FormalPara Take-home messages

This expert statement covers the available evidence on the epidemiology, diagnosis and treatment of bloodstream infections in the ICU. Key elements are: knowledge of the local epidemiology and of the risk factors due to bacterial resistance and inadequate therapy; optimization of the antimicrobial dose and infection source control. The potential benefit of new rapid diagnostic tests, antibiotic de-escalation and short duration of antimicrobial is also discussed.

Introduction

Bloodstream infection (BSI) is defined by positive blood cultures in a patient with systemic signs of infection and may be either secondary to a documented source or primary—that is, without identified origin (https://www.cdc.gov/nhsn/pdfs/pscmanual/4psc_clabscurrent.pdf accessed December 22th 2019). Bloodstream infections (BSI) represent 40% of cases of community-acquired (CA) and hospital-acquired (HA) sepsis and septic shock and approximately 20% of the ICU-acquired cases (Table 1). It is invariably associated with poor outcomes especially when adequate antimicrobial therapy and source control are delayed [1,2,3]. This expert statement proposes key elements for early diagnosis and adequate therapy of both primary and secondary BSI (Table 2).

Table 1 Prevalence of bloodstream infections in selected recent randomized trials including adult patients with sepsis or septic shock
Full size table
Table 2 Twenty key points for the management of bloodstream infection in critically ill patients
Full size table

Epidemiological features of bloodstream infection in ICU patients

BSI may complicate the course of a myriad of severe CA infectious diseases (Fig. 1). Escherichia coli, Staphylococcus aureus, Klebsiella pneumoniae and Streptococcus pneumoniae account for more than 70% of all CA-BSI though pathogen distribution varies substantially depending on infection foci and patient characteristics [4, 5]. Of note, Pseudomonas aeruginosa causes up to 5% of community-onset BSI, essentially in patients with severe underlying conditions (e.g., immunosuppression) and/or recent healthcare exposure and suffering from urinary tract infection (UTI) or pneumonia—yet, causative strains remain usually susceptible to first-line antipseudomonal agents, with a restricted prevalence of multidrug-resistant (MDR) isolates [5, 6]. After a spectacular rise in the early 2000’s, the incidence of CA-BSI due to community-associated methicillin-resistant S. aureus (MRSA) now trends to plateau in the United States and most of other endemic regions [7]. Meanwhile, the global burden of CA-BSI due to extended-spectrum beta-lactamase-producing Enterobacterales (ESBL-PE) is amplifying steadily due to massive spread of these pathogens in the community [4, 8]. Nowadays, the prevalence of ESBL-producing isolates commonly exceeds 5% in E. coli and K. pneumoniae CA-BSI secondary to UTI or intra-abdominal infection and may reach 20% in certain geographical areas, thereby equalling the proportion reported in HA-BSI [9, 10].

Fig. 1
figure 1

Bloodstream infections in critically ill patients: main sources and leading pathogens. BSI bloodstream infection, CAP community-acquired pneumonia, HCAP healthcare-associated pneumonia, UTI urinary tract infection, ESBL-PE extended-spectrum beta-lactamase-producing Enterobacterales, SSTI skin and soft-tissue infection, HAP hospital-acquired pneumonia, VAP ventilator-associated pneumonia. Community-acquired BSI: BSI first identified [blood cultures sampling] less than 48 h following hospital admission in a patient without recent exposure to the healthcare system. Healthcare-associated BSI: community-onset BSI in patients requiring chronic haemodialysis, living in a nursing home, or recently exposed to antibiotics, in-home nursing care, or the hospital environment. Hospital-acquired BSI: BSI first identified more than 48 h after hospital admission. ICU-acquired BSI: BSI first identified more than 48 h after ICU admission. Primary BSI indicates BSI without identification of a definite source

Full size image

HA-BSI in critically ill patients are imported (i.e., documented at ICU admission) and acquired in the ICU in roughly 25% and 75% of cases, respectively [2, 5]. Overall, ICU-acquired BSI occurs during 5–7% of admissions, corresponding to an average of 6–10 episodes per 1000 patient-days [1, 3, 11,12,13,14]. Main risk factors for ICU-acquired BSI include high severity indexes at admission, prolonged stay, immunosuppression, liver disease, surgical admission, and the requirement for invasive devices or procedures [11]. In the EUROBACT-1 international study (n = 1156), ICU-acquired BSI mostly ensued from catheter-related infections (21%), nosocomial pneumonia (21%), and intra-abdominal infections (12%)—strikingly, no definite source was identified for 24% of episodes [2].

