Antibiotic de-escalation for bloodstream infections and pneumonia: systematic review and meta-analysis

Clinical Microbiology and Infection 22 (2016) 960e967

Contents lists available at ScienceDirect

Clinical Microbiology and Infection journal homepage: www.clinicalmicrobiologyandinfection.com

Review

Antibiotic de-escalation for bloodstream infections and pneumonia: systematic review and meta-analysis M. Paul 1, *, 3, Y. Dickstein 1, 3, A. Raz-Pasteur 1, 2 1)

Infectious Diseases Institute, Rambam Health Care Campus and The Ruth and Bruce Rappaport Faculty of Medicine, Technion, Israel Institute of Technology, Haifa, Israel 2) Medicine A, Rambam Health Care Campus, Haifa, Israel

a r t i c l e i n f o

a b s t r a c t

Article history: Received 7 March 2016 Received in revised form 23 May 2016 Accepted 24 May 2016 Available online 6 June 2016

Antibiotic de-escalation is an appealing strategy in antibiotic stewardship programmes. We aimed to assess its safety and effects using a systematic review and meta-analysis. We included randomized controlled trials (RCTs) and observational studies assessing adults with bacteraemia, microbiologically documented pneumonia or severe sepsis, comparing between antibiotic de-escalation and no deescalation. De-escalation was defined as changing an initially covering antibiotic regimen to a narrower spectrum regimen based on antibiotic susceptibility testing results within 96 hours. The primary outcome was 30-day all-cause mortality. A search of published articles and conference proceedings was last updated in September 2015. Crude and adjusted ORs with 95% CI were pooled in random-effects meta-analyses. Sixteen observational studies and three RCTs were included. Risk of bias related to confounding was high in the observational studies. De-escalation was associated with fewer deaths in the unadjusted analysis (OR 0.53, 95% CI 0.39e0.73), 19 studies, moderate heterogeneity. In the adjusted analysis there was no significant difference in mortality (adjusted OR 0.83, 95% CI 0.59e1.16), 11 studies, moderate heterogeneity and the RCTs showed non-significant increased mortality with de-escalation (OR 1.73, 95% 0.97e3.06), three trials, no heterogeneity. There was a significant unadjusted association between de-escalation and survival in bacteraemia/severe sepsis (OR 0.45, 95% CI 0.30e0.67) and ventilator-associated pneumonia (OR 0.49, 95% CI 0.26e0.95), but not with other pneumonia (OR 0.97, 95% CI 0.45e2.12). Only two studies reported on the emergence of resistance with inconsistent findings. Observational studies suggest lower mortality with antibiotic susceptibility testing-based de-escalation for bacteraemia, severe sepsis and ventilator-associated pneumonia that was not demonstrated in RCTs. M. Paul, CMI 2016;22:960 © 2016 European Society of Clinical Microbiology and Infectious Diseases. Published by Elsevier Ltd. All rights reserved.

Editor: A. Huttner Keywords: Antibiotic treatment Bias De-escalation Empirical antibiotic treatment Meta-analysis Observational study Randomized controlled trial Susceptibility testing Systematic review

Background Bloodstream infections and pneumonia requiring hospitalization are responsible for significant morbidity and mortality, with mortality rates ranging between 27% and 54% [1,2]. Numerous studies have pointed to the importance of appropriate empiric antibiotics in reducing mortality for severe infections [3,4]. This had led to the widespread use of broad-spectrum drugs as first-line treatment, potentially contributing to the increase in bacterial resistance to antibiotics since, without intervening, empirical

* Corresponding author. M. Paul, Director, Infectious Diseases Institute, Rambam Health Care Campus, Haifa, 3109601, Israel. E-mail address: [email protected] (M. Paul). 3 Authors M. Paul and Y. Dickstein contributed equally to the study.

therapy is frequently continued. Several strategies have evolved to limit the appearance and spread of such organisms, among them antibiotic de-escalation [5,6]. De-escalation (also termed streamlining) refers to tailoring of empirical antibiotic treatment to the susceptibilities of the bacteria isolated, selecting the narrowest spectrum antibiotic. It can follow any empirical treatment, but is also applied with a policy of initial broad-spectrum treatment mainly in intensive-care units (ICUs). Deescalation might be easier to implement in antibiotic stewardship programmes than interventions targeting empirical antibiotics [5,7]. More information is available at the latter time-point, the patient's course is known and there is time for discussion and consideration. Promoting de-escalation entails increasing awareness among all antibiotic prescribers and education regarding antibiotic hierarchy or local preferences for targeted antibiotic treatment.

http://dx.doi.org/10.1016/j.cmi.2016.05.023 1198-743X/© 2016 European Society of Clinical Microbiology and Infectious Diseases. Published by Elsevier Ltd. All rights reserved.

