Background

Patients managed in the emergency room can be considered as having infection and requiring prompt antibiotic treatment because of organ failure. However, frequently the primary focus may remain unknown during the first hours after admission: 10 to 30% of patients do not have a definitive diagnosis following first clinical evaluation, which in turn has a negative impact on the final outcome [1,2,3,4,5]. No guidelines are available to help the clinician in such situation to choose the most appropriate treatment.

Due to antibiotic misuse, multidrug-resistant (MDR) bacteria have become a widespread concern in clinical medicine [6]. Several guidelines now take into consideration the risk of MDR bacteria involvement for antibiotic choice during empirical treatment [7, 8], even in the community setting. However, MDR bacteria prevalence is highly variable, depending on both patient-related and environmental factors [6, 8]. Data on MDR bacteria in the context of community-acquired infections are scarce except for cutaneous infections in the USA and urinary infections worldwide [2, 7, 9, 10].

The conjunction of diagnostic uncertainty with the background of enhanced rate of MDR bacteria leads the clinician to prescribe a combination of antibiotics as empirical treatment, especially if organ dysfunction is observed [11,12,13]. Based on microbiological data of bacteremic patients in the community setting, we aimed to identify the most suitable antibiotic combination for patient with diagnostic uncertainty, therefore at risk for unfavorable outcome.

Methods

Patient selection and characteristics

This was an observational study realized in Nice University Hospital, a tertiary care center with only one infectious diseases department. It was based on our medical dashboard, which was put into practice since July 2005 and previously described in [2, 3]. This dashboard works as a database, declared and approved by the French Data Protection Authority number 1430722.

As the software allows diagnosis or diagnosis-related group (DRG) selection, it is easy to study the main patient’s characteristics and evolution of a specific disease. Regarding severity, terminology used in patient’s final report was translated in the dashboard.

We included all patients with community-acquired bacteremia from July 2005 to June 2018.

Diagnostic uncertainty (DU) was defined by a discrepancy between diagnosis suspected at admission and final diagnosis at discharge. This definition included patients for whom no clear diagnosis appeared as the reason for hospitalization, e.g., mostly fever of unknown origin.

Bacteriological studies

We specifically checked in the patient’s chart the accuracy of the blood culture results and the community-acquired infection characteristics when the bacteria isolated was usually involved in nosocomial infections, such as ESBL-producing Enterobacteriaceae and Pseudomonas aeruginosa. All polymicrobial blood cultures were also assessed in the patient’s chart.

Blood cultures were collected directly during the venipuncture procedure using aerobic (Bact/AERT® FA Plus, Biomérieux, France) and anaerobic (Bact/AERT® FN Plus, Biomérieux, France) blood culture bottles, and were then sent to the laboratory and processed with an automated Bact/ALERT 3D system (BioMérieux, France). Specific culture for Mycobacterium spp. was also performed if clinically suspected. Bottles that showed a positive signal in the Bact/ALERT 3D system were routinely subjected to Gram staining and subcultured at least on blood agar plates and upon results of Gram on Drigalski agar or on chocolate agar. Colonies were identified using the API system (bioMérieux) and, since 2013, MALDI-TOF MS Microflex LT (Bruker Daltonics GmbH, Bremen, Germany) according to the manufacturer’s recommendation. Antibiograms were carried out by the diffusion method in Mueller-Hinton agar (MH BioMerieux SA, Marcy-l’Étoile, France) with BioRad discs (Marnes-la-Coquette, France) and interpreted according to the Antibiogram Committee of the French Microbiology Society recommendations using the Sirweb (I2A) software. Synergy was observed by placing third generation cephalosporin discs around discs containing clavulanic acid.

We specifically recorded microbial data regarding in vitro susceptibility to amoxicillin/clavulanic acid (AMC), cefotaxime or ceftriaxone (3GC), gentamicin (G), and their combinations, AMC+G and 3GC+G.

According to recent consensual definitions [14], inappropriate antibiotic therapy was defined as the use of antimicrobials to which the pathogen was resistant.

An effective antibiotic reassessment was defined as any modification (including the first introduction) of the initial antibiotic treatment, irrespective of the time of change.

Unfavorable outcome was defined as patient death during the hospital stay.

Statistical analysis

Data were analyzed with StatView software version 5.0 and statistical significance was established at α = 0.05. Continuous variables were compared with the Student’s t test or the Mann–Whitney non-parametric test. Proportions were compared with the χ2 statistic or Fisher’s exact test when appropriate. Logistic regression was used to determine in multivariate analysis the risk factor for all-cause in-hospital mortality. The results are presented as adjusted odds ratios (AORs), along with their 95% confidence intervals (CIs).

Results

Patient selection is described in Fig. 1. A total of 1034 CAB were included from July 2005 to June 2018, representing 8.7% of all community-acquired infections admitted in our department that had blood cultures collected.

Fig. 1
figure 1

Population study. Selection of community-acquired bacteremia according to successive exclusion criteria

Full size image

Main patient characteristics are presented in Table 1, according to diagnostic accuracy. Diagnostic uncertainty was observed for 357 patients (35%), mainly in respiratory infections DRG: 57/149 (38%). Also, these diagnostic uncertainties included 164/357 (46%) patients presented with fever of unknown origin. Regarding antibiotic treatment, 7 patients died before any therapeutic prescriptions, including 2 patients benefiting of palliative care; there were analyzed as inefficient treatment. Efficient empirical antibiotic treatment (eEAT) was observed in 924 cases (89%) and was less frequent in case of uncertain diagnosis: 87% vs 91%, p = 0.055. The rate of antibiotic reassessment (2 antibiotic treatments prescribed successively) was 53% (see Table 1).

