Bacteremia is defined by the presence of viable bacterial agent in the bloodstream and is diagnosed in daily clinical practice with the use of blood cultures. The annual incidence of community-onset bacteremia is evaluated between 40 and 154/100000 [1]. The range of mortality rates is extremely large depending on the occurrence of septic shock, antimicrobial resistance, age, comorbidities and community or healthcare-acquired infections. Bacteremia corresponds to bacterial dissemination through the bloodstream from an initial infectious focus, among which urinary sepsis, pneumonia and intravascular catheter-related infection are the most frequent [2, 3]. Nevertheless, even with an exhaustive diagnostic procedure, this portal of entry sometimes remains unknown.

The definitions of BUO vary significantly according to the different authors. In a study of Vallès et al. [4] the absence of a distal source documented is sufficient, whether other authors use a composite criterion [5] combining non-contributing physical exam, lack of microbiological investigations identifying the same bacteria as in bloodcultures and normal radiological exams.

BUO is associated with significant morbidity compared to bacteremic patients with an identified infectious focus [6]. Even in intensive care units (ICU) where huge risk factors as septic shock or ICU scoring systems often overshadow other clinical determinants, BUO has been associated with inappropriate antibiotic treatment [4] and poor outcome [7].

When empirical antibiotic treatment is required, the decision process usually includes the infection focus identified by physical exam or the first radiological results. During BUO, presenting as sepsis of unknown origin, these data are missing and could favour inappropriate antibiotic administration, which is known to increase mortality [8]. Nowadays, the challenge for empirical antibiotic prescription is to combine appropriate antibiotic use and to consider the risk of antimicrobial resistance in the community setting especially for 3rd generation cephalosporin resistant Enterobacteriaceae [9].

While studies focusing on BUO are scarce we present herein a cohort of 101 community-acquired BUO seen in our department. This study aims to describe clinical presentation and the main features of patients involved. On the basis of the microbiological results of our cohort we have tested the efficiency of two antibiotic associations which both include a beta-lactam with an aminoglycoside. The main objective is the characterization of a treatment that could be used, by extension, during sepsis of unknown origin.

Between July 2005 and April 2016 the medical dashboard recorded 13,576 hospitalizations. A bacteremia was detected among 1989 (15%) patients with 261/1989 (13%) classified in the group BUO. After applying the inclusion criteria to the 261 patients, 101 community-acquired BUO were retained (Fig. 1). The characteristics of the population studied are presented in Table 1. Thirty-one (31%) patients were immunocompromised: 12 with neoplasms (diagnosed during the last year), six with cirrhosis, six HIV-infected (two with CD4 T-cells under 200/mm³), three splenectomized patients, three with immunosuppressor drugs and 1 chronic renal insufficiency undergoing dialysis.

Clinical characteristics of patients	Values (%)
Age, years mean (SD)	71 (17)
Sex, men (%)	68 (67)
Comorbidities
Cardiovascular	61 (60)
Immunosuppressed	31 (31)
Pulmonary	27 (27)
Diabetes mellitus	17 (17)
Dyslipidemia	14 (14)
Neurological	14 (14)
Medical device	17 (17)
Severe sepsis	52 (51)
Death	9 (9)
ICU transfer	12 (12)
Symptoms at admission
Fever	99 (98)
Isolated fever	22 (22)
Dyspnea	14 (14)
Fatigue	13 (13)
Diarrhea/vomiting	13 (13)
Confusion	13 (13)
Loss of consciousness	11 (11)
Back pain	8 (8)
Arthralgia/myalgia	6 (6)
Fall	6 (6)
Chest pain	4
Abdominal pain	3
Neurological deficit	2
Jaundice	1
Cough	1
Headache	1

An antibiotic treatment had been prescribed the month preceding the BUO in 11/101 (11%) patients. Fever was detected at day 1 in 99/101 (98%) patients. The median duration of symptoms leading to the emergency room visit was 3 days. Severe sepsis was identified in 52/101 (51%) patients. ICU management was required for 12/101 (12%) patients. CRP and PCT values at admission were available in 98/101 (97%) and 47/101 (46%) of the cases respectively. The CRP value was below 20 mg/L in 17/98 (17%) patients and PCT value below 0.5 ng/L in 10/47 (21%) patients. A thoracic and abdominal CT-Scan was performed in 69/101 (68%) patients, colonoscopy in 37/101 (36%), TTE in 56/101 (55%) patients and TOE in 29/101 (28%).