In ICUs applying current prevention bundles for the insertion and maintenance of central venous catheter (CVC), CVC-related BSI occurs in 0.5–1.5% of exposed patients, with a median incidence density ranging from 0.5 to 2.5 episodes per 1000 catheter-days [15,16,17]. Defective asepsis, a jugular or femoral insertion (versus the subclavian site) and the duration of catheterisation remain the leading risk factors for CVC-related BSI [18,19,20]. The hazard of arterial catheter-related BSI appears similar to what is observed with CVCs (that is, around 1 episode per 1000 catheter-days), with a nearly two-fold risk increase with femoral accesses when compared to the radial site [21]. Lastly, patients under extra-corporeal membrane oxygenation (ECMO) are at major risk for ICU-acquired BSI with an incidence density reaching 20 episodes per 1000 ECMO-days [22]. Most of BSIs in this particular population with extended mechanical ventilation and ICU stay are related to ventilator-associated pneumonia or other infectious foci rather than cannula infection [23].

The main pathogens responsible for HA-BSI in critically ill patients are listed in Table 3. The epidemiology of MDR pathogens widely differs from one ICU to another according to case-mix, local policies for infection control and antimicrobial stewardship, and geographical location—BSI due to non-fermenting Gram-negative bacilli such as P. aeruginosa and Acinetobacter baumannii are notably more prevalent in warm countries or during warm periods in temperate areas [24]. However, and as for other ICU-acquired infections, the incidence of BSI due to ESBL-PE, carbapenemase-producing Enterobacterales, MDR P. aeruginosa, MDR Acinetobacter baumannii, MRSA and methicillin-resistant coagulase-negative staphylococci is high and even continues to increase in most parts of the world [25]. Table 4 indicates the current resistance rates in major pathogens responsible for hospital-acquired infections—including HA-BSI—in large surveillance networks.

Table 3 Hospital-acquired bloodstream infection in ICU patients: pathogen distribution in selected multicentre studies published after 2010
Full size table
Table 4 Current resistance rates in major pathogens responsible for hospital-acquired infections according to World Health Organization regions—available data from large surveillance networks
Full size table

Early microbiological diagnosis in BSI

Culture-based methods remain the gold standard to identify the causative microorganism in sepsis, with a recommended sampling of at least two sets of aerobic and anaerobic blood cultures (10–20 mL per bottle) following rigorous skin disinfection [26]; yet the rhythm imposed by the growth time requirements of the latter is barely compatible with the ‘need for speed’ in the context of sepsis (Fig. 2). It should be kept in mind that the initiation of empirical antimicrobial therapy significantly reduces the sensitivity of blood cultures drawn shortly after treatment initiation [27].

Fig. 2
figure 2

Current workflow of microbiological diagnosis in bloodstream infection. PCR polymerase chain reaction, CfDNA cell-free DNA, AST antibiotic susceptibility testing, BC blood culture. Biochemical tests such as C-reactive protein of Procalcitonin is most of time elevated in case of BSI but not sufficiently accurate to discard the diagnosis. A significant decrease of these biomarkers should be used to shorten the duration of antimicrobial therapy

Full size image

Molecular assays are increasingly deployed in bacteriology laboratories as rapid alternatives to culture-based methods. Attempts have been made to directly detect pathogens and resistance markers by PCR on blood samples without prior incubation (Roche LightCycler®SeptiFast, SeeGene MagicPlex® Sepsis Test, Abbott Iridica); however, these tests have not met a broad success because of their medium sensitivity and specificity (Table 5) and the lack of full automation. Furthermore, these tests only seek for a limited number of antibiotic resistance genes so that the probabilistic regimen can only be adapted according to the bacterial species. More recently, a magnetic resonance-based test (T2Bacteria Panel, T2Biosystems) was made available and showed a higher sensitivity (90%) than previous methods together with a shorter turn-around time (3.5 h vs. 5–8 h) [28].

