Health Care Associated Infection (HAI): A Critical Appraisal of the Emerging Threat Proceedings of the HAI Summit

Transcription

1 SUPPLEMENT ARTICLE Health Care Associated Infection (HAI): A Critical Appraisal of the Emerging Threat Proceedings of the HAI Summit Marin H. Kollef, 1,2 Lena M. Napolitano, 3 Joseph S. Solomkin, 4 Richard G. Wunderink, 5 In-Gyu Bae, 7 Vance G. Fowler, 7 Robert A. Balk, 6 Dennis L. Stevens, 8 James J. Rahal, 9,10 Andrew F. Shorr, 11,12 Peter K. Linden, 13 and Scott T. Micek 1,2 1 Washington University School of Medicine and 2 Barnes-Jewish Hospital, St. Louis, Missouri; 3 University of Michigan Health Center, Ann Arbor; 4 University of Cincinnati College of Medicine, Cincinnati, Ohio; 5 Feinberg School of Medicine, Northwestern University, and 6 Rush University Medical Center and Rush Medical College, Chicago, Illinois; 7 Duke University Medical Center, Durham, North Carolina; 8 Veterans Affairs Medical Center, Boise, Idaho; 9 New York Hospital Queens, Flushing, and 10 Weill Medical College of Cornell University, New York, New York; 11 Washington Hospital Center and 12 Georgetown University, Washington, DC; and 13 University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania During the Health Care Associated Pneumonia Summit conducted in June 2007, it was found that there is a need for educational efforts in several areas of health care associated infections (HAI) that extend beyond pneumonia. This supplement to Clinical Infectious Diseases represents the proceedings of the HAI Summit, a diverse panel of clinical investigators whose goal was to assess the quality of evidence regarding issues surrounding HAI and to discuss potential implications for its diagnosis and treatment in the future. The classification of bacterial infections is in a state of flux. Most of the prior classification schemes have segregated these infections according to the site of infection (e.g., lung, urinary tract, soft tissue and skin, and intra-abdominal) and the location of the patient at the time the infection developed. The latter has historically been divided into community-acquired and nosocomial (hospital-acquired) infections [1, 2]. Unfortunately, this simple classification scheme is no longer adequate, because of changing patient demographics and risk profiles for infection with potentially antibiotic-resistant bacteria, which historically have been encountered primarily in the hospital setting. Patients with serious infections (e.g., pneumonia, bacteremia, and septic shock) should be given treatment initially with antibiotics active against the bacterial pathogens causing the infection (i.e., appropriate antibiotic therapy). Additionally, appropriate antibiotic Reprints or correspondence: Dr. Marin H. Kollef, Div. of Pulmonary and Critical Care Medicine, Washington University School of Medicine, 660 S. Euclid Ave., Campus Box 8052, St. Louis, MO (mkollef@im.wustl.edu). Clinical Infectious Diseases 2008; 47:S by the Infectious Diseases Society of America. All rights reserved /2008/4707S2-0002$15.00 DOI: / therapy should be administered in a timely manner to optimize the likelihood of a clinical response. The support for these recommendations comes from investigations demonstrating that patients initially given treatment with antibiotic regimens that are not active against the causative bacterial species (i.e., inappropriate antibiotic therapy) have a greater risk for in-hospital mortality than do patients receiving appropriate therapy [3 5]. Classification schemes should assist clinicians in identifying patients at risk for antibiotic-resistant infections, thereby requiring initial treatment with broadspectrum antimicrobials. The recognition of potentially antibiotic-resistant infections occurring outside the hospital setting has resulted in the formulation of the new category, termed health care associated infections (HAIs). Implicit in the definition of HAIs is that patients will require initial therapy with more broadspectrum antibiotics, compared with patients with community-acquired infections. HAIs have been defined using various criteria (table 1). Friedman et al. [6] evaluated patients admitted to the hospital with bloodstream infections (BSIs) and showed that individuals with HAI risk factors were statistically more likely than were patients with community-acquired infections to be infected with anti- HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S55

2 Table 1. Risk factors used to define health care associated infections. Infection type [source] and criteria Bacteremia [6] Health care associated BSI was defined by a positive culture result for a blood specimen obtained from a patient at the time of hospital admission or within 48 h after admission if the patient fulfilled any of the following criteria: 1. Received intravenous therapy at home; received wound care or specialized nursing care through a health care agency, family, or friends; or had self-administered intravenous medical therapy in the 30 days before the BSI. Patients whose only home therapy was oxygen use were excluded. 2. Attended a hospital or hemodialysis clinic or received intravenous chemotherapy in the 30 days before the BSI 3. Was hospitalized in an acute care hospital for 2 daysin the 90 days before the BSI 4. Resided in a nursing home or long-term-care facility Pneumonia [7] HCAP was defined as a diagnosis of pneumonia in patients with a first positive bacterial respiratory culture finding within 2 days of admission and any of the following: 1. Admission source indicates a transfer from another health care facility 2. Receiving long-term hemodialysis 3. Prior hospitalization within 30 days for those whose condition does not meet VAP definition Pneumonia [8] HCAP was defined as a diagnosis of pneumonia in patients admitted to the hospital who met at least 1 of the following criteria: 1. Admission from a nursing home, rehabilitation hospital, or other long-term nursing care facility 2. Previous hospitalization within the immediately preceding 12 months 3. Receiving outpatient hemodialysis, peritoneal dialysis, or infusion therapy necessitating regular visits to a hospitalbased clinic 4. Having an immunocompromised state NOTE. Adapted from [6], from [7], and from [8]. BSI, bloodstream infection; HCAP, health care associated pneumonia; VAP, ventilator-associated pneumonia. biotic-resistant bacteria, including methicillin-resistant Staphylococcus aureus (MRSA) and antibiotic-resistant enterococci. In an accompanying overview, the importance of the classification of HAIs was emphasized in terms of identifying a group of patients who would potentially benefit from initial treatment with broad-spectrum antibiotics [9]. Similarly, Kollef et al. [7] examined 4543 patients with microbiologically confirmed pneumonia from a multicenter administrative database. They separated patients into 4 categories: community-acquired pneumonia (CAP), health care associated pneumonia (HCAP), hospital-acquired pneumonia (HAP), and ventilator-associated pneumonia (VAP). Patients with HCAP had underlying comorbidities and bacterial pathogens similar to those of patients with HAP and VAP. The most common bacterial pathogen isolated in patients with HCAP was MRSA. The in-hospital mortality rate among patients with HCAP was similar to that observed among patients with HAP (19.8% vs. 18.1%; P.005), both being almost twice the mortality rate observed among patients with CAP (10%; P!.001 for both comparisons) [7]. The American Thoracic Society (ATS) and Infectious Diseases Society of America (IDSA) guidelines for HCAP, HAP, and VAP have summarized potential risk factors for HAIs (table 1) [1, 2]. These are the first published guidelines to recognize the category of HCAP in terms of recommending initial broad-spectrum antimicrobial treatment because of the high prevalence of antibiotic-resistant bacteria as the causative agents of infection. Further support for this recommendation comes from a large singlecenter study evaluating patients with microbiologically confirmed pneumonia admitted to an urban teaching hospital [8]. Among the 639 patients with microbiologically confirmed pneumonia evaluated in that study, HCAP made up 67.4% of the pneumonia cases, and CAP accounted for 32.6%. Patients with HCAP were statistically more likely to be infected with MRSA, Pseudomonas aeruginosa, and other nonfermenting gram-negative rods, compared with patients with CAP. Patients with HCAP were also significantly more likely to have received inappropriate initial antimicrobial therapy (28.3% vs. 13.0%; P!.001) and had greater in-hospital mortality (24.6% vs. 9.1%; P!.001), compared with patients with CAP. The importance of correctly classifying patients with HAI risk profiles is demonstrated by 2 recent studies. Schramm et al. [5] evaluated patients with MRSA sterile-site infections and showed that patients with a positive sterile-site culture specimen obtained during the first 48 h of hospitalization were significantly less likely to have received empirical treatment for MRSA. This occurred despite the fact that most patients had readily identifiable risk factors for HAI, which suggests that the treating clinicians did not recognize either the presence of these risk factors or the associated therapeutic implications. In a prospective before-after study using a protocol and standardized order set for the management of septic shock in the emergency department, a statistically significant reduction (from 48.3% to 30%; P p.04) in 28-day mortality was associated with the prescription of broadspectrum antibiotics to patients with risk factors for HAI [10]. These studies suggest that many patients evaluated during the early periods of their hospitalization may benefit from having their infection identified as an HAI, so that more-appropriate initial antibiotic therapy can be prescribed. During the HCAP Summit conducted in June 2007, it was found that there is a need for educational efforts in several areas of HAI that extend beyond pneumonia. This supplement to Clinical Infectious Diseases represents the proceedings of a diverse panel of clinical investigators whose goal was to assess the quality of evidence in support of the clinical classification S56 CID 2008:47 (Suppl 2) Kollef et al.