This study conducted in one French University Hospital shows that community BUO mainly involves old male patients, with half of them presenting at least one organ dysfunction and requiring ICU admission in 12% of the cases. Isolated fever with a recent onset (less than 3 days) represents the primary clinical feature for 22% of the patients whereas symptoms leading to emergency room visit vary widely for the rest of the patients. S. aureus BUO was associated to unfavourable outcome. Based on the analysis of the antibiotic susceptibility tests AMC + GM provided a higher rate of adequacy than 3GC + GM when used empirically.

The dashboard of our ward classifies infective endocarditis and vertebral osteomyelitis in other groups. Unfortunately in those groups i.e. bone and joint infections and vascular infections, the portal of entry is not recorded. So a number of BUO with vascular or bone secondary focus has not been included thus representing a selection bias. In the same way exclusion of chronic cutaneous conditions could have lead to selection bias. Bacteriological analysis of samples collected on chronic cutaneous conditions as pressure ulcers is often polymicrobial and variable over time. Deep tissue biopsies are not routinely performed for this purpose so the exclusion of a cutaneous primary focus is never obvious. Since the most frequent clinical presentation of coagulase-negative Staphylococci is device-associated health care-associated infections [12] BUO with those bacteria had been excluded from our work focusing on community-acquired infection. The choice of the two beta-lactams studied associated to an aminoglycoside results from the research for a spectrum broad enough but with consideration for appropriate antibiotic use and ecologic impact.

In our dashboard, BUO corresponds to 13% of all the bacteremia. This result is close to the rates reported in a Spanish Emergency Department (12%) [13] and in a Spanish tertiary care center (16%) [5]. During the challenging management of BUO biomarkers may be helpful to the clinician. However, in our study, with a cut-off value of 0.5 ng/L the PCT yield a 21% rate of false negative. PCT is usually recognized as a biomarker able to rule out a bacteremia in patients presenting with fever in the emergency department [14]. In daily clinical practice a CRP value below 20 mg/L is usually considered not to be associated with a bacterial infection [15]. Compared to PCT, this cut-off provides a lower rate of false negative: 17%.

Bacterial epidemiology for S. aureus, Enterobacteriaceae and polymicrobial infections among the BUO are consistent with the few previously published data [5, 13, 16] while viridans Streptococci, Enterococci and group D Streptococci are more frequent in our study.

The high rate of patients presenting with at least one organ failure (51%) together with the 29% increase of the in-hospital mortality rate compare to the other patients hospitalized during the study period underscore the need to identify the best-suited empirical treatment. As in former studies, the definition of appropriate antibiotic used in our study considers monotherapy of aminoglycoside as efficient [17]. Compared to 3GC + GM, the use of AMC + GM enabled appropriate treatment in case of infection due to Bacteroides spp. and E. faecalis. ECDC antimicrobial resistance surveillance report [18] is the only tool providing resistance rate to antibiotic associations such as cephalosporins + aminoglycosides at an international level. Nonetheless those results include both community and health-care acquired infections and do not enable an assessment of our therapeutic strategy.

The fact that diarrhea and/or vomiting at admission is associated to unfavourable outcome remains unexplained. Similar to prior findings during BUO, S. aureus is associated with unfavourable outcome [5]. The use of beta-lactam/beta-lactamase inhibitor association for empirical treatment f S. aureus bacteremia has been associated with increased mortality rate compared to oxacillin or cefazolin [19]. Nevertheless in this study we propose the systematic use of a combination with an aminoglycoside known to significantly impact the course of S. aureus bacteremia [20].

Sepsis of unknown origin refers to a clinical situation encountered in the emergency room or just after ICU admission which includes: future BUO, sepsis with a primary source which will be documented in the following hours because of changes in clinical examination or results of paraclinical exams, sepsis which will never be documented and also non-infectious aetiologies [21]. So BUO represents only a subgroup of sepsis of unknown origin defined a posteriori on the basis of bloodcultures. Then our results regarding AMC + GM cannot be entirely and directly applied to sepsis of unknown origin. Our work based on patient hospitalized in an ID department aimed to use microbiological data of BUO as a surrogate marker to guide antimicrobial treatment during sepsis of unknown origin. As all surrogate markers it presents necessarily some drawbacks but to our knowledge no other work has intended to use a different one.

When considering an empirical antibiotic treatment for community-acquired sepsis of unknown origin, microbiological features of BUO presenting half of the time with organ failure has to be integrated in the decision process. With our results, AMC + GM should be considered.