Table 5 Rapid diagnostic tools for early optimization of antimicrobial therapy in patients with bloodstream infections: methods, turn-around time and diagnostic performances
Full size table

Since then, PCR-based tests have re-focused on positive blood cultures (BC) (such as the BioFire FilmArray Blood Culture Identification and the Luminex Verigene), meaning that the test comes after a first culture-based test. The multiplex PCR (mPCR) tests applied on positive blood culture have excellent performances and have been showed to decrease the time to an optimized antibiotic regimen (spectrum narrowing or broadening or even cessation when a contaminant was identified) but not the mortality or the length of stay [29]. One major limitation of these genotypic methods is the limitation of the number of PCR probes. A negative PCR should be interpreted in view of the overall findings, possible source of infection and other available bacteriological results. Consequently, a solid expertise and strong collaboration between microbiologists and clinicians are needed [30]. Besides PCR, matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF) can also be used to identify bacteria directly from positive BC after a purification protocol (not automatized yet), with good performances for Gram-negative bacteria (> 90% concordance with subsequent culture) but not for Gram-positive bacteria (~ 80% concordance) [31]. MALDI-TOF can also be used for antibiotic susceptibility testing, with the possibility to compare spectrum after a short incubation (1–4 h) with or without antibiotics, or to directly detect peaks matching the resistance mechanisms [31]. While the proof of concept has been established, direct AST from BC using MALDI-TOF still needs standardization to enter the routine workflow.

Recently, next-generation sequencing (NGS) methods and application of machine-learning methods have showed promising results in the diagnostic of BSI. Clinical metagenomics (CMg) refers to the sequencing of the nucleic acids present in a clinical sample to identify pathogens and to infer their susceptibility to antimicrobials [32]. A variant of CMg is cell-free DNA (cfDNA) sequencing, i.e. the sequencing of extracellular cell-free DNA in clinical samples. It has been showed that the absolute concentrations of plasmatic cfDNA in patients with sepsis were elevated when compared to healthy volunteers and that the cfDNA sequences could identify potential bacterial pathogens missed by conventional, culture-based methods [33]. A recent work including 348 patients reported a 93% sensitivity but only a 63% specificity for cfDNA sequencing [34]. Indeed, metagenomic sequencing identified much more bacteria than culture, with 62/166 samples negative with conventional methods but with microorganisms found in cfDNA sequencing. Of note, cfDNA sequencing results were delivered the day after the sample arrival. Another work based on metagenomic sequencing of plasma from immunosuppressed patients found similar results, with a 95% negative predictive value [35].

Finally, while molecular tests are interesting, the performance of old-fashioned culture methods may be improved to provide more timely results. AST performed from the morning positive BC can be read at the end of the working day [36], but this would not apply to BC found positive later. In this perspective, the continuous processing of samples through lab automation could break the barriers intrinsic to the lab workflow. Similarly, the Accelerate Pheno provides an automated solution to deliver identification and a phenotypic AST within 6–8 h [37].

Choice of antimicrobial therapy

Decisions on which antimicrobials should be employed for treating bloodstream infections (BSI) in critically ill patients depend on several, overlapped factors: (1) the empirical or targeted nature of the treatment; (2) the presumed or proven origin site of the infection; (3) the suspected or proven presence of antimicrobial resistance (notably in healthcare settings with endemic MDR pathogens and/or in patients with recent exposure to antimicrobial drugs); (4) the suspected or proven presence of candidemia [25, 38,39,40,41]. Immunosuppression (e.g., neutropenia, HIV infection, or current or recent immunosuppressive therapy) should also been taken into account since immunocompromised hosts are at increasing risk of both infection with MDR bacteria—due to more frequent antimicrobial and healthcare exposure—and non-bacterial sepsis, notably resulting from invasive fungal infection [26, 42, 43].