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

A Cochrane review published in 2013 found insufficient evidence to recommend for or against de-escalation in adults with sepsis after a review of the literature failed to reveal randomized controlled trials (RCTs) testing the intervention [8]. RCTs constitute the reference standard design to assess such an intervention. However, as de-escalation has been appraised in many observational studies, systematically reviewing them and appraising the risk of bias in observational studies might prove useful to guide practice and further research. We aimed to assess the outcome of de-escalation therapy in patients with bloodstream infections, severe sepsis and pneumonia. Materials and methods We included RCTs and prospective or retrospective observational studies, conducted in non-ICU and ICU settings. Patients, interventions, comparisons and outcomes are summarized in the Supplementary material (Table S1). We included adults 18 years of age and older with pneumonia, bacteraemia and severe sepsis/ septic shock with microbiologically documented infections, who received appropriate empirical antibiotic treatment. Bloodstream infections had to be defined as clinically significant using valid definitions to exclude contaminants and pneumonia had to be defined using valid clinical and microbiological definitions [9]. The studies had to compare de-escalation therapy versus continued empiric antibiotic therapy. De-escalation was defined as changing an initially appropriate (covering) antimicrobial therapy to a narrower spectrum regimen based on culture results within 96 hours. A narrower spectrum regimen was defined as downgrading from a broad spectrum to a narrower spectrum agent within the same antibiotic class, changing a broad-spectrum antibiotic to a narrower-spectrum antibiotic of a different class (e.g. vancomycin to oxacillin), or discontinuation of one or more drugs of a combined regimen. Downgrading antibiotics from a broad to a narrow spectrum necessitates a hierarchy of antibiotics; we documented whether a hierarchy was used and accepted the study definitions for de-escalation and antibiotic hierarchy, as long as defined by antibiotic susceptibility testing and compatible with our definitions. We separated between studies in which empiric antibiotic treatment was intentionally broad-spectrum and those that did not specifically direct the empirical regimen. The primary outcome assessed was all-cause mortality at 30 days. If not reported, we used all-cause mortality at the end of study follow up. Secondary outcomes included clinical failure, as defined in the study, examined at the end of treatment; duration of hospital and ICU stay; duration of antibiotic treatment; resistance development and superinfections, defined as secondary clinically significant infections developing within a 30-day follow up. Antibiotic resistance development was assessed as isolation of bacteria resistant to the antibiotics given to the patient in clinical or surveillance samples; and as isolation of MDR bacteria of epidemiological significance that were not present initially, including: extended-spectrum b-lactamase-producing bacteria, methicillinresistant Staphylococcus aureus, vancomycin-resistant Enterobacteriaceae and carbapenem-resistant Enterobacteriaceae. We assessed adverse effects of antibiotic therapy, including nausea or vomiting, rash, antibiotic-associated diarrhoea, renal failure and hepatotoxicity and Clostridium-difficile infection. We conducted a broad search for randomized and observational studies in PubMed, The Cochrane Library from inception until September 2015, and conference proceedings for the last 3 years of the ECCMID and ICAAC. In addition, we examined the bibliographies of identified trials as well as previous systematic reviews. No restrictions on language, date of publication or publication status were applied. We tailored the following search string by database:

961

(blood stream infection OR bloodstream infection OR bacteremia OR sepsis OR septic shock OR pneumonia) AND (de-escalation OR de-escalate OR streamlining OR streamline OR targeted OR targeting OR narrowing OR narrow) AND (antibiotic OR antibiotics). We applied the Cochrane Risk of Bias Assessment Tool for NonRandomized Studies of Interventions (ACROBAT-NRSI, http://bmg. cochrane.org/cochrane-risk-bias-assessment-tool-nonrandomized-studies-interventions-acrobat-nrsi) to both observational studies and RCTs, addressing the outcome of all-cause mortality. The tool includes signalling questions in seven domains of bias, to which the responses are yes, probably yes, probably no and no. RCTs could achieve low risk of bias for all domains, depending on randomization methods. Low risk of bias in an observational study implies that it is comparable to a well-performed RCT. We tailored the tool to our review (see Supplementary material, Appendix S1). Two reviewers independently applied inclusion/ exclusion criteria and extracted all data. Data were compared and differences were resolved by discussion. We extracted crude mortality rates and adjusted effect estimates in observational studies. For continuous outcomes we computed means and standard deviations from the data reported in the study using methods specified in the Cochrane Handbook and Wan et al. [10,11]. Adjusted risk ratios were converted to odds ratios and the hazard ratio reported in one trial was assumed to represent the risk ratio following the proportional hazards assumption. We compiled crude ORs or absolute mean differences using a random effects meta-analysis and adjusted ORs (or ORs computed from RCTs) using an inverse variable random effects meta-analysis. Heterogeneity was assessed using a chi-square test (p <0.1) and the I2 test (>50%). Analyses were conducted in REVIEW MANAGER 5.3 [12]. Results Our literature review identified 558 potential articles for evaluation. Of these, 19 fulfilled review eligibility criteria and were included in the analysis [13e31] (Fig. 1). One study contributed to two analyses of different, non-overlapping periods in the study [24]. The studies comprised a total of 3973 patients, all adults, with a study mean or median age between 51 and 71 years. Eight studies enrolled patients with bacteria, two addressed severe sepsis or septic shock in the ICU and 14 studies enrolled patients with pneumonia (Table 1). Four studies included community-acquired infections only [14,17,25,29], eight included exclusively hospitalacquired infections [13,19,22,23,27,28,30,31] and six studies included both [15,18,20,21,24,26]. De-escalation was performed according to MDI in all of the included studies and was defined variably (see Supplementary material, Table S2): ten defined deescalation by narrowing the spectrum and 13 by stopping one or more drugs of a combination. In five studies on hospital-acquired pneumonia de-escalation was performed from a broad-spectrum empirical regimen [13,19,22,30,31]. All-cause mortality was reported in all studies. The majority (11) reported 28- to 30-day allcause mortality, whereas the others reported on in-hospital, inICU, 90 days [26] or in relation to completion of antibiotic treatment [13]. Risk of bias assessment Three trials were randomized [17,22,26], whereas the remainder were observational studies. The RCTs did not score low risk of bias for all items, one trial failing to match groups for important confounders [26] and two open-label trials did not report on cointerventions [17,26] (Fig. 2). Overall, 11/19 studies reported on pre-defined important confounders (age, renal function baseline

962

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

Records idenƟfied through database searching (n = 525)

AddiƟonal records idenƟfied through other sources (n = 33)

Records aŌer duplicates removed (n = 549)

Records screened (n = 549)

Records excluded (n = 519)

Full-text arƟcles assessed for eligibility (n = 30)

Studies included in qualitaƟve synthesis (n = 19)

Studies included in quanƟtaƟve synthesis (n = 19)

Full-text arƟcles excluded (n = 11) Not all paƟents with microbiological documentaƟon/ deescalaƟon not based on AST [1–6] No comparator group [7,8] Review [9,10] No clinical data on deescalaƟon [11]

Fig. 1. Study flow; references refer to supplementary Appendix S2.

and acute, haemodynamic status, respiratory status, functional status, cardiac disease, see Supplementary material, Appendix S1) and performed an adjusted analysis. The mean severity-of-illness scores (e.g. APACHE II) were well-matched between the study groups in 13 of the studies [13e15,18e20,22,25,27,29,30]. However, items scoring at high risk in most studies included confounder assessment, matching for confounders and co-interventions (Fig. 2). Primary outcome: mortality Overall, the pooled OR for mortality with de-escalation of all studies using crude, unadjusted, results from the observational studies was 0.53 (95% CI 0.39e0.73), OR <1 favouring de-escalation (Fig. 3). The difference was statistically significant in studies of bacteraemic patients (OR 0.45, 95% CI 0.30e0.67, ten studies) and non-significant for pneumonia (OR 0.64, 95% CI 0.39e1.06, nine studies). Both subgroups and the overall analysis had moderate heterogeneity (I2 47e69%). Heterogeneity in the pneumonia analysis could be partially explained by the type of pneumonia: for ventilator-associated pneumonia there was a significant advantage to de-escalation (OR 0.49, 95% CI 0.26e0.95), whereas in studies including mostly non-ventilated pneumonia there was no significant difference between groups (OR 0.97, 95% CI 0.45e2.12; see Supplementary material, Fig. S1), with no differences observed when empirical treatment was broad-spectrum or non-directed. However, heterogeneity remained also in the stratified pneumonia analysis. ORs were similar for studies in which the mortality rate was above the median (mean 32%) and in those with lower mortality (mean 16%).