3 Table 2. Health Care Associated Infection (HAI) Summit clinical practice statements. Workshop 1: Treatment by Sites of Infection (statements 1 5) 1. Patients at risk for health care associated complicated skin and soft tissue infections are more likely to have both resistant gram-negative and gram-positive pathogens. (L.M.N.) 2. Patients with health care associated intra-abdominal infections should receive dual empiric therapy for resistant gramnegative and gram-positive pathogens. (J.S.S.) 3. Early aggressive, appropriate empiric treatment and de-escalation for HCAP reduces mortality and minimizes resistance. (R.G.W.) 4. Health care associated BSIs require empiric coverage for MDR gram-negative bacteria and MRSA, as well as coverage for fungal pathogens in patients with specific risk factors. (V.G.F.) 5. Initial appropriate antimicrobial therapy and source control are the most important determinants of outcome in severe sepsis and septic shock. (R.A.B.) Workshop 2: Treatment by Organism (statements 6 10) 6. Vancomycin is obsolete for treating MRSA infections. (D.L.S.) 7. Serious HAIs due to suspected gram-negative bacteria should be treated empirically with dual coverage that includes an aminoglycoside. (J.J.R.) 8. Patients with serious HAIs who have risk factors for fungal infections require early empiric antifungal therapy to reduce mortality. (A.F.S.) 9. All infections in immunocompromised patients should be considered HAIs until proven otherwise. (P.K.L.) 10. Adjunctive therapy should be utilized to prevent and treat serious HAIs. (S.T.M.) NOTE. BSI, bloodstream infection; HCAP, health care associated pneumonia; MDR, multidrug resistant; MRSA, methicillin-resistant Staphylococcus aureus. of HAI as a distinct entity and the need for specific therapeutic interventions for HAI. Ten clinical practice statements were drafted by the chair (M.H.K.) and the 2 workshop leaders (L.M.N. and D.L.S.) and were subsequently evaluated by an 11-member panel with expertise in infectious diseases, surgery, critical care, pharmacology, and outcomes research (table 2). Before the summit was convened, each participant was assigned a statement and was instructed to systematically review and summarize the evidence supporting or refuting that statement. In the first phase of the live meeting, the simultaneously conducted workshops Treatment by Sites of Infection and Treatment by Organism included a leader and 4 or 5 content experts and served as a forum for each individual to present the evidence for his or her assigned statement. When the data were presented, primary attention was given to the study methodology, the number of patients enrolled, and the outcome events. After the presentation of data for each statement, workshop members discussed the evidence, graded the strength of the evidence, and assigned the statement a consensus numeric grade through a voting process (table 3). In the second phase of the live meeting, all summit panelists reconvened as a single group, reviewed the workshop summaries, and discussed each statement further. After each discussion, all participants voted on their individual levels of support, using the grading scheme shown in table 3. In addition to defining the level of evidence available for each statement, the panel members also outlined additional data required to further refine the statements for future clinical uses. Before the summit meeting, clinical perspectives of practicing physicians were measured via a Web-based survey. E- mail polling was done to ascertain their level of support for the same 10 statements. The invitation to participate in the electronic survey was sent to 3300 members of the IDSA (all active addresses). Of the IDSA members surveyed, 744 (23%) responded. The purpose of the electronic surveys was to provide information that would allow for the comparison of data-driven responses from the content experts at the summit with those from clinicians practicing in the field. The summit participants and the surveyed physicians used the same voting scheme for Individual Level of Support to grade the 10 statements (table 3). This exercise was performed to determine the prevailing current opinions regarding HAIs and areas where additional research and knowledge is required. In this era of increasing antimicrobial resistance, clinical decision making regarding the Table 3. Workshop and Health Care Associated Infection Summit panel voting schemes. Category I II III IV V A B C D E Nature of evidence Evidence obtained from at least 1 well-designed, randomized, controlled trial Evidence obtained from well-designed cohort or case-control studies Evidence obtained from case series, case reports, or flawed clinical trials Opinions of respected authorities based on clinical experience, descriptive studies, or reports of expert committees Insufficient evidence to form an opinion Level of workshop support for statement There is good evidence to support the statement There is fair evidence to support the statement There is poor evidence to support the statement, but recommendations may be made on other grounds There is fair evidence to reject the statement There is good evidence to reject the statement Individual level of support 1 Accept recommendation completely 2 Accept recommendation with some reservations 3 Accept recommendation with major reservations 4 Reject recommendation with reservations 5 Reject recommendation completely HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S57

4 management of suspected bacterial infections has become increasingly complex. Given factors such as the aging of our population, the increasing use of immunomodulating therapies, and the practice of caring for patients with more-complicated cases outside of the hospital setting, it is very likely that the prevalence of HAIs will increase. Research to better define this category of infection and its management appears to be very relevant. STATEMENT 1: PATIENTS AT RISK FOR HEALTH CARE ASSOCIATED COMPLICATED SKIN AND SOFT-TISSUE INFECTIONS ARE MORE LIKELY TO HAVE BOTH RESISTANT GRAM-NEGATIVE AND GRAM-POSITIVE PATHOGENS Rationale and Definition of Statement Presently, there is no standard definition for health care associated complicated skin and soft-tissue infection (cssti). The terminology of HAIs was first devised as a new classification scheme for BSIs to distinguish patients with community-acquired, health care associated, and nosocomial infections [6]. Skin and soft-tissue infections (SSTIs) have traditionally been categorized as either uncomplicated or complicated infections, by use of the US Food and Drug Administration (FDA) criteria [11]. Uncomplicated skin infections include simple abscesses, impetiginous lesions, furuncles, and cellulitis. Complicated skin infections include deeper soft-tissue infections or those requiring significant surgical intervention, such as infected ulcers, burns, and major abscesses or a significant underlying disease state that complicates the response to treatment. Superficial infections or abscesses in an anatomical site, such as the rectal area, where the risk of anaerobic or gram-negative pathogen involvement is higher, should be considered complicated infections. The microbiology of uncomplicated and complicated skin infections is not the same. In uncomplicated skin infections, S. aureus and Streptococcus pyogenes are the 2 most commonly seen pathogens. Among complicated skin infections, the possible pathogens are numerous, may be monomicrobial or polymicrobial, and are dependent on the clinical situation, the location of the infection, and the medical history of the individual patient. Because no standard definition for health care associated cssti is available, we will review data regarding HAIs in general and the changing epidemiology of csstis. This section aims to assess the strength of evidence supporting the assertion that patients at risk for health care associated cssti are more likely to be infected with both resistant gram-negative and grampositive pathogens. Methods A PubMed database search to identify studies related to the clinical and microbiological features of health care associated csstis was completed on 4 September The search term skin infections yielded a total of 78,866 articles. The search term complicated skin and skin structure infection (csssi) yielded 244 articles, and the search term complicated skin and soft tissue infection (cssti) yielded 100 articles. The search terms health care associated, healthcare associated, and healthcare-associated yielded 51,504, 38,460, and 288 articles, respectively. Combining the search terms health care associated, healthcare associated, and healthcare-associated with infection, using the AND function, produced 5154, 3759, and 250 articles, respectively. Combining the search terms health care associated, healthcare associated, and healthcare-associated with skin infections, using the AND function, produced 138, 109, and 5 articles, respectively. After limiting these articles to the English language, a total of 147 articles were reviewed, and 2 articles were deemed relevant to the statement. Evidence Prevalence of health care associated cssti. No studies were identified as specifically focusing on the prevalence of health care associated cssti. One study specifically addressed the issue of overall prevalence of HAIs in general and included a cohort of patients with skin infections. This study involved a cross-sectional population survey of patients, aged 19 years, admitted to 25 acute-care hospitals participating in the Canadian Nosocomial Infection Surveillance Program, to determine the prevalence of HAIs. A 1-day HAI point-prevalence survey was conducted in February Adult patients who had been admitted at least 48 h before the day of the survey were identified, and the primary outcome was the presence of an HAI, which was identified as an infection not present at admission and with onset at least 72 h after admission. Some would consider these nosocomial infections. The study was limited to the following infections: pneumonia, urinary-tract infection, BSI, surgical-site infection, and Clostridium difficile infection. Centers for Disease Control and Prevention (CDC) definitions were used for all HAIs. A total of 5750 adults were surveyed, 2086 (36%) of whom were receiving at least 1 systemic antimicrobial agent; 601 patients had 667 HAIs, giving a prevalence of 10.5% for infection and 11.6% for HAI. The only skin infection reported was surgical-site infection, which was identified in 146 patients (2.5%). In the multivariate logistic regression model for HAI, the following characteristics were all independently associated with HAI: extended hospital stays of 17 days before the day of the survey, having a central venous S58 CID 2008:47 (Suppl 2) Kollef et al.

5 catheter, having an indwelling urinary catheter, or having an endotracheal tube with or without mechanical ventilation [12]. Epidemiology and microbiology of health care associated cssti. No studies were identified that specifically focused on the microbiology of health care associated cssti. However, multiple studies reported the microbiology of SSTIs in hospitalized patients and patients presenting to the emergency department. The SENTRY Antimicrobial Surveillance Program, established in 1997 by the Jones Group/JMI laboratories and funded by SmithKline Beecham, is designed to monitor antimicrobial resistance among various pathogens around the globe [13]. The SENTRY program recently reported data regarding causative isolates from SSTIs from 3 continents during a 7-year period ( ) [14]. Each year, participating medical centers sent 50 consecutive pathogens from hospitalized patients that were determined to be significant causes of pyogenic wound infections. The isolates were from an SSTI or surgical-site infection and were either community acquired or nosocomial in origin. The predominant pathogens included S. aureus (ranked first in all geographic regions), P. aeruginosa, Escherichia coli, and Enterococcus species. On the global scale, S. aureus was the most frequently occurring pathogen from an SSTI, with MRSA being the greatest resistance concern. Considerable variation in the MRSA rate was noted between countries and continents, with the overall rate highest in North America (35.9%), followed by Latin America (29.4%) and Europe (22.8%). It was noted that the rate of MRSA in North America increased from 26.2% of isolates in 1998 to 47.4% of isolates in Gram-negative isolates as causes of SSTIs were common, and, among non- Enterobacteriaceae gram-negative bacilli, P. aeruginosa had the highest occurrence in SSTIs in all geographic regions. Community-associated MRSA has increased markedly to become the greatest problem facing treatment of SSTIs in the outpatient setting. A comparison of community-associated and health care associated MRSA infections was performed as a prospective cohort study of patients with MRSA infection identified at 12 Minnesota laboratory facilities from 1 January through 31 December Of 1100 MRSA infections, 131 (12%) were community associated, and 937 (85%) were health care associated. SSTIs were more common among communityassociated cases (75%) than among health care associated cases (37%) (OR, 4.25; 95% CI, ) [15]. A prospective, observational study examined patients with SSTIs presenting to the emergency department in an urban public hospital in Oakland, California. Among the 137 patients enrolled, MRSA was present in 51% of infection-site cultures. Of 119 S. aureus isolates (from infection site and nares), 89 (75%) were MRSA, and almost all (99%) of the MRSA isolates possessed the staphylococcal cassette chromosome (SCC) mec type IV allele (typical of community-associated MRSA). Among predictor variables independently associated with MRSA infection, the strongest was presence of furunculosis (OR, 28.6). In this urban population, MRSA was the leading pathogen in SSTIs [16]. The CDC and 3 sites participating in the Emerging Infections Program began a specialized, prospective MRSA surveillance project in 2001 using the Active Bacterial Core Surveillance program, a population-based surveillance component of the Emerging Infections Program Network designed to study the epidemiologic features of invasive bacterial disease and to track drug resistance in the United States. The MRSA Active Bacterial Core Surveillance project monitored all MRSA isolates from all body sites from patients in select hospitals in Baltimore, Atlanta, and Minnesota. From 2001 through 2002, 1647 cases of community-acquired MRSA infections were reported, and 77% involved skin and soft tissue. Overall, 23% of patients were hospitalized for the MRSA infection. This study concluded that community-associated MRSA skin infections were now a common problem [17]. A prospective multicenter study confirmed this finding. Adult patients with acute, purulent SSTIs presenting to 11 university-affiliated emergency departments during the month of August 2004 were enrolled to determine the causative bacterial isolates. S. aureus was isolated from 320 (76%) of 422 patients with SSTIs. The prevalence of MRSA was 59% overall, and USA 300 isolates accounted for 97% of MRSA isolates; SCC mec type IV and the Panton-Valentine leukocidin toxin gene were detected in 98% of MRSA isolates, consistent with communityassociated MRSA infection. Methicillin-susceptible S. aureus (MSSA) was identified in only 17% of patients with SSTIs. In this study, MRSA was the most common identifiable cause of SSTIs among patients presenting to emergency departments in 11 US cities [18]. None of these studies specifically differentiate between the epidemiologic characteristics of community-associated (i.e., with no established risk factors) versus health care associated (i.e., with health care associated risk factors) SSTIs. They do, however, address the issue of differences in the microbiological characteristics between the community-associated and health care associated MRSA isolates (table 4) [19]. An active, prospective, laboratory surveillance study conducted at a 1000-bed urban hospital and its affiliated outpatient clinics in Atlanta, Georgia, identified S. aureus that was recovered from SSTIs in 384 persons and 389 episodes of infection, with MRSA accounting for 72% (279 of 389 episodes). Among all S. aureus isolates, 63% (244 of 389 isolates) were community-acquired MRSA. Among MRSA isolates, 87% (244 of 279 isolates) were community-acquired MRSA, and 99% were USA 300 clones. Factors independently associated with communityacquired MRSA infection were black race (prevalence ratio, HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S59