Considering that antimicrobials are mainly used empirically in critically ill patients [44], both a reasoned administration of empirical agents on the basis of the suspected pathogen/s and efforts to pursue a rapid etiological diagnosis for allowing de-escalation are essential measures for treatment optimization [25, 45,46,47]. In this scenario, there is certainly a need for a balanced use of recently approved agents active against MDR organisms, in order not to delay the administration of an effective therapy and, on the other hand, not to accelerate the selection of further resistance using them indiscriminately [48, 49]. In addition, the availability of novel beta-lactams/beta-lactamases inhibitor (BL-BLI) combinations, which express selected activity against MDR Gram-negative bacteria expressing different determinants of resistance, has already started to change clinical reasoning at the bedside of septic patients. For example, the type of locally prevalent carbapenemases should now be taken into account when prompting empirical therapies [50]. Trying to balance all the above-reported factors, possible choices for treating BSI in critically ill patients, together with their activity against MDR pathogens and dosage recommendations, are detailed in Table 6.

Table 6 Characteristics of antibacterial drugs indicated (or used off-label in selected cases) for treating bloodstream infections (BSI) in critically ill patients
Full size table

Role of therapeutic drug monitoring (TDM)

Useful pharmacokinetic (PK) parameters for deciding antimicrobial dosages are not routinely measurable in critically ill patients. However, albeit imperfect, some practical and immediately available proxies exist that may help optimizing dosages. First, higher loading dosages of hydrophilic antimicrobials are required in critically ill patients with a positive fluid balance indicating a high volume of distribution (Vd) [51,52,53]. Second, the facts that most antimicrobials used in ICUs are excreted by the kidneys, that either augmented renal clearance (ARC) or acute kidney injury (AKI) can be present in critically ill patients with BSI, and that renal replacement therapies (RRT) are not infrequently used in these patients imply that careful attention should be devoted to the adjustment of maintenance dosages according to the fluctuations in renal function during the course of treatment [25, 52, 54,55,56]. With regard to pharmacodynamics (PD), knowledge of the different PD index of choice (T > minimum inhibitory concentration (MIC), Cmax/MIC, or area under the curve(AUC)/MIC) pertaining to the different antimicrobial classes is crucial both for selecting the most appropriate type of infusion (e.g., continuous vs. intermittent) and for measuring the impact of suspected/measured pathogens MIC on the probability of target attainment, taking into account possible variability in MIC measurements [25, 52, 57].

However, TDM remains desirable for antimicrobial treatments in critically ill patients, owing to the imperfect prediction of PK and PD in this population without measurement, even when carefully taking into account both patients chronic and acute characteristics and the expected drug behavior [52, 58, 59]. Practically, TDM appears beneficial for minimizing toxicity and/or improving clinical responses in patients treated with vancomycin or aminoglycosides [60, 61], while further evidence and standardization are needed to clearly delineate and maximize any possible clinical impact of TDM on the use of beta-lactams [25, 62, 63]. For some antimicrobial classes with inherent variable serum concentrations, technical difficulties and its frequent unavailability outside research laboratories prevent a widespread use of TDM (e.g., polymyxins, for which nonetheless TDM remains desirable whenever feasible) [58]. Detailed discussion on possible PK/PD targets (either for improving bacterial killing/clinical outcome or reducing toxicity) and sampling times for different antibiotic classes in critically ill patients undergoing TDM are available elsewhere [64, 65].

Single-drug or combination therapy for bloodstream infection in ICU patients

In an era of increasing resistance prevalence, the primary objective of an empirical combination regimen (usually a beta-lactam plus an aminoglycoside or a fluoroquinolone) is to maximize the likelihood of administering at least one drug with activity against the causative pathogen. Yet, once antimicrobial susceptibility results become available, the benefit of continuing with a dual regimen rather than a single active agent remains equivocal owing to fragmentary or conflicting evidence.