The adjusted analysis showed a smaller association between deescalation and survival that was no longer statistical significant (adjusted OR 0.83, 95% CI 0.59e1.16) and in the subgroup of four studies assessing patients with pneumonia there was no difference (adjusted OR 1.17, 95% CI 0.73e1.86), without significant heterogeneity (Fig. 4). Pooling of the three RCTs, assessing different types of infection (community-acquired pneumonia, hospital-acquired pneumonia and severe sepsis in ICU) and different de-escalation definitions showed increased mortality with de-escalation without statistical significance, OR 1.73, 95% CI 0.97e3.06), without heterogeneity (I2 ¼ 0%). Overall, the large association between de-escalation and survival observed crudely was smaller in adjusted analyses of these observational studies and non-existent in RCTs (Fig. 5). Secondary outcomes Clinical failure was variably defined and reported in only six studies [13,18e20,22,24]. Significantly fewer patients with bacteraemia experienced treatment failure with de-escalation (unadjusted OR 0.44, 95% CI 0.26e0.82, three studies), without heterogeneity. For pneumonia there was no difference between groups (OR 1.30, 95% CI 0.76e2.23, three studies), without heterogeneity. Only one of the studies in the pneumonia analysis was randomized and the others did not report an adjusted analysis. There were no significant differences between groups for hospital or ICU length of stay, whether only patients discharged alive were considered or when all patients were analysed. The duration of

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

963

Table 1 Characteristics of included studies Study ID

Location

Study years N patients Type of infectiona Infection definitionsb

Alvarez-Lerma 2006 [13]

Spain

2000e2001 213

HAP

Carugati 2015 [14]

multicenter, 35 countries

2005

CAP with BSI

Cremers 2014 [15]

The Netherlands 2001e2011 275

BSI

Eachampati 2009 [16] Falguera 2010 [17] GarnachoMontero 2014 [18] Giantsou 2007 [31] Joffe 2008 [19]

USA Spain Spain

2005e2007 134 2006e2008 177 2008e2012 628

Greece USA

NS 143 2000e2005 412

VAP CAP Severe sepsis/ septic shock, any VAP VAP

Khasawneh pneumonia 2014 [20]

USA

2008

60

Pneumonia with BSI

Khasawneh UTI 2014 [21] USA

2008

65

UTI with BSI

Kim 2012 [22]

Korea

2004e2006 108

HAP

Kollef 2006 [23]

USA

2003e2004 394

VAP

Koupetori 2014 [24]

Greece

2006e2013 94 þ 129

BSI

Lee 2015 [25]

Taiwan

2005e2012 189

BSI

Leone 2014 [26]

France

2012e2013 116

Rello 2004 [27]

Spain

2000e2001 331

Severe sepsis, any VAP

Shime 2011 [29]

Japan

2004e2009 201

BSI

Shime 2013 [28]

Japan

2006e2011 49

BSI

Soo Hoo 2005 [30]

USA

1998e2002 106

HAP

261

Place of acquisition

CDC criteria with positive BAL, Hospital TA or PBS or Legionella antigen New infiltrate on CXR with Community clinical criteria and bacteraemia with plausible pathogen Blood-culture-proven pneumococcal infection Hospital/ Community Clinical suspicion of pneumonia-positive BAL Hospital Infiltrate on CXR with clinical criteria Community Established criteria for severe Hospital/ sepsis and septic shock Community Physician diagnosis with BAL or CXR Hospital New or persistent infiltrate on CXR Hospital with clinical criteria Hospital/ Same pathogen in blood and respiratory Community sample and pneumonia (non-defined) documented in medical chart Same pathogen isolated in Hospital/ blood and urine cultures Community New and persistent infiltrate on Hospital CXR with clinical criteria New and persistent infiltrate on Hospital CXR with clinical criteria BSI with at least two SIRS criteria Hospital/ Community community-onset monomicrobial Community Enterobacteriaceae BSI Established definitions for severe sepsis Hospital/ Community Hospital New, persistent infiltrate on CXR with purulent respiratory secretions and clinical criteria Monomicrobial bacteraemia caused by Community antibiotic-susceptible bacteria Hospital Monomicrobial bacteraemia caused by typically MDR Gram-negative bacteria Hospital New, persistent infiltrate on CXR with purulent respiratory secretions and clinical criteria