6 Table 4. Comparison of community-associated and health care associated methicillin-resistant Staphylococcus aureus (MRSA). Characteristic Community-associated MRSA Health care associated MRSA Susceptibility, drug Chloramphenicol Usually susceptible Frequently resistant Clindamycin Usually susceptible Frequently resistant Erythromycin Usually resistant Usually resistant Fluoroquinolone Geographic variability Usually resistant TMP-SMZ Usually susceptible Usually susceptible SCC mec type IV II Lineage USA 300, USA 400 USA 100, USA 200 Toxin producing More Fewer Panton-Valentine leukocidin producing Common Rare Health care exposure Less frequent More frequent NOTE. SCC, staphylococcal chromosome cassette; TMP-SMZ, trimethoprim-sulfamethoxazole. Adapted from [19]. 1.53; 95% CI, ), female sex (prevalence ratio, 1.16; 95% CI, ), and hospitalization within the previous 12 months (prevalence ratio, 0.80; 95% CI, ). Inadequate initial antibiotic therapy was statistically significantly more common among those with community-acquired MRSA (65%) than among those with MSSA (1%) SSTI. This study concluded that the community-acquired MRSA USA 300 clone was the predominant cause of community-onset S. aureus SSTIs, and therefore empirical use of agents active against communityacquired MRSA is warranted for patients presenting with serious SSTIs. The study setting was a 1000-bed, urban hospital and its affiliated outpatient clinics in Atlanta, Georgia [20]. Epidemiology and microbiology of cssti. Because no published data are available for health care associated csstis, we reviewed recent studies that served as FDA registration trials for antimicrobials used to treat cssti. In 2 randomized international trials involving 1092 patients with cssti, daptomycin was compared with conventional antibiotics (penicillinase-resistant penicillin or vancomycin). S. aureus was the leading causative pathogen, isolated in 70% of patients; MRSA accounted for only 10% of the S. aureus isolates. Streptococci and enterococci were also common causative pathogens [21]. In another phase 3 cssti study, patients ( n p 854) were randomized to receive dalbavancin or linezolid. Baseline cultures yielded at least 1 gram-positive pathogen for 550 patients (64%; the microbiological intent-to-treat population). Of these, 90% presented with a single gram-positive pathogen. S. aureus was predominant (89% of all patients). Of the S. aureus isolates, 278 (57%) of 492 were MRSA. Overall, 51% of patients presented with cssti that involved MRSA [22]. Two phase 3, double-blind studies randomized hospitalized adults with cssti to receive tigecycline or vancomycin-aztreonam ( n p 1116). S. aureus, with a majority of isolates being MSSA, was the leading pathogen, and streptococci were also common. Gram-negative isolates (only E. coli) were uncommon and were isolated in only 59 patients [23]. Two large, multinational, double-blind, randomized, phase 3 clinical studies (ATLAS 1 and ATLAS 2) enrolled 1867 patients with cssti, 719 of whom were infected with MRSA, and determined that televancin was not inferior to vancomycin. S. aureus was the primary pathogen isolated in these studies, as it was in the 2 prior phase 2 trials (FAST 1 and FAST 2) [24 26]. A multicenter, global, randomized, double-blind trial compared ceftobiprole with vancomycin for patients ( ) n p 784 with cssti and confirmed the noninferiority of ceftobiprole, and S. aureus was the primary causative isolate [27]. A second cssti trial also included patients with diabetic foot infections, and gram-negative pathogens were more common. Gram-positive pathogens were isolated from 79% of patients and MRSA was the most common pathogen (42.4%). Gram-negative pathogens were isolated from 29% of patients, and E. coli (11.0%) and Pseudomonas isolates (6.6%) were the most common [28]. All these studies confirm that the most common causative pathogens in csstis are aerobic gram-positive cocci, with S. aureus and MRSA as the leading isolates. Epidemiology and microbiology of surgical-site infections. Surgical-site infections are also included in the cssti category. In a report from the National Nosocomial Infections Surveillance System from , an analysis of 1410,000 bacterial isolates associated with hospital-acquired infections (BSIs, pneumonia, surgical-site infection, and urinary-tract infection) in intensive care units (ICUs) were reported. For surgical-site infections, the percentage of bacterial isolates that were gram negative decreased during the study period (from 56.5% in 1986 to 33.8% in 2003). By the mid-1990s, gram-positive bacterial pathogens were more commonly reported as causative isolates in surgical-site infections, with S. aureus as the leading pathogen [9]. MRSA has emerged as the most common isolate S60 CID 2008:47 (Suppl 2) Kollef et al.

7 Figure 1. Voting comparison for statement 1 ( Patients at risk for health care associated complicated skin and soft tissue infections are more likely to have both resistant gram-negative and gram-positive pathogens ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Web-based survey; Summit refers to the Health Care Associated Infection Summit panel. causing surgical-site infections in most institutions [29]. MRSA surgical-site infections are associated with significantly increased mortality (OR, 3.4; P p.003), length of hospital stay, and costs, compared with MSSA surgical-site infections [30]. Community-associated MRSA strains are increasingly recovered from hospital settings, and a recent retrospective review of surgical-site infection in in Alabama determined that 57% of MRSA strains from surgical-site infections were of the USA 300 genotype, confirming that community-associated MRSA was a prominent cause of surgical-site infection at that institution [31]. Epidemiology and microbiology of diabetic foot infection. The bacteriology of diabetic foot infections was assessed in a recent study of 371 patients (infected ulcer and cellulitis were the most common types of infection). Overall, one-half of all patients had only gram-positive cocci isolated (355 isolates). Of these, S. aureus, coagulase-negative staphylococci, streptococci, and enterococci were the most common isolates. Gramnegative bacteria, predominantly Pseudomonas and Enterobacteriaceae species, were isolated in 105 patients, and 32% had mixed infections with both gram-positive and gram-negative pathogens [32]. In the SIDESTEP study (of ertapenem vs. piperacillin/tazobactam for treatment of diabetic foot infection; n p 586), infections were polymicrobial in 47% of evaluable patients, and 9% had both gram-positive and gram-negative aerobic organisms isolated by culture. The most commonly isolated pathogens were gram-positive aerobic cocci (257 isolates), with S. aureus as the leading isolate, followed by gramnegative aerobic bacilli isolates (102 isolates), with Enterobacteriaceae species as the leading isolates [33]. The IDSA guidelines for diagnosis and treatment of diabetic foot infection state, Aerobic gram-positive cocci (especially S. aureus) are the predominant pathogens in diabetic foot infections. Patients who have chronic wounds or who have recently received antibiotic therapy may also be infected with gramnegative rods, and those with foot ischemia or gangrene may have obligate anaerobic pathogens [34, p. 885]. Grading of Evidence On the basis of a review of the studies cited above, the workshop members agreed that there was substantial evidence available to reject this statement. In evaluating the nature of the evidence, 20% voted category I, 60% voted category II, and 20% voted category III (table 3). Level of Support When voting on the individual level of support for this statement, 0% of the summit participants voted to accept the statement completely, 18% voted to accept the statement with some reservations, 9% voted to accept the statement with major reservations, 45% voted to reject the statement with reservations, and 27% voted to reject the statement completely. In comparison, of the 744 IDSA members who participated in the online survey, 32% voted to accept the statement completely, 42% voted to accept the statement with some reservations, 11% voted to accept the statement with major reservations, 12% voted to reject the statement with reservations, and 3% voted to reject the statement completely (figure 1). Discussion Presently, there is no true category or definition of health care associated cssti, and no studies were identified in the published literature. The traditional categories of SSTI include uncomplicated versus complicated (initially proposed by the FDA for the conduct of clinical trials for SSTIs) and community acquired or community onset versus hospital acquired or nosocomial. The leading causative pathogen of SSTIs in both community and hospitalized patients is MRSA. This has been confirmed with an in-depth review of the recent registration trials for csstis that discusses the microbiology of new antimicrobials (daptomycin, dalbavancin, telavancin, tigecycline, and ceftobiprole). S. aureus was the leading pathogen in all studies, with rising rates of MRSA. The SENTRY Antimicrobial Surveillance Program has documented that the rate of MRSA in SSTIs in HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S61