First, experimental models suggest that antimicrobial associations may synergistically prevent or postpone the selection of resistant mutants, especially in P. aeruginosa and other non-fermenting Gram-negative pathogens [66]. However, clinical data are lacking to appraise the relevance of these findings and whether combination therapy effectively protects from the emergence of resistance at the infection site is still unsettled. Interestingly, in a randomized controlled trial (RCT) including 551 patients with sepsis, receiving a meropenem–moxifloxacin combination was associated with a lower risk of persistent or subsequent infection with meropenem-resistant pathogens than meropenem alone (1.3% versus 9.1%, respectively, P = 0.04) [67]. This endpoint was unfortunately not addressed in the gut microbiota—that is, the main reservoir of MDR Gram-negative bacteria in ICU patients. Intuitively, adding a second drug to a broad-spectrum beta-lactam may amplify the ecological side-effects on commensal ecosystems and the routine use of combination therapy can probably not be justified on the sole basis of preventing resistance at the infection site.

Next, several meta-analyses failed to demonstrate that the use of a beta-lactam/aminoglycoside association reduces fatality rates in patients with BSI—including those with neutropenia or sepsis—when compared to a monotherapy with the same beta-lactam [68,69,70]. Besides, adding an aminoglycoside to a beta-lactam-based regimen has been consistently linked with an increased hazard of acute renal failure at the acute phase of infection [68, 69, 71]. On a pathogen-specific approach, there is currently no evidence that a double-active regimen impacts the outcome of patients with BSI due to methicillin-susceptible S. aureus (except in those with implanted devices) or Enterobacterales, including AmpC-hyperproducers and ESBL-PE [72,73,74,75]. Along this line, the benefit of a polymyxin-based combination has not been convincingly proven in patients infected with carbapenem-resistant A. baumannii though this strategy may perform better than polymyxin alone in patients with BSI and/or when high-dose colistin regimen are administered (i.e., ≥ 6 MUI per day) [76,77,78,79]. Controversies equally persist about the prognostic effect of combination therapy in P. aeruginosa BSI [73, 80, 81]; yet, no survival improvement was demonstrated in a meta-analysis of RCTs comparing beta-lactam plus aminoglycoside or fluoroquinolone versus beta-lactam alone in patients with this condition [82]. It should be emphasized, however, that most of available studies are relatively ancient, include a limited number of ICU patients, and display substantial heterogeneity in terms of antimicrobial administration schemes, sepsis definition, and severity indexes at BSI onset.

The benefit of combination therapy could actually be restricted to the most severely ill patients. Indeed, in a meta-regression analysis of observational studies and RCTs, combination therapy was associated with improved survival in patients with a baseline risk of death > 25% [odds ratio (OR) for death 0.51, 95% CI 0.41–0.64], while a harmful effect was strikingly observed in less severe patients (OR 1.53, 95% CI 1.16–2.03) [83], putatively due to toxic and/or ecological adverse events. Similar results were reported in a cohort of 437 patients with BSI due to carbapenemase-producing Enterobacterales (mostly KPC-producing K. pneumoniae), with a survival benefit of combination therapy in those with a high probability of death (OR 0.56, 95% CI 0.34–0.91) but not in the lower mortality stratum [84]. Pending for confirmatory studies to definitely solve this essential issue [85], combination therapy remains recommended in patients with septic shock but should not be routinely prescribed for the definite treatment of those with other severe infections, including sepsis without circulatory failure [26].

Early appropriate source control

The search for the source of BSI (that is, secondary BSI) should be guided by the patient clinical presentation. The most common conditions that may require a specific approach for source control are obstructive UTI, skin and soft-tissue infections and intra-abdominal infections for secondary CA-BSI, and vascular device and surgical site infection for secondary HA-BSI. Source control to eliminate infectious foci follows principles of damage control and should be limited to drainage, debridement, device removal and compartment decompression in case of hemodynamic instability and organ failures [86].

An important body of the literature argues for a systematic catheter removal in case of catheter-related BSI in critically ill patients [87,88,89]. However, the device is actually the source of sepsis in less than half of those with a suspected catheter-related infection [90]. The systematic removal should thus be balanced with a more conservative attitude in the absence of septic shock but remains the rule in case of septic shock, immune suppression, or persistent bacteremia under appropriate antimicrobial therapy.

