin-tazobactam (TZP) is commonly used as an empirical treatment in our setting.The multinational, randomized, controlled trial of patients with ESBL-producing bacteremia (MERINO study) showed that definitive treatment with TZP increased 30-day mortality compared to meropenem, and no difference in mortality between urinary versus non-urinary source [23].However, Sharara et al. reported no differences between TZP versus carbapenems in the clinical resolution or mortality for the treatment of ESBL-producing pyelonephritis [24].A urinary pharmacokinetics study found that high TZP concentrations in urine and could result in treatment efficacy [25].A small randomized trial showed that indwelling catheter replacement before initiating antimicrobial therapy significantly decreased bacteriuria and time to clinical improvement [26].High TZP concentration in urine and biofilm removal in catheter-associated UTI may contribute to better outcomes compared to non-urinary-source infections.
The most frequent inappropriate prescribing was empirical broad spectrum antimicrobial treatment was 57.9% among NAMR-BSI in our study.A study evaluating practice at ER reported that inappropriate antimicrobial prescription in adult patients was 36.9% [27].Short courses of antimicrobial therapy (6-10 days) have been shown to have comparable clinical outcomes as prolonged courses of therapy (11-16 days) for Gram-negative bacteremia [28].The median duration of antimicrobial treatment for the uncomplicated bacterial BSIs was 13 days in our study; shorter courses were possible in more than half of cases in our study.Although we have institutional empirical treatment guidelines and weekly handshake stewardship in the ER, there is a more pressing need to develop initia-tives to improve ER-based antimicrobial prescribing and emphasis on optimal treatment duration [29].
Potential risk factors driving the emergence of antimicrobial-resistant bacterial infections have been identified in various studies.However, integration of multiple risk factors into actual practice is scarce.Clinicians continue to face a significant challenge when treating serious Gram-negative infections due to the difficult balance between the risk of ineffective agents versus overly broad empiric antimicrobial treatment.A prior study developed an easy-to-use clinical decision algorithm to determine the probability of an extended-spectrum beta-lactamase (ESBL)-producing bacterial BSI in a bacteremic patient that could aid in selecting appropriate empiric treatment [3].However, it could not be applied in regions with high ESBL prevalence.
From the analysis of risk factors for antimicrobial resistance, we developed a decision tree algorithm with three predictors; the presence of MDROs within 90 days, antibiotic exposure in the previous 90 days, and urinary tract infection source.There is always a trade-off between sensitivity and specificity.The ability to correctly predict NAMR-BSI cases (specificity) is essential to ensure the lowest risk of ineffective therapy.Patients classified as AMR-BSI cases by the algorithm were 53.3% more likely to be true AMR-BSI cases (PPV), and patients classified as NAMR-BSI cases were 83.2% more likely to be true NAMR-BSI (NPV).The subset of patients with Enterobacteriaceae BSIs yielded 6% higher PPV and 2.5% lower NPV.Empirical treatment following the algorithm would reduce broad spectrum antimicrobial therapy in NAMR-BSI cases by 43.0% and inactive spectrum antimicrobial in AMR-BSI cases by 12.4% in this dataset.This easy-to-use algorithm could improve the prediction of AMR-BSIs and reduce inappropriate empirical antimicrobial use.However, this algorithm cannot replace clinical judgment.Relevant components, such as clinical appearance, underlying conditions, and concern level of clinicians should be incorporated into decision-making.
There are several limitations in our study.First, this study was conducted at a single center.Our results may not be generalizable to patients in other settings with different prevalences of antimicrobial-resistant bacteria.Our findings should be validated in other cohorts.Second, the presence of MDRO was reviewed from our clinical microbiology laboratory reports, and previous antimicrobial exposure was retrospectively reviewed from the medical records.However, cases visiting outpatient clinics and ER were usually established patients who had received healthcare services at our hospital, and missing data on these two independent factors were likely small.Finally, the broad clinical approach included both Gram-positive and Gram-negative bacteria in our algorithm.Ceftriaxone is recommended and the most commonly used empirical treatment for community-acquired sepsis at our institution.Many Gram-positive bacteria non-susceptible to penicillin are susceptible to ceftriaxone.A subset of AMR-BSI predicted by this proposed algorithm would include ceftriaxone-susceptible Gram-positives, in which ceftriaxone is reasonable to use as an empirical treatment.We performed the sensitivity analyses on Enterobacteriaceae BSIs, which yielded 6% increased PPV and 2.5% decreased NPV.This finding suggests that the algorithm is robust to Gram-positive and Enterobacteriaceae.This broad-range approach would better represent real-world practice when the initial presentation cannot distinguish between Gram-positive versus Gram-negative organisms.