8 North America increased substantially, from 26.2% of isolates in 1998 to 47.4% of isolates in Community-associated MRSA is the primary pathogen in patients without health care associated risk factors. S. aureus is also the leading pathogen in surgical-site infections, with rising rates of MRSA. Given the high prevalence of MRSA cssti at present, is it important to standardize the classification of these invasive MRSA infections? Several methods are used to classify MRSA as health care associated or community associated, including (1) genotypic testing, based on the results of PFGE or other molecular techniques; (2) phenotypic testing, based on antimicrobial susceptibility testing; and (3) epidemiologic analysis, based on the time from hospital admission to a positive culture result. Definitions of community-associated MRSA often use time-based criteria in which the recovery of MRSA isolates within 48 or 72 h after hospital admission is considered indicative of community-associated MRSA. However, time-based criteria do not consider patients with MRSA infection after recent health care exposure. Furthermore, community-associated MRSA has emerged as a health care associated and nosocomial pathogen [35 38]. Community-associated MRSA strains differ from hospital-acquired MRSA strains in that they are generally susceptible to most antibiotics, whereas nosocomial strains are generally multidrug resistant (MDR). However, these data are not available to the prescribing clinician when empirical antibiotics are selected. The recent epidemiologic reports of invasive MRSA infections ( n p 8987) in the United States, which were associated with 1598 in-hospital deaths, classified cases into mutually exclusive groups (health care associated vs. community associated), first on the basis of health care risk factors. HAIs, in turn, were classified as either community onset or hospital onset (table 5) [39]. In contrast, in diabetic foot infections, gram-negative pathogens and polymicrobial infections are more common than are surgical-site infections and csstis. There are, however, some Table 5. Definitions used for epidemiologic classification of invasive methicillin-resistant Staphylococcus aureus (MRSA) infections. Classification Definition additional cssti categories in which resistant gram-positive and gram-negative pathogens would be likely. These include perineal infections, necrotizing soft-tissue polymicrobial infections, pressure ulcer and decubitus infections, and surgical-site infections related to abdominal and genitourinary surgical procedures. Future Directions Future directions discussed by the summit members include the need to evaluate whether a category of health care associated csstis is appropriate at this time. The use of HAI categories in bacteremia and pneumonia are thought to be important for improving the recognition of those patients who may be infected with MDR pathogens and therefore warrant more broad-spectrum empirical antimicrobial therapy. There is minimal evidence to suggest that the addition of health care associated csstis would have significant implications for the selection of empirical antimicrobial therapy for these patients with skin infections. Other potential classification schemes could be considered for csstis, such as monomicrobial versus polymicrobial, necrotizing versus nonnecrotizing, and pyogenic versus nonpyogenic. Additional detailed studies of SSTIs are warranted to further delineate changes in the microbial etiology of csstis, to optimize treatment strategies and also to evaluate risk factors for recurrence. STATEMENT 2: PATIENTS WITH HEALTH CARE ASSOCIATED INTRA-ABDOMINAL INFECTIONS SHOULD RECEIVE DUAL EMPIRIC THERAPY FOR RESISTANT GRAM-NEGATIVE AND GRAM- POSITIVE PATHOGENS Rationale and Definition of Statement Complicated intra-abdominal infections (ciais) are defined as infections that extend beyond the hollow viscus of origin into the peritoneal space and are associated with either abcess for- Health care associated Community onset Cases with at least 1 of the following health care risk factors: (1) presence of an invasive device at time of admission; (2) history of MRSA infection or colonization; (3) history of surgery, hospitalization, dialysis, or residence in a long-term-care facility in previous 12 months preceding culture date Hospital onset Cases with positive culture result from a normally sterile site obtained 148 h after hospital admission. These cases might also have 1 of the community-onset risk factors. Community associated Cases with no documented community-onset health care risk factor NOTE. Reprinted from JAMA 2007;298: [39]. Copyright 2007, American Medical Association. All rights reserved. S62 CID 2008:47 (Suppl 2) Kollef et al.

9 mation or peritonitis [40]. Intra-abdominal infections pose serious challenges to the treating physicians, and mortality rates approach 60% [41]. Rapid diagnosis, appropriate intervention, and timely and efficacious anti-infective therapy are of critical importance and have been shown to lead to improved patient outcomes. The traditional binary classification scheme for ciais has consisted of nosocomial and community-acquired infections. At present, there is no defined category for health care associated ciai, unlike the distinction that has been made recently with BSI and pneumonia. Many of the data regarding epidemiology and antimicrobial treatment of ciai are derived from antimicrobial trials, and most patients who entered into those trials had community-acquired infections, such as perforated or complicated appendicitis, and have not been severely ill. It is estimated that 80% of all intra-abdominal infections (IAIs) are acquired in a community setting [42]. In community-acquired infections, the location of the gastrointestinal perforation defines the infecting flora: infections that occur beyond the proximal small bowel are caused by facultative and aerobic gram-negative organisms; infections that occur past the proximal ileum can be caused by a variety of anaerobic microorganisms. The IDSA guidelines for ciai used the term health care associated infections to describe nosocomial infections, including ciais acquired postoperatively. HAIs were specifically defined as infections that are most commonly acquired as complications of previous elective or emergent intra-abdominal operations and are caused by nosocomial isolates particular to the site of the operation and to the specific hospital and unit [40, p. 997]. In the context of the HAI Summit s reference to other infections, the term HAI was used to describe infections in individuals who regularly interact with the health care environment. It has been suggested that HAIs represent a unique population of patients. These patients are thought to be infected not only with a different spectrum of pathogens but also with potentially more-resistant flora. The term HAI was first used to characterize a spectrum of BSIs [6]. Similarly, patients with ciai with these same risk factors for HAI may have ample opportunity to acquire resistant bacteria. It is undetermined at present whether an expanded classification scheme, to include health care associated ciai, may be necessary for patients with IAIs. There is currently no standard category or definition for HAIs in the broader category of IAIs. This review focuses on the available literature that characterizes the microbiology of ciais (both community acquired and nosocomial), the importance of appropriate initial empirical therapy, and the incidence of MDR organisms. By assessment of the strength of this evidence, it is possible to ascertain whether an expanded classification system that includes health care associated ciai is needed and would benefit a potential new subgroup of patients. Methods A PubMed database search was conducted on 4 September 2007 to identify relevant reports involving the treatment and microbiological features of health care associated IAIs. The search term intra-abdominal infections yielded a total of 2347 articles. The search terms health care associated, healthcareassociated, and healthcare associated yielded 51,504, 38,460, and 288 articles, respectively. When these terms were combined with intra-abdominal infections, using the AND function, 22 articles were found. After the results were limited to the English language, 0 articles were found to be relevant to the statement. In a second PubMed database search, the search term postoperative peritonitis OR secondary peritonitis yielded 9237 articles. This term was combined with microbiology, using the AND function, yielding 168 articles; with the search term drug resistance, yielding 198 articles; and with the search term appropriate therapy, yielding 220 articles. After results were limited to humans and the English language, 5 articles were found to be relevant to the statement. The IDSA and the Surgical Infection Society guidelines for the treatment of ciais were also reviewed. Evidence No studies specifically related to health care associated intraabdominal infections were identified. The 2 issues of empirical antibiotic therapy and dual empirical therapy for treatment of infection with resistant gram-positive and gram-negative pathogens will be addressed separately. Empirical antimicrobial therapy for ciai. The first portion of the statement recommends that patients with health care associated IAI should receive empirical antimicrobial therapy. A retrospective case study by Krobot et al. [43] assessed the effect of inappropriate initial empirical antibiotic therapy in 425 patients with community-acquired secondary peritonitis. E. coli was the most commonly isolated pathogen. A total of 54 patients (13%) received inappropriate initial therapy. Clinical success, predefined as resolution of infection with initial or step-down therapy after primary surgery, was achieved for 322 patients (75.7%). However, patients were more likely to have clinical success if the initial antibiotic therapy was appropriate than if it was inappropriate (75.7% vs. 53.4%). Patients who had clinical success had a mean length of stay of 13.9 days, compared with 19.8 days for those who had clinical failure. Furthermore, multinomial analyses (with adjustment for patient age, sex, and comorbidities) revealed that inappropriate HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S63

10 antimicrobial therapy was associated with the need for secondline antibiotic therapy and repeated operation. A more recent multicenter study of 425 patients with community-acquired IAI in Spain examined the consequences of inappropriate initial empirical parenteral antibiotic therapy [44]. Initial empirical therapy was classified as appropriate if all isolates were susceptible to at least 1 of the antibiotics administered. A total of 387 patients (92%) received appropriate initial empirical therapy. Patients receiving inappropriate therapy were less likely to have clinical success (79% vs. 26%; P!.001), more likely to require additional antibiotic therapy (40% vs. 7%; ), and more likely to be rehospitalized P!.01 within 30 days after discharge (18% vs. 3%; P!.01). Multi- variate analyses also showed that inappropriate therapy was associated with an almost 16% increase in length of stay and a 26% increase in the number of days of antibiotic therapy. Inappropriate initial antibiotic therapy was associated with a significantly higher proportion of unsuccessful patient outcomes, including death, repeated operation, rehospitalization, additional antibiotic therapy, and increased length of stay. Other studies have confirmed similar findings [45 47]. These data clearly confirm that patients with IAI should receive appropriate empirical antimicrobial therapy. Dual empirical antimicrobial therapy for ciai with resistant gram-positive and gram-negative pathogens. The second portion of the statement recommends that dual empirical therapy for resistant gram-positive and gram-negative pathogens should be used for patients with health care associated ciai. The IDSA guidelines for ciai separate the recommendations regarding selection of anti-infective agents into 2 categories mild-to-moderate and high-severity infections and these may occur in both patients with community-acquired and patients with nosocomial infections [40]. Similarly, the Surgical Infection Society guidelines for ciai separate the recommendations into lower-risk patient and higher-risk patient [48]. In general, for less severely ill patients with community-acquired infections, antimicrobial agents with a narrow spectrum of activity are adequate. The IDSA recommends that community-acquired infections may be managed with a variety of single- and multiple-agent therapeutic regimens that are based, in part, on in vitro activities. The IDSA advises that no particular antimicrobial regimen has consistently been demonstrated to be superior or inferior (table 6). For higher-risk patients or for those with high-severity IAIs, broader-spectrum empirical antimicrobial therapy is recommended to cover potential MDR pathogens. Nosocomial IAIs are typically caused by a more-resistant flora, which may include P. aeruginosa, Acinetobacter species, Enterobacter species, Proteus species, MRSA, enterococci, and Candida species. The IDSA s treatment recommendations for nosocomial ciai suggest multidrug regimens guided by knowledge of nosocomial flora and susceptibility patterns. Several studies have documented that infections involving resistant organisms, particularly those likely to be acquired in the health care setting, are associated with an increased risk of treatment failure, morbidity, and mortality [49 52]. Prolonged preoperative length of stay and prolonged (12 days) preoperative antimicrobial therapy are significant predictors of antimicrobial failure leading to recurrent infection, which suggests that organisms resistant to the empirical antimicrobial regimen may be responsible for infection. Patients with these risk factors should be given treatment for nosocomial infection. Montravers et al. [52] evaluated the incidence of resistant bacterial strains among patients with postoperative peritonitis, as well as the efficacy of empirical antimicrobial therapy. In this study, 100 resistant pathogens were isolated from 70 patients who underwent repeated operation for generalized postoperative peritonitis. Candida species and both gram-negative and gram-positive anaerobic bacteria were isolated (table 7) [52]. The relative frequencies of different pathogens cultured in this patient population differed from those typically found in patients with community-acquired peritonitis. Furthermore, the authors determined that 54% of the patients who received inadequate initial empirical therapy for these resistant pathogens had poorer outcomes, compared with patients who received adequate therapy ( ). P!.05 Roehrborn et al. [53] examined the microbiology of postoperative peritonitis in a prospective case study involving 67 patients. The most common isolates from patients with postoperative peritonitis were E. coli and Enterococcus, Enterobacter, Bacteroides, and Klebsiella species. In addition, patients with postoperative peritonitis were significantly more likely than were patients with community-acquired infections to have the following isolates: enterococci (23 vs. 6), Enterobacter species (13 vs. 4), S. aureus (7 vs. 1), and coagulase-negative staphylococci (7 vs. 1). Patients with community-acquired infections were significantly more likely to have streptococci and E. coli isolated. The Study for Monitoring Antimicrobial Resistance Trends, begun in 2002 and developed by the Merck research program, is designed to monitor resistance patterns among aerobic and facultative gram-negative bacilli isolated worldwide from intraabdominal bacterial clinical isolates collected from multiple centers (including both teaching and community hospitals) [54]. Data from the 2004 report [55] were used in the evaluation of 6156 unique aerobic and facultatively anaerobic gram-negative bacilli isolated from IAIs. Enterobacteriaceae composed 86% of the total isolates, with E. coli (48%), Klebsiella species (16%), and Enterobacter species (9%) comprising the majority of isolates. Quinolone susceptibility rates for E. coli were significantly reduced (60% 70% susceptible), with the lowest rates S64 CID 2008:47 (Suppl 2) Kollef et al.