For BSIs related to surgical site infection, source control is a major determinant of outcome [91]. However, the optimal delay for source control is debated—from less than 6 to less than 12 h [91,92,93,94]. Indeed, the cost–benefit ratio of an immediate drainage in unstabilized patients or a hemodynamic and physiological stabilization-first and secondary source control is a matter of debate. Immediate damage control using less risky, minimally invasive surgical debridement and/or percutaneous drainage (delaying the need for definitive surgery until the patient is stabilized) is probably the best option [95]. It should be discussed with anesthetists and surgeons on an individual basis.

Key elements of surgical source control include drainage, debridement, cleansing, irrigation, and control of the source of contamination [95]. Yet, the quality of source control is difficult to evaluate [96], somewhat subjective and depends on the surgeon’s perception. A standardized reporting file of the operative procedure may help. A closed collaboration between surgeons and intensivists during the patients’ follow-up is, therefore, fundamental.

De-escalation strategy

Antimicrobial de-escalation (ADE) is a component of antimicrobial stewardship strategies (AMS) aiming at both reducing the spectrum of antibiotic therapy and decreasing the emergence of antimicrobial resistance [97].

ADE is variably defined, and usually includes narrowing the spectrum of a pivotal antibiotic, often a β-lactam, and/or decreasing the number of agents [98, 99]. Those components should be scrutinized separately but in general ADE is part of the re-evaluation of the antimicrobial regimen that happens 2–3 days after the infection was diagnosed when results of microbiological specimens become available. BSI is very particular as it is the only kind of infection where the pathogen is always known (by definition), and as such a perfect candidate for this re-evaluation.

Where the source and the pathogen of the BSI are known, it can be safely recommended, even in immunocompromised patients [99, 100], to stop companion antibiotics intended to broaden the spectrum of therapy. Indeed, for a Gram-negative BSI, anti-MRSA or anti-fungal agents should not be continued longer than it is needed to know those are not in cause.

The case of the pivotal antibiotic is more complex because of multiple factors:

  • While resistance to carbapenems may increase after extended courses, a lot of the harm has already been done after 1–3 days of therapy [101].

  • Ranking the spectrum of antibiotics is complex and yields variable results depending on the method, the region of the world and the priorities that are considered [98, 102, 103].

  • Whether narrowing the clinical antimicrobial spectrum decreases the emergence of resistance has not been adequately studied, and while it has some rationale, in some cases, the opposite may be true [104].

  • Some ADE regimens such as switching a beta-lactam for a fluoroquinolone may be useful in the ward to allow for oral treatment and discharge from hospital. This potential benefit does not exist for ICU patients, and those regimens are likely to cause additional emergence of resistance.

  • Caution is important for sources that are frequently polymicrobial such as intra-abdominal infections. A positive BC may yield a single pathogen, while specimens from the source may identify multiple pathogens. Furthermore, there may be other important pathogens that were not cultured and require the broad-spectrum antimicrobial that was initially started.

  • In silico PK/PD modeling has warned that with conventional dosing strategies narrower spectrum beta-lactams may have higher risks of non-target attainment than their broad-spectrum alternatives [105].

  • Some narrower spectrum alternatives are more effective than broad-spectrum regimens. For instance, oxacillin or cephazolin are superior to piperacillin/tazobactam in S. aureus BSI [106].

  • Risk exists that ADE may cause an increase in the total duration of antimicrobial therapy [107]. Multiple studies on different sources of infection lead to recommend shorter rather than longer duration of antimicrobial therapy and this may be a more important target than changing molecules within a treatment [108, 109].

  • In ADE, it is particularly important to not increase the duration of treatment because of days where treatments overlap. We recommend that a stop date and time is calculated from the time of effective treatment and that this is maintained for the treatment after ADE.

We recommend consideration is given to all or a combination of those reasons before deciding if narrowing the pivotal antimicrobial is the appropriate course of action in critically ill patients with a BSI. ADE should be an integral part of the global AMS strategy, targeting the optimization of the treatment of patients with an infection. ADE is part of the clinical and microbiological re-evaluation that should happen for every patient with a BSI every time the laboratory provides new information. Those time points include the alert for positivity and Gram stain, the speciation and sensitivities of the pathogen.