ICU or non-ICU settingc ICU Both

Both ICU Non-ICU ICU ICU ICU Both

Both ICU ICU Both Both ICU ICU

Both Both

Both

a Types of infection: BSI, bloodstream infection; CAP, community-acquired pneumonia; HAP, hospital-acquired pneumonia; UTI, urinary tract infection; VAP, ventilatorassociated pneumonia. b BAL, bronchoalveolar lavage; BSI, bloodstream infection; CDC, Centers for Disease Control; CSR, chest X-ray; MDR, multidrug-resistant bacteria; PBS, protected brush specimen; SIRS, sepsis inflammatory response syndrome; TA, tracheal aspirate. c The studies' settings: ICU, intensive care unit.

antibiotic treatment was significantly longer with de-escalation in studies of bacteraemia, with a mean difference of 3.06 days (95% CI 1.98e4.14), four studies, with no heterogeneity (I2 ¼ 0%). However, survival in the de-escalation group was higher, as shown above, and treatment duration reported in the studies was not adjusted for survival. With pneumonia, there was no difference in antibiotic treatment duration (four studies). Only two studies compared the emergence of resistance. One reported that the emergence of extended-spectrum b-lactamaseproducing bacteria occurred less often in the de-escalation group than in the non-de-escalation group (1/86 (1.16%) versus 9/103 (8.7%)) [25], whereas the other showed no difference between groups in the detection of various multi-drug resistant bacteria (methicillin-resistant S. aureus, carbapenem-resistant Gram-negatives, extended-spectrum b-lactamase-producing bacteria and Stenotrophomonas maltophilia [22]. Superinfections were assessed in two studies of ventilator-associated pneumonia in the ICU [19,26] with no significant pooled difference between groups. A single RCT reported on any adverse effect and found a rate of 8/88 (9%) in the de-escalation group versus 16/89 (18%) in the non-de-

escalation group [17] and a single RCT reported on C. difficileassociated diarrhea with no events in either group [26]. Discussion In a systematic review of studies examining the strategy of antibiotic susceptibility testing-guided antibiotic de-escalation from covering empirical treatment for bacteraemia, pneumonia or severe sepsis/septic shock, we found major differences in results for observational studies and RCTs. Observational studies reported a significant and large unadjusted association between de-escalation and lower mortality (OR 0.53, 95% CI 0.39e0.73) and most claimed this as their conclusion. The adjusted effect estimate from observational studies was smaller and non-existent when the analysis was limited to three RCTs that showed an opposite trend (OR 1.73, 95% CI 0.97e3.06). It is likely that the crude lower mortality with de-escalation in observational studies is due to bias. Physicians were probably more likely to change to a narrow-spectrum antibiotic in patients who had improved at the time when culture results became available. It is difficult to measure and adjust for this

Co-interventions PN

PN

PN

DN

PN

PY

DY

DY

PN

PY

DY

DY

DY

PY

DY

DN

DN

DY

PY

DN

DY

PY

DY

DY

DY

DN

PN

DN

DN

PN

DY

PN

DY

PN

DN

DN

DN

DY

PN

DY

DY

DY

PY

DY

PY

DY

Prospective

PN

DY

DN

DN

DN

DN

PN

Prospective

DY

PY

PY

PY

DY

DY

PN

Lee 2015

Retrospective

DY

PN

PY

DN

PY

DY

PN

Leone 2014

RCT

DY

DY

DN

DN

DY

PY

PN

Rello 2004

Prospective

DY

PY

DN

DN

DN

PN

PN

Shime 2011

Retrospective

DY

PN

DN

DN

DN

DY

PN

Shime 2013

Retrospective

DY

PN

DN

DN

PY

PY

PN

Soo Hoo 2005

Retrospective

DN

PN

PY

DN

PY

DY

DN

Study

Study design

Alvarez-Lerma 2006

Prospective

DY

DY

DY

DN

Carugati 2015

Retrospective

DY

PN

PY

Cremers 2014

Retrospective

DY

PN

PY

Eachampati 2009

Retrospective

DY

DN

PY

Falguera 2010

RCT

DY

DY

PY

Garnacho-Montero 2014

Prospective

DY

PY

Giantsou 2007

Prospective

DY

Joffe 2008

Retrospective

DY

Retrospective

DY

Khasawneh UTI 2014

Retrospective

Kim 2012

RCT

Kollef 2006 Koupetori 2014

Outcome

PN

DY

assessment

DN

DN

Confounder

PY

PN

matching

PN

Confounder

PN

assessment

PY

Exposure

DN

Cohort selection

Missing data

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

assessment

964

Khasawneh pneumonia 2014

DY

Definitely yes (low risk of bias)