11 Table 6. Therapeutic regimens for complicated intra-abdominal infections, according to guidelines from the Infectious Diseases Society of America (IDSA) and the Surgical Infection Society (SIS). Single agent Combination regimen Classification IDSA SIS IDSA SIS Low risk (categorized as mild-tomoderate infections by the IDSA and lower-risk patients by the SIS) High risk (categorized as high-severity infections by the IDSA and higher-risk patients by the SIS) Ampicillin/sulbactam a Ticarcillin/clavulanic acid Ertapenem Piperacillin/tazobactam Imipenem/cilastatin Meropenem Ampicillin/sulbactam Ticarcillin/clavulanic acid Ertapenem Cefotetan Cefoxitin Piperacillin/tazobactam Imipenem/cilastatin Meropenem Piperacillin/tazobactam Imipenem/cilastatin Meropenem Cefuroxime or cefazolin plus metronidazole Fluoroquinolone b plus metronidazole Ciprofloxacin plus metronidazole Aztreonam plus metronidazole Third/fourth-generation cephalosporin c plus metronidazole Cefuroxime plus metronidazole Ciprofloxacin plus metronidazole Aztreonam plus clindamycin Third/fourth-generation cephalosporin c plus an antianaerobe Aminoglycoside d plus an antianaerobe Ciprofloxacin plus metronidazole Aztreonam plus clindamycin Third/fourth-generation cephalosporin c plus an antianaerobe Aminoglycoside d plus an antianaerobe e NOTE. Adapted from [40] and from [48]. Because increasing resistance of Escherichia coli to ampicillin and to ampicillin/sulbactam has been reported, local susceptibility profiles should be reviewed before use. b Ciprofloxacin, levofloxacin, moxifloxacin, or gatifloxacin. Cefepime, cefotaxime, ceftazidime, ceftizoxime, or ceftriaxone. d Amikacin, gentamicin, netilmicin, or tobramycin. e Clindamycin or metronidazole. a c

12 Table 7. Organisms reported in a study of postoperative peritonitis by Montravers et al. [52]. Organism No. of isolates Gram negative Escherichia coli 53 Proteus/Morganella species 25 Pseudomonas species 21 Klebsiella species 14 Enterobacter cloacae 10 Acinetobacter/Citrobacter/Serratia species 10 Gram positive Methicillin-resistant Staphylococcus species 24 Enterococcus faecalis 19 Enterococcus faecium 19 Anaerobe Bacteroides species 13 Candida species 23 NOTE. Adapted from [52], with permission from the University of Chicago Press. Some patients had 11 isolate. in the Asia/Pacific region and Latin America. Extended-spectrum b-lactamases (ESBLs) were detected phenotypically in 10% of E. coli, 17% of Klebsiella species, and 22% of Enterobacter species worldwide, representing an increase from the 2 previous years. In this large surveillance program, an additional analysis of 7002 E. coli isolates documented that increasing resistance rates have been seen in both community-acquired and hospital-acquired E. coli infections [56]. Ampicillin-sulbactam was the least active agent (45.1% 67.6% of isolates were susceptible). Quinolones (ciprofloxacin and levofloxacin) also demonstrated low activity (69% 75% susceptible). E. coli isolated!48 h after hospital admission (presumed to be community acquired) were more often susceptible to the agents tested than were E. coli isolated 148 h after hospitalization (presumed to be hospital acquired). There were small differences in susceptibility rates between community-acquired and hospital-acquired E. coli for the carbapenems and amikacin, but there were more sizable differences for other agents, including ampicillin-sulbactam (60.3% vs. 48.4%), ciprofloxacin (83.7% vs. 71.6%), and levofloxacin (83.8% vs. 73.5%). Antimicrobial resistance among gram-negative bacteria isolated from IAIs, both community acquired and nosocomial, is emerging as a more significant problem worldwide. Although resistance rates are of growing concern, there are rare studies that examine the consequences of resistance and adequate empirical treatment for outcomes. A retrospective cohort study by Peralta et al. [57] analyzed patients with E. coli bacteremia to identify associations between antibiotic resistance, adequacy of empirical therapy, and mortality. Of the 663 patients included in the study, those with MDR E. coli bacteremia had a significantly lower frequency of correct empirical treatment than did patients with non-mdr E. coli bacteremia (relative risk [RR], 0.53; 95% CI, ), coupled with a considerably higher mortality rate (RR, 3.31; 95% CI, ). A prospective observational study by Seguin et al. [58] reported factors associated with MDR bacteria in secondary peritonitis. Forty-four cases of community-acquired peritonitis and 49 cases of nosocomial peritonitis (35 postoperative cases) were reported. In univariate analysis, the risk of acquiring an MDR organism was significantly associated with a higher Acute Physiology and Chronic Health Evaluation (APACHE) II score. In addition, preoperative length of hospital stay, previous antimicrobial therapy, and the duration and modification of postoperative antimicrobial therapy were significantly associated with the presence of MDR bacteria. Multivariate analysis confirmed that patients with a preoperative length of hospital stay of 5 days had a higher risk for developing an MDR IAI, especially if antibiotics had been used previously. The authors concluded that knowledge of these 2 risk factors for acquiring MDR bacteria (preoperative length of stay and prior use of antibiotics) enables the use of expanded-spectrum empirical antimicrobial therapy for these specific high-risk patients. No studies were identified that specifically focused on the epidemiology and/or incidence of MDR organisms of health care associated IAIs. Additional discussion regarding the potential definition of health care associated IAIs and patients who would be included in such a category suggested that it could include patients with ciais such as peritoneal dialysiscatheter infections, patients with spontaneous bacterial peritonitis with multiple prior episodes, and patients in nursing homes or long-term-care facilities who develop ciais including appendicitis, cholecystitis, and diverticulitis. There is no consensus as to whether this category of health care associated ciai should be created. Grading of Evidence On the basis of a review of the studies cited above, the workshop members agreed that there was substantial evidence to accept the statement. In evaluating the nature of the evidence, 40% voted category II, 20% voted category III, and 40% voted category IV (table 3). Level of Support When voting on the support for this statement, 0% of the summit participants voted to accept the statement completely, 27% voted to accept the statement with some reservations, 64% voted to accept the statement with major reservations, 9% voted to reject the statement with reservations, and 0% voted to reject the statement completely. In comparison, of the 744 IDSA members who participated in the online survey, 30% voted to accept the statement completely, 38% voted to accept the state- S66 CID 2008:47 (Suppl 2) Kollef et al.

13 ment with some reservations, 11% voted to accept the statement with major reservations, 18% voted to reject the statement with reservations, and 3% voted to reject the statement completely (figure 2). Discussion In summary, a review of the literature produced very limited retrospective studies in general support of the statement. However, in practice, patients suspected of having risk factors for a health care associated infection typically receive empirical therapy for MDR gram-positive and gram-negative organisms. On the basis of the presented studies of patients with secondary peritonitis, it is reasonable to assume that certain patient subgroups were infected with a different spectrum of bacteria, as well as with MDR bacteria. The current studies of patients with secondary peritonitis document that appropriate empirical antibiotic coverage as well as coverage for MDR organisms lead to improved outcomes. Future Directions Traditional categorization of IAIs has segregated them as nosocomial or community-acquired infections. In recent years, epidemiologic studies have identified that pathogens associated with ciai demonstrate rising levels of drug resistance in both groups. It has also been shown that inadequate initial empirical therapy is associated with a significantly higher rate of failures and death. On the basis of studies of patients with postoperative peritonitis, it is reasonable to suggest that select patients may benefit from broad-spectrum empirical therapy. For patients with peritonitis, several attempts have been made to identify clinical features that increase the risk of adverse outcomes. For these patients, the IDSA suggests that antimicrobial regimens with expanded spectra may be warranted. Finally, given the different spectrum of pathogens and the varying levels of resistance seen in patients with peritonitis, an effort should be made to identify other patient types and specific risk factors for IAIs due to resistant pathogens. Because unnecessary broadspectrum therapy is associated with its own problems, caution should be exercised. Future studies will need to be conducted to examine whether health care associated ciai should be delineated as a separate category of IAIs before specific recommendations can be made. STATEMENT 3: EARLY AGGRESSIVE, APPROPRIATE EMPIRIC TREATMENT AND DE-ESCALATION FOR HCAP REDUCES MORTALITY AND MINIMIZES RESISTANCE Rationale and Definition of Statement A designation of health care associated infection was first used for cases of bacteremia in which patients who acquired bacteremia as outpatients were found to have pathogens usually associated with hospital-acquired infections [6]. Of significance, the term referred only to patients who were hospitalized with an infection, not to those who remained in their nonhospital setting. The term health care associated seemed to apply to a variety of infections, including pneumonia, with a similar propensity to be caused by typically nosocomial pathogen. The concept of HCAP was therefore readily embraced. For this reason, HCAP was included in the latest statement on HAP from the ATS and IDSA [1] and was essentially excluded from discussion in the recent IDSA-ATS consensus guidelines on the management of CAP [2]. Because several principals of treatment have been thought to be important for outcomes among patients with HAP and VAP, logic would suggest that these principals are applicable to HCAP as well. These principals include: (1) early initiation of empirical antibiotic treatment; (2) use of broad-spectrum, empirical, antibiotic therapy to avoid inappropriate therapy; and (3) narrowing or de-escalation of empirical antibiotic therapy on the basis of results of respiratory-tract cultures [1]. The purported benefits of such an approach were to decrease the mortality associated with inappropriate initial antibiotics while, at the same time, lessening the emergence of antibiotic-resistant Figure 2. Voting comparison for statement 2 ( Patients with health care associated intra-abdominal infections should receive dual empiric therapy for resistant gram-negative and gram-positive pathogens ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Web-based survey; Summit refers to the Health Care Associated Infection Summit panel. HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S67