Table 1 Patient comparison according to diagnosis accuracy. Univariate analysis. The presumed clinical diagnosis, at the time of the empirical antibiotic prescriptions, had to be considered by comparison with the final diagnosis. Diagnostic uncertainty was defined by a discrepancy between diagnosis suspected at admission and final diagnosis at discharge
Full size table

Unfavorable outcome was observed in 61 cases (5.8%). Risk factors for unfavorable outcome are shown in Table 2. eEAT (n = 553) was associated with a lower death rate compared to inefficient therapies without reaching statistical significance: 5.4 vs 10.0% (p = 0.053). Urinary source of CAB and effective antibiotic reassessment were protective factors of unfavorable outcome: AOR = 0.42, p = 0.003 and 0.34, p = 0.012 respectively. In contrast, neurological and/or psychiatric co-morbid conditions were associated with unfavorable outcome: AOR = 3.05, p < 0.001, as well as severe forms of infections: AOR = 5.09, p < 0.001, and ESBL positive strains bacteremia: AOR = 7.48, p < 0.001.

Table 2 Risk factors for unfavorable outcome. Univariate analysis and logistic regression analysis. Each 10 years or more was associated with an increase of the risk of death of 11%. AMC+G-R = resistance to amoxicillin/clavulanate+gentamicin; 3GC+G-R = resistance to cefotaxim+gentamicin
Full size table

As eEAT was consistent with a protective factor for unfavorable outcome, but was also inconstant, the question was the determination of the best empirical therapy. Blood culture results as well as susceptibility to both antibiotic combinations AMC+G and 3GC+G are indicated in Table 3. As expected, Enterobacteriaceae were predominant, accounting for 437 cases (42%), including 56 ESBL-producing strains (13%). Streptococci were more frequently isolated (21%) than Staphylococcus aureus (16%), among which only 3 cases of methicillin resistant S. aureus were detected. Of note, polymicrobial CAB amounted to 10% of the cases. In vitro data indicates that 250 bacteria were resistant to AMC (24%), 201 were resistant to 3GC (19%) and 376 were resistant to G (36%). Among the 56 ESBL-producing strains, 24 were also resistant to G, but all were susceptible to amikacin. Considering antibiotic combination, 5.8% of bacteria were resistant to AMC+G and 10.1% were resistant to 3GC+G (see Tables 1 and 3). A total of 47 (4.5%) strains was resistant to both antibiotic combinations, in particularly 24 ESBL producing Enterobacteriaceae. Accordingly, based on in vitro antibiotic susceptibility data, the most efficient antibiotic combination in the subgroup of uncertain diagnosis was AMC+G compared with 3GC+G: 92% vs 87%.

Table 3 Bacteria involved in 1034 community-acquired bacteremia over 13 years in one tertiary care center and antimicrobial resistance. Genre is indicated as well as the 3 main species involved. AMC+G-R = resistance to amoxicillin/clavulanate+gentamicin; 3GC+G-R = cefotaxim+gentamicin
Full size table

Discussion

Our work confirms previous studies showing that uncertain diagnosis is frequent and associated with a trend towards inappropriate empirical antibiotic therapy [1, 2, 12, 13]. Also, effective antibiotic reassessment is associated with a better outcome [1, 12, 15, 16]. Therefore, our study designates the vicious circle between uncertain diagnosis, inappropriate empirical antibiotic therapy, and unfavorable outcome in the absence of antibiotic reassessment. Thus, defining the best empirical antibiotic combination for patients with uncertain diagnosis is of paramount importance, especially because antibiotic reassessment is still limited in real-life practice [1, 15]. As the first criteria of drug choice in empirical antibiotic therapy is determined through the in vitro susceptibility of the suspected bacteria, our data indicated that AMC+G was superior to 3GC+G. Moreover, the superiority of AMC+G was significant in the subgroup of patients with uncertain diagnosis, at least in part due to the resistance of enteroccocal infections and polymicrobial bacteremia to 3GC+G (see Tables 1 and 3).

One limit of our study is the monocentric characteristic of the resistance epidemiology, so the resluts will be applicable to geographical areas and health care facilities displaying the same rate of MDR bacteria in the community setting with a similar CAB epidemiology. For example, our results will be not relevent in the USA, where methicillin-resistant S. aureus is common, especially in skin and soft tissue infections [7]. Also, the choice for the aminoglycoside compound is not unequivocal: gentamicin is certainly a major molecule in combination with a penicillin for Gram-positive cocci such as streptococci, but amikacin is usually a more effective drug for ESBL-positive strains [17]. The negative impact of MDR bacteria such as ESBL-positive strains on the outcome has been reported and have to be considered even in the community settings [9, 18]. Lastly, in accordance with our results, the bacteremic urinary infections was associated with a better outcome compared to the digestive tract infections [18, 19]. We have previously reported that the AMC+aminoglycoside combination is also a good choice for primary bacteremia, defined by the absence of clinical diagnosis and fruitless investigations [20].

Conclusion

Our studies and others suggest that in front of diagnostic uncertainty, but still in the community setting, the empirical treatment of choice could be AMC+aminoglycosides. These results have to be further validated in a prospective comparative study in order to reduce the negative impact of diagnostic uncertainty.