PY

Probably yes

PN

Probably no

DN

Definitely no (high risk of bias)

Fig. 2. Risk of bias in included studies; RCT, randomized controlled trial; Prospective/retrospective, refers to observational cohort studies.

trend in clinical status to which the treating physicians react. RCTs, despite risk for confounding in one of the studies, are profoundly different from the observational studies in that patient selection is not based on the initial response to treatment or source control, and a protocol-guided intervention is applied. RCTs are the only appropriate design to assess the effect of de-escalation on survival. We do not actually expect a mortality effect with de-escalation. It should be enough to show that de-escalation is not associated with adverse consequences (mortality or clinical failure) to adopt it. Our main expectation from de-escalation is to limit the selection and development of bacteria resistant to broad-spectrum antibiotics in the individual treated. Unfortunately, only 2/19 of the studies reported on isolation of resistant bacteria with versus without de-escalation and none performed active surveillance for resistance assessment. Hence, we do not have evidence showing that de-escalation actually results in less resistance. The effects of de-escalation on resistance in the unit, ward or hospital and in other patients can be demonstrated only in studies comparing the

different strategies at the unit, ward or hospital level; such studies were not identified. This could be achieved in cluster-randomized trials, cross-over and beforeeafter studies or other comparative studies designed at the cluster level. The large clinical heterogeneity in the patients assessed and interventions applied in the RCTs included in our review does not allow strong confidence in the effect estimate of their metaanalysis suggesting higher mortality with de-escalation. De-escalation could result in higher mortality if the isolated bacteria are not the actual, or only, bacteria causing the infection. This is more likely to happen with pneumonia where respiratory samples might not represent well the aetiology of the pneumonia, and the diagnosis of ventilator-associated pneumonia as the source of infection is not totally specific [32,33]. Indeed, in the observational studies, the association between de-escalation and survival observed among patients with bacteraemia was not observed for patients with pneumonia. Another possibility is that the narrow-spectrum antibiotics were older and less effective than the empirical, newer

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

965

Fig. 3. Unadjusted analysis for all-cause mortality for de-escalation (DE) versus non-de-escalation (non-DE) antibiotic strategies.

Fig. 4. Adjusted analysis for all-cause mortality for de-escalation (DE) versus non-de-escalation (non-DE) antibiotic strategies; RCTs are denoted by a star in the forest plot.

966

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

Fig. 5. Pooled odds ratios for all-cause mortality with de-escalation (DE) versus non-de-escalation (non-DE) by study design.

agents. Differences between antibiotics probably exist and cannot be appreciated from the non-inferiority trials used for their approval [34,35]. The main limitation of our analysis concerns the heterogeneity of the patient populations, types of infection and the intervention. A systematic review recently summarized the variability in the definitions of de-escalation, despite having limited study inclusion to those conducted in the ICU [36]. A consensus statement on blactam hierarchy might prove useful to standardize future deescalation studies [37]. Heterogeneity in the unadjusted mortality analysis was partially explained by the type of infection and perhaps, thereby, by the degree of certainty in the pathogens as responsible for the infection that directed de-escalation. Severity of infection and the empirical antibiotic regimen (broad-spectrum or not) probably did not introduce heterogeneity into our analyses. We restricted inclusion to patients treated with appropriate empirical antibiotic treatment to standardize an important determinant of survival from sepsis [4]. In summary, observational studies show lower mortality following antibiotic de-escalation guided by culture results among patients with bacteraemia, pneumonia or severe sepsis, whereas three small RCTs favoured no de-escalation without reaching statistical significance. We have no information on the effects of deescalation on resistance in the individual treated or in the environment of the patient. Observational studies need to focus on the ecological impact of de-escalation on resistance rather than on mortality. RCTs are needed to assess the safety of antibiotic deescalation, especially in the context of pneumonia and to assess the effects of a policy implementing antibiotic de-escalation on resistance. Transparency Declaration The authors declare that they have no conflicts of interest. Appendix A. Supplementary data Additional Supporting Information may be found in the online version of this article http://dx.doi.org/10.1016/j.cmi.2016.05.023. References [1] Paul M, Greub G. The hidden killer: are we improving the management of bacteremia? Clin Microbiol Infect 2015;21:291e4. [2] Weir DL, Majumdar SR, McAlister FA, Marrie TJ, Eurich DT. The impact of multimorbidity on short-term events in patients with community-acquired pneumonia: prospective cohort study. Clin Microbiol Infect 2015;21: 264e7e264e213. [3] Leibovici L, Shraga I, Drucker M, Konigsberger H, Samra Z, Pitlik SD. The benefit of appropriate empirical antibiotic treatment in patients with bloodstream infection. J Intern Med 1998;244:379e86. [4] Paul M, Shani V, Muchtar E, Kariv G, Robenshtok E, Leibovici L. Systematic review and meta-analysis of the efficacy of appropriate empiric antibiotic therapy for sepsis. Antimicrob Agents Chemother 2010;54:4851e63. [5] Kaye KS. Antimicrobial de-escalation strategies in hospitalized patients with pneumonia, intra-abdominal infections, and bacteremia. J Hosp Med 2012;7(Suppl 1):S13e21.