14 pathogens. Although it may be logical to assume that these principals and expected results apply to HCAP, this section aims to assess the strength of evidence supporting this assertion. Methods A PubMed search was performed on 1 October With the search limited to the English language, the term health care associated OR healthcare associated OR health care-associated OR healthcare-associated gave a total of 33,408 articles. This result combined with the term pneumonia resulted in a total of 333 articles. This result combined with the term treatment yielded 309 articles. The abstracts were reviewed for pertinence, and additional related articles were also screened. The text word antibiotic was also combined with the 333 health care associated/pneumonia articles, resulting in 87 common articles. Abstracts of all these articles were examined, as were the related articles for each. One apposite article was found. As a consequence of the overlap of HCAP with other pneumonia terms, additional searches were performed. Combination of the terms nursing home AND pneumonia AND treatment resulted in 262 English language articles and 1 relevant article. Evidence No randomized, controlled trials of treatment specific to hospitalized patients with HCAP were found. No concurrent cohort studies of antibiotic treatment in general for hospitalized patients with HCAP were found. Only 1 randomized, controlled trial involving nursing home patients that specifically addressed treatment in the nursing home was available [59]. No category I evidence for any aspect of the statement. Because the entity of HCAP has been defined only recently, studies of either CAP or HAP may have bearing on the statement; therefore, this evidence will also be reviewed. Furthermore, because the statement is multifaceted, each statement component will be discussed separately. Appropriate empirical therapy. Only 1 study specifically addressed the issue of appropriate empirical therapy for HCAP [59]. This was a preintervention and postintervention study of the management of nursing home acquired pneumonia with either oral or parenteral antibiotic therapy. The actual intervention was a guideline for the indication for parenteral antibiotics in a randomized study of 10 skilled-nursing facilities involving either a multidisciplinary or a physician-only training program. After the intervention, use of parenteral antibiotics, when indicated by guidelines, increased significantly ( P!.02) without differences by randomization. No overall mortality benefit was seen. Emergence of resistance was not addressed. Because 35% 40% of patients ultimately required hospitalization, the results have some pertinence to the issue of HCAP. The issue of appropriate antibiotic therapy for HCAP revolves around the microbial etiology of HCAP, specifically whether broad-spectrum antibiotic therapy is needed to empirically cover MDR pathogens, such as P. aeruginosa, MRSA, and ESBL-producing Enterobacteriaceae. Surprisingly, only 3 epidemiologic studies address this issue specifically. Two retrospective US studies focused on culture-positive cases; the first study examined a large administrative database [7], and the second analyzed a single large tertiary care referral hospital [8]. Both studies demonstrated that HCAP was more common than CAP, with a high frequency (20% 25%) of each of the MDR pathogens listed above. Conversely, another study involving a Spanish, multicenter, prospective, observational cohort of patients admitted with pneumonia found that only 17.3% of cases could be classified as HCAP and that the incidences of cases of pneumonia caused by gram-negative organisms (other than Legionella species and Hemophilus influenzae, typical CAP pathogens) and S. aureus were both!5% [60]. However, 32% of patients with HCAP in this study did not receive a microbiological diagnosis, and an additional 20% received a diagnosis of aspiration pneumonia, which left a positive microbiological diagnosis for!50%. Several explanations for these major differences in etiology of HCAP exist. By far the most important is the inclusion of HCAP cases without a microbiological diagnosis. Others include differences in criteria (all immunocompromised patients included vs. only severely immunocompromised patients included; previous hospitalization in the past 12 months vs. the past 3 months), different types of hospitals (major referral centers vs. smaller local hospitals), and study design (retrospective vs. prospective). Even if MDR pathogens occur at high frequency, the use of broad-spectrum therapy is still of unclear benefit with regard to mortality. Two studies, both using the before-after intervention format, specifically addressed this issue. Ibrahim et al. [61] found that the use of a 3-drug, broad-spectrum protocol for late-onset VAP was able to decrease the percentage of patients administered inappropriate initial empirical antibiotic therapy to 5.8%, as opposed to 52% before protocol introduction. Mortality was unaffected, although the incidence of recurrent VAP and subsequent infection with MDR pathogens decreased. The second study [62], which used a similar type of empirical protocol, also demonstrated that the use of inappropriate initial antibiotic therapy was decreased significantly, and broaderspectrum therapy resulted in decreased mortality at 14 days after treatment (27% vs. 8%; P p.03). However, the statistically significant reduction in mortality was not maintained for 30- day or in-hospital mortality. An additional article from the CAP literature on cases of pneumonia caused by gram-negative pathogens from a prospective CAP database was pertinent and thus was reviewed [63]. Most patients with CAP caused by gram-negative path- S68 CID 2008:47 (Suppl 2) Kollef et al.

15 ogens had risk factors that would likely qualify them for HCAP status. Provision of appropriate initial therapy was not associated with a significant improvement in mortality (32% vs. 13%; P p.27). In summary, despite being intuitively logical and supported by multiple retrospective studies, no prospective study of VAP, nursing home acquired pneumonia, HCAP, or CAP has demonstrated a mortality benefit from broader-spectrum protocolized antibiotic regimens, despite efforts made to consistently decrease rates of inappropriate therapy to low levels. Thus, the support for a mortality benefit of aggressive broad-spectrum therapy for HCAP cannot even be extrapolated from studies of other types of pneumonia. Aggressive empirical therapy. The use of the broad-spectrum multiple-drug regimens discussed above can be considered aggressive empirical therapy. However, the only published study specifically addressing HCAP is a retrospective review of vancomycin dosing for patients with HAP and HCAP [64]. The authors compared dosing that was adjusted to achieve a serum trough level of 115 mg/ml, as recommended by the ATS-IDSA guidelines because of poor outcomes with standard dosing. The group that achieved trough levels greater than this threshold did not have a mortality benefit and had more adverse effects [65]. Early empirical therapy. The timing of appropriate antibiotic therapy has received significant attention. No study of HCAP has specifically examined the timing of antibiotic therapy. However, 2 large retrospective reviews of Medicare patients suggested a survival advantage when there was earlier provision of antibiotics [66, 67]. Many of these patients were likely to have HCAP. Although significant differences in mortality among patients receiving antibiotics in the first 4 8 h were documented, the trend toward increased mortality was heterogeneous, with some of the highest mortality rates found among those who received antibiotics in the first 2 h after presentation to the emergency department. More importantly, prospective studies of CAP guideline implementation have demonstrated that mortality is unchanged despite significant increases in the proportion of patients receiving antibiotics within 4 or 8 h [68]. The only prospective trial involving VAP did not show a difference in mortality if antibiotics were started empirically when VAP was suspected, compared with when the culture results were returned [69]. However, the study was limited to trauma patients for whom an attributable mortality due to VAP was unclear and VAP was less likely to have been caused by MDR pathogens. Duration of ventilation was increased for patients randomized to receive culture-directed treatment. Retrospective data from a medical ICU population suggest that a delay of 24 h in initiating therapy is associated with excess mortality [70]. The strongest evidence in favor of early antibiotic therapy is from a retrospective review of septic shock, in which every 1- hour delay in initiation of antibiotic therapy was associated with a 7.6% increase in mortality [4]. In this study, 37% of patients had pneumonia. Early de-escalation of empirical therapy. De-escalation has a variety of definitions. The most accepted definition is a decrease in the number of different antibiotics being used for treatment; however, de-escalation may also include switching to a narrower-spectrum agent, shortening the duration of therapy, or even ceasing the administration of antibiotics altogether when culture results are negative. Both before-after antibiotic treatment protocols included decreasing the number of empirical antibiotics (once culture results were known), as well as the duration of use [61, 62]. Although neither protocol was associated with a mortality benefit, the use of very-broad-spectrum antibiotics and de-escalation was associated with a decrease in the subsequent occurrence of colonization or infection with MDR pathogens [61, 62, 71]. Some evidence of antibiotic pressure was seen in the study by Soo Hoo et al. [62], in which 6 of the 7 imipenemresistant isolates occurred in patients given the more aggressive empirical regimen (which included imipenem). Two additional randomized trials are significant. Singh et al. [71] demonstrated that, among patients with HAP or VAP who have a persistently low clinical pulmonary infection score, the discontinuation of antibiotic therapy after 3 days was associated with a decrease in the percentage of pathogens that were MDR (14% vs. 38%; P p.017) and a trend toward mortality differences. Chastre et al. [72] demonstrated that patients with VAP randomized to receive 8 days of therapy had lower rates of emergence of MDR pathogens than did those who received 15 days of therapy (42.1% vs. 62%; P p.04). A randomized, controlled trial of diagnostic methods also demonstrated that, when fewer antibiotics were used, the 14-day mortality (16.2% vs. 25.8%; P p.022) and severity-adjusted 28-day mortality were decreased, although no differences in the emergence of MDR pathogens was demonstrated (61.3% vs. 59.8%; P 1.2) [73]. In summary, early de-escalation of therapy has an unclear association with decreased mortality. The strongest support comes from avoiding or discontinuing antibiotic therapy completely, rather than narrowing the spectrum or decreasing the number of antibiotics. Conversely, any type of de-escalation is associated with a decrease in the emergence of MDR pathogens. The major benefit appears to occur with a decrease in the overall duration of therapy, rather than de-escalation per se. Grading of Evidence On the basis of a review of the studies cited above, 83% of the members of this workshop agreed that the nature of the evidence available to support this statement was category II for the statement in general, with the remainder grading the evidence as category III (table 3). HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S69