Duration of therapy

Sufficient duration of antimicrobial therapy is required to prevent clinical failure and relapse. It should, however, not exceed what is required to achieve that target as longer courses may cause adverse events, toxicity, emergence of antimicrobial resistance, increase costs and resource use.

Duration of therapy should be defined as starting on the first day after an adequate antimicrobial is administered, and the source has been treated, and the blood cultures have become negative. To define clearance of the bacteremia, we require at least one set of negative blood cultures obtained 2–4 days after the infection [110]. Sampling more than one follow-up sets of blood cultures is preferable to avoid the skip phenomenon. This was described for S. aureus as when persisting bacteremia may be missed if insufficient follow-up cultures are performed [111].

From a recently published cohort study of 1202 ICU patients with BSI, we learn that current practice consists in extended duration of treatment for those patients with a median of 14 days (IQR, 9–17.5 days) [112]. After proper adjustment and excluding early deaths, duration of treatment showed no association with either survival or bacteremia relapse [15]. Most importantly, data extrapolated from observational studies on duration of treatment should be analyzed with extreme caution in populations with an inherently high risk of death. In survivor bias, the patients who have died early have had less days alive where the treatment could be given, hence received a shorter course of treatment. This artificially increases the risk of death associated with shorter courses and may have led clinicians to favor unnecessarily longer treatments.

A multicenter RCT involving 3598 ICU patients with BSI to 7 vs 14 days of antibiotics is ongoing (planned enrollment of 3598 patients). Results will not be available until 2022 (Balance trial-NCT03005145) and, until then, we will need to rely on a lower quality of evidence from uncontrolled or underpowered randomized trials that are described below.

The safety of a shorter antibiotic therapy for Gram-negative uncomplicated BSI was recently shown in a RCT including 604 patients across 3 centers. The authors enrolled hemodynamically stable patients without fever for at least 48 h at day 7 after the BSI onset They established non-inferiority of 7 against 14 days of treatment for a primary composite outcome of mortality, clinical failure, readmissions and extended hospitalization at day 90 [113]. The validity of these results in Pseudomonas aeruginosa BSI and in population with higher severity or prevalence of immunosuppression was suggested in a multicenter cohort of 249 patients included from 5 hospitals [114]. They used a causal inference model with adjustment on the inverse probability of treatment weighting. The composite outcome of recurrent P. aeruginosa infection at any site or mortality at 30 days was similar in both groups (OR 1.06; 95% CI 0.42–2.68; p = 0.91). Recurrent infections occurred in 7% of the short course and 11% of the prolonged course groups, thereby invalidating the reasoning to continue antibiotics beyond the recommended duration to prevent relapse [114].

This is in line with the findings of a meta-analysis of short versus long antibiotic treatments in patients with bacteremia caused by the most common infectious syndromes [108]. Only one trial conducted in children with acute nephronia—that is an intermediate stage between acute pyelonephritis and renal abscess—favored longer compared to shorter antibiotic courses [115]. For other trials and in pooled results, there was no difference in survival, clinical or microbiological cure with owing to treatment duration.

Trials investigating the infectious syndromes causing BSI in the ICU are important as in most cases, it may be the source that should guide our treatment rather than its consequence (the bacteremia). Short treatment should safely be used for ventilator-associated pneumonia (VAP) [116], or post-operative intra-abdominal infections provided source control was optimal [109].

When judging of duration of antibiotics, there is this second time point at 5–7 days. The decision of escalation/de-escalation/no change or dose adjustments should be taken after 2–3 days when microbiological specimens became available [98]. The effectiveness of therapy should be judged after 1 week of treatment on clinical and microbiological resolution of the infection. This will include defervescence and resolution of organ failures and shock, negative follow-up cultures, the absence of endocarditis or metastatic sites of infection and no implanted prosthesis which are all required to define an uncomplicated infection [110]. Problems with source control and/or superinfections at the source will also uncover around that time point. If those are resolved and the pathogen or the source is not specifically requiring extended treatments antibiotic regimen can be safely stopped.