[6] Laxminarayan R, Duse A, Wattal C, Zaidi AK, Wertheim HF, Sumpradit N, et al. Antibiotic resistance-the need for global solutions. Lancet Infect Dis 2013;13: 1057e98. [7] Schulz L, Osterby K, Fox B. The use of best practice alerts with the development of an antimicrobial stewardship navigator to promote antibiotic deescalation in the electronic medical record. Infect Control Hosp Epidemiol 2013;34:1259e65. [8] Silva BN, Andriolo RB, Atallah AN, Salomao R. De-escalation of antimicrobial treatment for adults with sepsis, severe sepsis or septic shock. Cochrane Database Syst Rev 2013;3:CD007934. [9] Horan TC, Gaynes RP. Surveillance of nosocomial infections. In: Mayhall CG, editor. Hospital Epidemiology and Infection Control. 3rd ed. Philadelphia: Lippincott Williams & Wilkins; 2004. p. 1659e702. [10] Higgins JPT, Green S, editors. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March 2011]. The Cochrane Collaboration; 2011. Available from, www.cochrane-handbook.org. [11] Wan X, Wang W, Liu J, Tong T. Estimating the sample mean and standard deviation from the sample size, median, range and/or interquartile range. BMC Med Res Methodol 2014;14:135. [12] Review Manager (RevMan) [Computer program]. Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration; 2014. [13] Alvarez-Lerma F, Alvarez B, Luque P, Ruiz F, Dominguez-Roldan JM, Quintana E, et al. Empiric broad-spectrum antibiotic therapy of nosocomial pneumonia in the intensive care unit: a prospective observational study. Crit Care 2006;10:R78. [14] Carugati M, Franzetti F, Wiemken T, Kelly R, Peyrani P, Blasi F, et al. Deescalation therapy among bacteraemic patients with community-acquired pneumonia. Clin Microbiol Infect 2015;21:936e911e38. [15] Cremers AJ, Sprong T, Schouten JA, Walraven G, Hermans PW, Meis JF, et al. Effect of antibiotic streamlining on patient outcome in pneumococcal bacteraemia. J Antimicrob Chemother 2014;69:2258e64. [16] Eachempati SR, Hydo LJ, Shou J, Barie PS. Does de-escalation of antibiotic therapy for ventilator-associated pneumonia affect the likelihood of recurrent pneumonia or mortality in critically ill surgical patients? J Trauma 2009;66: 1343e8. [17] Falguera M, Ruiz-Gonzalez A, Schoenenberger JA, Touzon C, Gazquez I, Galindo C, et al. Prospective, randomised study to compare empirical treatment versus targeted treatment on the basis of the urine antigen results in hospitalised patients with community-acquired pneumonia. Thorax 2010;65: 101e6. [18] Garnacho-Montero J, Gutierrez-Pizarraya A, Escoresca-Ortega A, CorciaPalomo Y, Fernandez-Delgado E, Herrera-Melero I, et al. De-escalation of empirical therapy is associated with lower mortality in patients with severe sepsis and septic shock. Intensive Care Med 2014;40:32e40. [19] Joffe AR, Muscedere J, Marshall JC, Su Y, Heyland DK. Canadian Critical Care Trials G. The safety of targeted antibiotic therapy for ventilator-associated pneumonia: a multicenter observational study. J Crit Care 2008;23:82e90. [20] Khasawneh FA, Karim A, Mahmood T, Ahmed S, Jaffri SF, Mehmood M. Safety and feasibility of antibiotic de-escalation in bacteremic pneumonia. Infect Drug Resist 2014;7:177e82. [21] Khasawneh FA, Karim A, Mahmood T, Ahmed S, Jaffri SF, Tate ME, et al. Antibiotic de-escalation in bacteremic urinary tract infections: potential opportunities and effect on outcome. Infection 2014;42:829e34. [22] Kim JW, Chung J, Choi SH, Jang HJ, Hong SB, Lim CM, et al. Early use of imipenem/cilastatin and vancomycin followed by de-escalation versus conventional antimicrobials without de-escalation for patients with hospitalacquired pneumonia in a medical ICU: a randomized clinical trial. Crit Care 2012;16:R28. [23] Kollef MH, Morrow LE, Niederman MS, Leeper KV, Anzueto A, Benz-Scott L, et al. Clinical characteristics and treatment patterns among patients with ventilator-associated pneumonia. Chest 2006;129:1210e8. [24] Koupetori M, Retsas T, Antonakos N, Vlachogiannis G, Perdios I, Nathanail C, et al. Bloodstream infections and sepsis in Greece: over-time change of epidemiology and impact of de-escalation on final outcome. BMC Infect Dis 2014;14:272. [25] Lee CC, Lee NY, Chen PL, Hong MY, Chan TY, Chi CH, et al. Impact of antimicrobial strategies on clinical outcomes of adults with septic shock and community-onset Enterobacteriaceae bacteremia: de-escalation is beneficial. Diagn Microbiol Infect Dis 2015;82:158e64. [26] Leone M, Bechis C, Baumstarck K, Lefrant JY, Albanese J, Jaber S, et al. Deescalation versus continuation of empirical antimicrobial treatment in severe