Study Population
We conducted a retrospective cohort study at Ramathibodi Hospital, a 1300-bed tertiary-care hospital in Bangkok, Thailand, between 1 August 2021 and 15 April 2022.All positive blood cultures from patients aged > 18 years at ER and outpatient clinics were identified.Only the initial episode of bacterial BSI with AST results was included in the comparative analysis of risk factors for antimicrobial-resistant infection.

Data Collection
Information regarding demographics, pre-existing medical conditions, and the severity of acute illness on day 1 of BSI, including Quick Sequential Organ Failure Assessment (qSOFA) score, Pitt bacteremia score, intensive care unit (ICU) admission, mechanical ventilation, vasopressor administration, receipt of antimicrobial treatment during preceding 90 days, antimicrobial-resistant bacterial colonization or infection during the prior 90 days, and microbiological and mortality data were obtained from medical records.Mortality and cause of death were assessed at 30 days.Duplicate isolates of the same species with the identical AST profile recovered from consecutive blood cultures on the same patient after the index BSI were excluded from cumulative AST.

Definitions
Community-onset BSI refers to the location of the onset of BSI episodes which includes community and long-term healthcare facilities.BSI is defined as the positive growth of the organism (s) from blood specimen (s) in ≥1 blood culture bottle taken from a patient with compatible clinical features of infection.The isolated bacteria are classified as contaminants if they are common commensal organisms on the skin or environment e.g., coagulase-negative staphylococci, Bacillus spp., Corynebacterium spp., Propionibacterium spp., Aerococcus spp., Micrococcus spp., and the patient has no compatible clinical syndrome that could be caused by such organisms.Polymicrobial BSI is defined as isolation of ≥2 different pathogens from the same blood sample.The source of BSI is determined based on clear clinical evidence that the BSI was linked to focal infection at another body site.
Antimicrobial resistance is the resistance of organisms to an antimicrobial to which they were previously sensitive.For the purpose of this study, antimicrobial-resistant bacteria are defined as follows; Gram-negative bacteria other than P. aeruginosa that exhibit in vitro non-susceptibility to ceftriaxone, P. aeruginosa that exhibits in vitro non-susceptibility to ceftazidime, and Gram-positive bacteria that exhibit in vitro non-susceptibility to penicillinclass drugs (penicillin, ampicillin, or oxacillin).

Microbiological Testing
All isolates were tested for their antimicrobial susceptibility by an automated microbroth dilution testing system (Sensititre™; Thermo Fisher Scientific, Cleveland, OH, USA).Clinical and Laboratory Standards Institute (CLSI) clinical breakpoints were used to interpret the minimum inhibitory concentration (MIC) values [30].

Assessment of the Appropriateness of Antimicrobial Use
Independent adjudication of the appropriateness of antimicrobial treatment was performed based on institutional antimicrobial guidelines by three infectious disease specialists.The local antimicrobial guidelines for common bacterial infections consisted of the recommended antimicrobials for empirical therapy and the recommended dosage of each antimicrobial [29].Antimicrobial treatment was considered active when isolated pathogens were susceptible in vitro to the prescribed antimicrobial.Antimicrobial de-escalation or escalation encompassed the antimicrobial change within 48-72 h of available culture and susceptibility test results.
All BSI episodes wherein at least one antimicrobial was prescribed were randomized, and each was assigned to two specialists.Each expert independently assessed the antimicrobial prescription into specific categories for appropriate antimicrobial use modified from a previous study [31] (Table S1 in the Supplementary Data).

Statistical Analysis
Continuous variables were summarized as the mean (standard deviation) for normally distributed data or median (interquartile range) for non-normally distributed data.Categorical variables were displayed using absolute counts and percentages.Comparative analysis of variables associated with antimicrobial resistance was conducted using the

Figure 1 .
Figure 1.Distribution of pathogenic organisms associated with community-onset bloodstream infection.

Figure 1 .
Figure 1.Distribution of pathogenic organisms associated with community-onset bloodstream infection.

Table 1 .
Baseline characteristics of 450 unique adult patients with AMR-BSI and NAMR-BSI.

Table 2 .
Univariable analysis and multivariable analysis of risk factors for antimicrobial resistance.

Table 3 .
Hazard ratio for 30-day all-cause mortality for adult patients with community-onset bacterial bloodstream infection.

Table 4 .
Appropriateness of antimicrobial treatment of 441 unique adult patients with AMR-BSI vs. NAMR-BSI receiving at least one dose of an antimicrobial agent.
Abbreviation: IQR, interquartile range.a In patients with uncomplicated BSI, and not in palliative care within 5 days.

Table 5 .
Sensitivity, specificity, positive predictive value, negative predictive value, and C-statistics of the algorithm in predicting antimicrobial-resistant infection.