For some pathogens, such as S. aureus or in cases of uncomplicated candidemia, treatment should be continued for 14 days after the first negative blood culture [110, 117]. Some particular source of infections where the pathogen is quiescent or living in biofilms, the presence of an untreatable source, septic metastasis or micro-abscesses also require prolonged therapy. Trans-thoracic/transoesophagal echocardiograph and funduscopy should be performed before deciding the duration of therapy. Indeed, short-term therapy (15 days) was shown to be effective only in selected cases of uncomplicated S. aureus right-sided infectious endocarditis or left-sided native valve infectious endocarditis due to highly susceptible streptococci. Most current recommendations emphasize prolonged antibiotic administration (4–6 weeks or even 8 weeks) for S. aureus prosthetic valve endocarditis. Valve cultures should be taken into account to decide how long to continue antimicrobial therapy after valve replacement [118]. Longer therapies are also needed for bone and joint infections (4–8 weeks), brain abscesses (8 weeks), empyema (4–6 weeks) or, in general, when the source control is impossible or incomplete. It is especially the case when infected devices or prosthesis are left in place.

The major limitation of systematically shortening the duration of therapy in uncomplicated infection is the lack of controlled trials confirming its safety. Importantly, the stabilization of the infection process may be difficult to define and often subjective. The use of individualized follow-up of procalcitonin (PCT) levels might be helpful in certain CA infections [119]—indeed, available data suggest that a PCT-driven reduction of treatment duration in patients with otherwise improving clinical status does not result in increased mortality, including in those with BSI [120, 121]. In case of incomplete source control, the duration of therapy might be guided by repeated morphological exams such as leucocyte scintigraphy, and PET scans.

The case of ongoing instability or clinical worsening is complex and may be due to multiple different causes. Often combined, interconnected and leading to diagnostic dilemma with a very high risk of death. Failure of treatment at the source, superinfection with a different or the same pathogen that has become resistant to the ongoing antimicrobial therapy, residual infected tissues or material at the source or at other sites via hematogenous dissemination will all require new specimens, possibly new percutaneous or surgical control, optimization and sometimes escalation of antibiotics. The duration will need to be recalculated from that point in time. Furthermore, it is always important to suspect infections related to the high intensity of care, such as VAP, CLABSI, or a CAUTI. Peripheral blood cultures, specimens of each clinically suspected source and changing the CVC, arterial line and any indwelling material with sending catheter tips for microbiology are most often necessary as part of the treatment and diagnostic workup. While in patients without shock, there are hints to a benefit for waiting for results of such investigations [122]. In cases with high severity, worsening shock and the risk of an untreated infection, it is often required to escalate the antimicrobial regimen in the meantime. The new regimen should take in account colonization and the most frequent pathogens and resistance patterns according to the source and patient category and always be preceded by new blood cultures and specimens of any potential source.

It is, however, important to always remember that non-infectious causes of fever may complicate the clinical picture, most prevalently drug reactions or antibiotic related fever and venous thromboembolism [123]. It is only with meticulous review of the history, available microbiology, clinical examination and targeted investigations that the decision can be taken to escalate the spectrum, extend the duration or sometimes stopping the antimicrobials altogether to allow for an effective microbiological workup.

Concluding remarks

Community-acquired and healthcare-associated BSIs are common situations to manage in ICU patients and are associated with impaired outcomes, especially in case of sepsis/septic shock, immune deficiency, and delayed adequate antimicrobial therapy and/or source control. The prevalence of MDR pathogens is high or even increasing in healthcare-associated BSI, thereby strengthening the need for prospective clinical evaluation of novel diagnostic tools that enable earlier identification of resistance markers. Pending for such data, the choice of empirical regimen depends on a variety of clinical parameters, with the patient’s individual risk of MDR pathogen being the leading one. Double-active regimen might improve survival in the most severely ill patients yet further studies should be focused on this essential issue. Antimicrobial de-escalation should be considered once culture results become available and the source of BSI is identified and controlled. Treatment duration longer than 5–8 days may be indicated only in certain clinical scenarios and/or in BSI due to particular pathogens such as S. aureus.