M. Paul et al. / Clinical Microbiology and Infection 22 (2016) 960e967

[27]

[28]

[29]

[30]

[31]

[32]

sepsis: a multicenter non-blinded randomized noninferiority trial. Intensive Care Med 2014;40:1399e408. Rello J, Vidaur L, Sandiumenge A, Rodriguez A, Gualis B, Boque C, et al. Deescalation therapy in ventilator-associated pneumonia. Crit Care Med 2004;32:2183e90. Shime N, Kosaka T, Fujita N. De-escalation of antimicrobial therapy for bacteraemia due to difficult-to-treat Gram-negative bacilli. Infection 2013;41: 203e10. Shime N, Satake S, Fujita N. De-escalation of antimicrobials in the treatment of bacteraemia due to antibiotic-sensitive pathogens in immunocompetent patients. Infection 2011;39:319e25. Soo Hoo GW, Wen YE, Nguyen TV, Goetz MB. Impact of clinical guidelines in the management of severe hospital-acquired pneumonia. Chest 2005;128: 2778e87. Giantsou E, Liratzopoulos N, Efraimidou E, Panopoulou M, Alepopoulou E, Kartali-Ktenidou S, et al. De-escalation therapy rates are significantly higher by bronchoalveolar lavage than by tracheal aspirate. Intensive Care Med 2007;33:1533e40. Fabregas N, Ewig S, Torres A, El-Ebiary M, Ramirez J, de La Bellacasa JP, et al. Clinical diagnosis of ventilator associated pneumonia revisited: comparative

[33]

[34]

[35] [36]

[37]

967

validation using immediate post-mortem lung biopsies. Thorax 1999;54: 867e73. Mondi MM, Chang MC, Bowton DL, Kilgo PD, Meredith JW, Miller PR. Prospective comparison of bronchoalveolar lavage and quantitative deep tracheal aspirate in the diagnosis of ventilator associated pneumonia. J Trauma 2005;59:891e5. discussion 895e6. Yahav D, Lador A, Paul M, Leibovici L. Efficacy and safety of tigecycline: a systematic review and meta-analysis. J Antimicrob Chemother 2011;66: 1963e71. Yahav D, Paul M, Fraser A, Sarid N, Leibovici L. Efficacy and safety of cefepime: a systematic review and meta-analysis. Lancet Infect Dis 2007;7:338e48. Tabah A, Cotta MO, Garnacho-Montero J, Schouten J, Roberts JA, Lipman J, et al. A Systematic review of the definitions, determinants, and clinical outcomes of antimicrobial de-escalation in the intensive care unit. Clin Infect Dis 2016;62:1009e17. Weiss E, Zahar JR, Lesprit P, Ruppe E, Leone M, Chastre J, et al. Elaboration of a consensual definition of de-escalation allowing a ranking of beta-lactams. Clin Microbiol Infect 2015;21(7):649e1e649e10.