16 Level of Support When voting on the support for this statement, 55% of the workshop members voted to accept the statement with some reservations, and 45% voted to accept the statement with major reservations. In comparison, of the 744 IDSA members who participated in the online survey, 56% voted to accept the statement completely, 36% voted to accept the statement with some reservations, 5% voted to accept the statement with major reservations, 2% voted to reject the statement with reservations, and 1% voted to reject the statement completely (figure 3). Discussion The difference between the voting of the workshop participants and that of the IDSA members is striking. The most likely explanation is an overestimation of the literature support for the concept of HCAP. Only 3 studies have specifically addressed HCAP [7, 8, 60]. Most of the other information is extrapolated from either HAP/VAP or CAP literature. The second major issue is the over-reliance on retrospective studies, which is particularly true for data on inappropriate initial empirical therapy, for which multiple retrospective studies consistently show excess mortality among patients receiving inappropriate initial empirical therapy [1]. The prospective trials of broad-spectrum empirical therapy with de-escalation do not demonstrate that providing appropriate initial antibiotics is sufficient to improve mortality [61, 62]. One explanation for this seemingly paradoxical finding between retrospective and prospective trials is that antibiotics for MDR pathogens may frequently be ineffective, despite being appropriate [1]. Another explanation may be that the patient s host response is unable to cure the pneumonia despite antibiotic therapy. Here, the difference between CAP and HAP, especially VAP, is likely to be great. Many VAP cases occur during a period of relative immunoparalysis after initial ICU admission for a critical illness [74]. In contrast, most CAP cases are characterized by a proinflammatory state. Although the pathogens associated with HCAP may resemble HAP and/or VAP, it is unclear whether the physiologic response will vary in the same way. Future Directions No prospective, randomized trial comparing appropriate versus inappropriate initial antibiotic therapy for HCAP has been performed. Therefore, the only information regarding the benefit of early appropriate initial therapy will have to come from studies of alternative empirical regimens, such as those for VAP. Given the wide discrepancy in the frequencies of MDR pathogens in HCAP cases in recent studies, this type of study is clearly needed. STATEMENT 4: HEALTH CARE ASSOCIATED BSIs REQUIRE EMPIRIC COVERAGE FOR MDR GRAM-NEGATIVE BACTERIA AND MRSA, AS WELL AS COVERAGE FOR FUNGAL PATHOGENS IN PATIENTS WITH SPECIFIC RISK FACTORS Rationale and Definition of Statement BSI is a common and potentially lethal complication of health care contact. A significant minority of hospitalized patients develop a BSI. Among these patients, mortality rates are high. This high mortality may be caused in part by the emergence of antimicrobial resistance in pathogens associated with the health care system. Such antimicrobial resistance increases the possibility of inadequate empirical antimicrobial therapy, which can delay the time until effective antimicrobial therapy is administered. The entity of health care associated BSI was first defined by Friedman et al. [6] as involving a positive culture result from a blood specimen that was obtained from a patient within 48 h after admission if the patient received intravenous therapy, wound care, or specialized nursing care or did any of the following: received self-administered intravenous medical therapy in the 30 days before the BSI; attended a hospital or hemo- Figure 3. Voting comparison for statement 3 ( Early aggressive, appropriate empiric treatment and de-escalation for HCAP reduces mortality and minimizes resistance ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Web-based survey; Summit refers to the Health Care Associated Infection Summit panel. HCAP, health care associated pneumonia. S70 CID 2008:47 (Suppl 2) Kollef et al.

17 dialysis clinic or received chemotherapy in the 30 days before the BSI; was hospitalized in an acute care hospital for 2 days in the 90 days before the BSI; or resided in a nursing home or long-term-care facility. A key finding of this study was that the prevalence of antimicrobial-resistant pathogens among patients with non-nosocomial health care associated BSI (i.e., BSI that did not originate in the hospital setting) resembled that among patients with nosocomial BSI. Thus, for the purposes of the present article, HAIs are defined as both nosocomial and nonnosocomial HAIs. Methods A PubMed search related to health care associated BSI was completed on 28 September The search terms health care associated, health care-associated, healthcare associated, and healthcare associated OR health care-associated OR health care associated OR healthcare-associated gave a total of 54,638 articles. The search terms blood stream infection, bloodstream infection, bacteremia, bloodstream infection, and blood stream infection, combined using the OR function, yielded a total of 27,839 articles. The search term ineffective therapy OR ineffective antibiotic therapy OR delayed antibiotic treatment OR delayed receipt of effective antimicrobial therapy OR inadequate antimicrobial treatment OR delay in effective therapy yielded a total of 24,230 articles. Combining the bloodstream infection search with the ineffective therapy search, using the AND function, resulted in a total of 232 articles. All these articles were reviewed; 13 were relevant to the statement. Evidence Health care associated status is a risk factor for ineffective antibiotic therapy of BSI. One study specifically focused on the impact of health care associated status on the likelihood of ineffective therapy for patients with BSI [75]. In this prospective, multicenter, cohort study of 466 adults with BSI, only 132 (28%) had community-acquired BSI. The most common pathogens in BSI were E. coli (14.2%) and MRSA (13.1%). Although the microbiological characteristics of nosocomial and non-nosocomial health care associated BSIs were similar, microbiological characteristics of both groups differed significantly from those of community-associated BSI. In multivariable logistic regression analysis, both health care associated (OR, 3.1; 95% CI, ) and nosocomial (OR, 4.3; 95% CI, ) status were independently associated with ineffective initial antibiotic therapy. Specific causes of BSI, including MRSA (OR, 1.7; 95% CI, ) and Enterococcus species (OR, 2.3; 95% CI, ), were also associated with ineffective initial therapy. Assessment of association between appropriate antibiotic therapy and mortality in patients with bacteremia. Studies of the association between inappropriate therapy and mortality among patients with bacteremia have yielded conflicting results. One recent article [76] systematically reviewed the published literature evaluating the association between inappropriate antibiotic therapy and mortality among patients with bacteremia. The authors found that 51 studies meeting their inclusion criteria exhibited significant heterogeneity in design, definition, measurement of variables, and statistics. Thirty-four studies (67%) measured the severity of illness, but only 6 (12%) specified when it was assessed. Only 8 studies (16%) defined inappropriate antibiotic therapy as that which was inactive in vitro against the isolated organism and was not consistent with current clinical practice recommendations and also distinguished between empirical and definitive treatment. McGregor et al. [76] identified key methodological recommendations to improve the validity and generalizability of future studies, including a robust, consistent definition of inappropriate therapy based on in vitro susceptibility data; separate consideration of empirical and definitive therapy; and appropriate statistical adjustment for the baseline severity of illness of the patient. Association between patient outcome and antibiotic therapy for BSI caused by MDR gram-negative pathogens. A recent meta-analysis of 16 peer-reviewed studies examined associations between ESBL production in Enterobacteriaceae species causing bacteremia, time to effective antibiotic therapy, and patient mortality [77]. Meta-analysis of crude RR demonstrated a significantly increased incidence of delay in effective therapy (pooled RR, 5.56; 95% CI, ; P!.001) and significantly increased mortality (pooled RR, 1.85; 95% CI, ; P!.001) in bacteremia caused by ESBL-producing bacteria. The meta-analysis was unable to evaluate adjusted mortality, because only 1 of the 16 included studies reported these data. A total of 7 additional reports were published after the enrollment period for the meta-analysis [78 84]. All were retrospective, and all but 1 was a single-center study [80]. Most [1, 5, 7, 10,78, 81, 82, 84] but not all [79, 83] of the 7 additional studies found an association between delayed effective therapy for BSI caused by MDR gram-negative pathogens and mortality. Consistent with the report by McGregor et al. [76], significant heterogeneity existed among the 7 studies in patient population, definitions of delayed antibiotic therapy, follow-up period, and statistical methodology. Moreover, establishing the risks of attributable mortality remains difficult. Using classification and regression tree (CART) analysis, Lodise et al. [81] evaluated the relationship between delayed appropriate antibiotic therapy and risk of 30-day mortality in 100 patients with nosocomial P. aeruginosa bacteremia. Delayed antibiotic therapy was defined using CART analysis as receipt of effective antibiotic therapy 152 h after the culture result was obtained. Mortality was significantly higher among patients with delayed appropriate antibiotic therapy than among patients whose therapy was not delayed (44% vs. 19%; P p HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S71

18 .008). Appropriate antibiotic therapy delayed 152 h was independently associated with resistance to 13 antibiotic classes (adjusted OR [AOR], 4.6; 95% CI, ; P p.001), chronic obstructive pulmonary disease (AOR, 5.4; 95% CI, ; P p.01), and 30-day mortality (OR, 4.1; 95% CI, ; P p.03) among patients with P. aeruginosa BSI. Tumbarello et al. [84] sought to identify the impact of inadequate initial antibiotic therapy (defined as initiation of treatment with active antimicrobial agents 172 h after collection of the first positive blood culture specimen) on 21-day mortality in 186 hospitalized patients with BSI caused by ESBL-producing organisms. Patients receiving inadequate treatment had a 3-fold increase in mortality, compared with the group receiving adequate treatment (59.5% vs. 18.5%; 95% CI, ; P!.001). In multivariate analysis, the significant predictors of mortality were inadequate initial antimicrobial therapy (OR, 6.28; 95% CI, ; P!.001) and unidentified primary infection site (OR, 2.69; 95% CI, ; P p.004). The antibiotic regimens most frequently classified as inadequate were based on oxyimino cephalosporin or fluoroquinolone therapy. Using a multicenter, nested, case-control study, Hyle et al. [80] evaluated the association of inadequate initial antimicrobial therapy with mortality in 187 patients with BSI caused by ESBL-producing organisms. Initial antimicrobial therapy was defined as inadequate when there was 148 h between the time a culture specimen was obtained and the initiation of therapy with an agent to which the infecting organism was susceptible. Infection with MDR ESBL-producing E. coli or Klebsiella species (AOR, 14.58; 95% CI, ) and health care acquired infection with ESBL-producing E. coli or Klebsiella species (AOR, 4.32; 95% CI, ) were independent risk factors for inadequate initial antimicrobial therapy, and inadequate initial antimicrobial therapy was an independent risk factor for mortality among patients with nonurinary infection with ESBLproducing E. coli or Klebsiella species (AOR, 10.04; 95% CI, ). Anderson et al. [78] used multivariable logistic regression to identify predictors of all-cause in-hospital mortality among 60 patients with bacteremia due to ceftazidime-resistant Klebsiella pneumoniae. Only 72% of patients received effective therapy within 5 days after the diagnosis of BSI. Delay in the initiation of effective therapy for 172 h after diagnosis of BSI was an independent predictor of mortality (OR, 3.32; 95% CI, ; P p.04). Micek et al. [82] evaluated 305 patients with P. aeruginosa BSI to determine whether the administration of appropriate initial antimicrobial treatment was associated with a better clinical outcome and to examine the relationship between the empirical administration of combination antimicrobial therapy for gram-negative pathogens and appropriate treatment for P. aeruginosa BSI [82]. In-hospital mortality was statistically greater for patients receiving inappropriate initial antimicrobial treatment than for patients receiving appropriate initial treatment (30.7% vs. 17.8%; P p.018 ). Multiple logistic regression analysis identified inappropriate initial antimicrobial treatment (AOR, 2.04; 95% CI, ; P p.048) as an independent predictor of in-hospital mortality. An appropriate initial antimicrobial regimen was administered more often to patients receiving empirical combination antimicrobial treatment for gram-negative bacteria than to those receiving empirical monotherapy (79.4% vs. 65.5%; P p.011). Two studies found no increased risk with delayed effective therapy for BSI caused by MDR gram-negative pathogens. Osih et al. [83] assessed the effect of appropriate empirical therapy on in-hospital mortality and length of stay among 167 patients with P. aeruginosa BSI. Adequate empirical antibiotic therapy was defined on the basis of in vitro susceptibility testing from 8 h before the first positive blood culture to the time the susceptibility results were known. After adjustment for age, severity of illness, and time at risk, appropriate empirical antibiotic therapy was not significantly associated with mortality (OR, 0.96; 95% CI, ; P p.58). Deal et al. [79] sought to identify predictors of in-hospital mortality among 124 patients with bacteremia caused by Enterobacter or Citrobacter species from 1998 through Appropriate empirical antibiotic therapy was administered to three-quarters of the patients and was similar among survivors and nonsurvivors (74% vs. 81%; P p.51). An important limitation to this investigation was sample size. Association between patient outcome and antibiotic therapy for MRSA bacteremia. Two meta-analyses involving staphylococcemic patients have shown that the mortality rate among patients with MRSA bacteremia was significantly greater than that among patients with MSSA bacteremia [85, 86]. Using data from patients from 30 studies, Cosgrove et al. [85] showed that mortality was significantly higher among patients with MRSA bacteremia than among patients with MSSA bacteremia (36% vs. 23%; RR, 1.42; 95% CI, ; P!.001). Whitby et al. [86] reviewed 9 studies of nosocomial S. aureus bacteremia published in In this analysis, the RR of death also was significantly higher among patients with MRSA bacteremia (29% vs.12%; RR, 2.12; 95% CI, ; P!.001). Several investigations have sought to quantify the impact of delayed effective therapy on outcomes for patients with MRSA bacteremia [87 90]. Results have varied, with 2 studies finding no difference in mortality, and 2 studies finding higher mortality rates among patients with MRSA bacteremia receiving delayed antibiotic therapy. Roghmann et al. [90] retrospectively evaluated 132 episodes in 128 patients with MRSA bacteremia to estimate the impact of delayed initiation of vancomycin on clinical outcomes. Patients with MRSA bacteremia were significantly less S72 CID 2008:47 (Suppl 2) Kollef et al.

19 likely to receive effective antibiotic therapy during the first 48 h of hospitalization (45% vs. 98%; P!.01) than were patients with MSSA bacteremia. However, this ineffective empirical therapy was not significantly associated with an increased mortality risk (RR, 0.82; 95% CI, ) and did not change significantly when adjusted for age, occurrence of sepsis, or nosocomial infection. Kim et al. [88] evaluated 238 retrospectively identified patients with MRSA bacteremia who received vancomycin or ineffective therapy. Using a propensity-matching case-control design to adjust for confounding introduced by the clinician s choice of antibiotic, these investigators compared the outcomes for patients with MRSA bacteremia who received inappropriate empirical therapy with those of control patients with a similar score but who received vancomycin. In the matched case-control analysis of 50 propensity score matched pairs with MRSA bacteremia, inappropriate empirical antibiotic therapy was not associated with a statistically significant difference in mortality (OR, 1.15; 95% CI, ). By contrast, 2 investigations found higher mortality among patients receiving delayed effective therapy for S. aureus bacteremia. Using CART analysis, Lodise et al. [89] evaluated the impact of delayed effective therapy on 167 retrospectively identified patients with nosocomial S. aureus bacteremia. The breakpoint between delayed and early therapy by use of CART analysis was h. In a multivariate analysis, delayed treatment was found to be an independent predictor of infection-related mortality (OR, 3.8; 95% CI, ; P p.01) and was associated with longer hospital stay when compared with early treatment (20.2 vs days; P p.05). The authors concluded that delay of therapy has deleterious effects on clinical outcomes and underscores the importance of early appropriate therapy. Similar conclusions were reached by Khatib et al. [87], who found that, in a cohort of 342 retrospectively identified patients with S. aureus bacteremia, the time to effective antibiotic therapy was longer for MRSA-infected patients than for MSSAinfected patients (25.5 vs. 9.6 h; P!.0005) and all-cause mortality was higher among patients receiving inappropriate therapy than among those receiving appropriate therapy (35.0% vs. 20.9%; P p.02). Association between patient outcome and antimicrobial therapy for fungal BSI in patients with specific risk factors. Two studies evaluated the risk of delayed effective therapy in fungemic patients. Garey et al. [91] evaluated the relationship between treatment delay and mortality in 230 retrospectively identified patients with Candida BSI. Although the mortality was the lowest among patients who began therapy on day 0 (15%), day 1 (24%), day 2 (37%), or day 3 or later (41%) ( P p.0009 for trend), only 40% of patients received antifungal therapy within the first day. By multivariate modeling, increased time (per day) to administration of fluconazole was independently associated with mortality (AOR, 1.5; 95% CI, ; P p.0138). In the second study, Morrell et al. [92] evaluated 157 candidemic hospitalized patients to identify the influence of delayed empirical antifungal treatment on clinical outcome. By multivariable analysis, administration of antifungal treatment 112 h after the first positive blood culture specimen was drawn (AOR, 2.09; 95% CI, ; P p.018 ) was independently associated with in-hospital mortality. Of note, only 5.7% of patients received antifungal therapy within 12 h after the initial positive result of blood culture. Investigators in both of these studies concluded that delay in initiation of fluconazole therapy for hospitalized patients with candidemia had a significant impact on mortality. Delayed treatment of Candida BSI could be minimized by the development of more-rapid diagnostic techniques for the identification of Candida BSI or through increased use of empirical antifungal treatment for selected patients at risk for fungemia. Grading of Evidence On the basis of a review of the studies cited above, the workshop members considered the nature of the evidence supporting this statement to be category II (67% of votes) or category III (33% of votes) (table 3). Level of Support When voting on the support for this statement, 9% of the summit participants voted to accept the statement completely, 73% voted to accept the statement with some reservations, and 18% voted to accept the statement with major reservations. In comparison, of the 744 IDSA members who participated in the online survey, 25% voted to accept the statement completely, 38% voted to accept the statement with some reservations, 16% voted to accept the statement with major reservations, 17% voted to reject the statement with reservations, and 4% voted to reject the statement completely (figure 4). Discussion This statement is critically important, given the growing problems of sepsis, bacteremia [93], and antimicrobial resistance [94]. The majority of the studies reviewed for this statement support the assertion that delayed appropriate antibiotic therapy is associated with higher mortality among patients with BSIs. Although none of the studies were able to accurately establish causal relationships between delayed appropriate antimicrobial therapy and increased mortality and most suffered in one way or another from methodologic limitations [76], their conclusions are generally consistent with current treatment guidelines for other HAIs [1] and with previous reports evaluating the impact of such treatment delays for patients with sepsis [3]. As evidenced by the results of the IDSA membership poll related to this statement, the important influence of time HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S73

20 Figure 4. Voting comparison for statement 4 ( Health care associated BSIs require empiric coverage for MDR gram-negative bacteria and MRSA, as well as coverage for fungal pathogens in patients with specific risk factors ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Web-based survey; Summit refers to the Health Care Associated Infection Summit panel. BSI, bloodstream infection; MDR, multidrug resistant; MRSA, methicillin-resistant Staphylococcus aureus. to administration of effective antimicrobial therapy on the clinical outcome also makes intuitive sense to many clinicians. The primary rate-limiting steps to effective antimicrobial therapy for health care associated BSI remain diagnostics and susceptibility testing. Even when guided by local antimicrobial susceptibility, empirical therapy often becomes little more than an educated guess. Until diagnostic strategies emerge to provide real-time, point-of-care information on the identification and susceptibility of a bloodstream pathogen, clinicians will be forced to make important decisions about initial antibiotic selection without the luxury of definitive data. In this light, observations from these studies are important. Among patients with BSI caused by gram-negative pathogens, early effective therapy was usually associated with reduced mortality, and the likelihood of accomplishing early effective therapy was higher when combination empirical antimicrobial therapy was employed. Obviously, clinicians should consider both the risks and benefits of adding a second antibiotic often an aminoglycoside to an empirical regimen to treat gramnegative pathogens in individual patients. However, the predominance of MRSA as a cause of health care associated bacteremia, the availability of an FDA-approved agent for the treatment of S. aureus bacteremia and right-sided endocarditis (e.g., daptomycin), and the prospects of several anti-mrsa agents in late stages of clinical development emphasize the need for appropriately designed clinical studies to better address this important issue. Significant controversy remains over the role of vancomycin for treatment of MRSA bacteremia whether empirical or targeted [95, 96]. Finally, the emerging importance of fungi as a cause of BSI and sepsis is a potentially important change to consider in the management of BSI. For example, in an evaluation of the hospital discharge records of 110 million cases of sepsis in the United States over 22 years, there was an annualized increase in the incidence of sepsis of 8.7%, from 164,000 cases (82.7 per 100,000 population) to nearly 660,000 cases (240.4 per 100,000 population). During this time, the rate of sepsis due to fungal organisms increased by 207% [93]. Given the increasing importance of fungemia and the suggestion that early empirical antifungal therapy may reduce mortality among patients with this infection, further studies are clearly needed to help determine which patients, if any, should receive empirical antifungal treatment. Future Directions Future directions discussed by the summit members reflected many of the limitations indicated by McGregor et al. [76]. Appropriately designed epidemiologic studies with rigorous attention to important design details are required, including a consistent definition of inappropriate therapy based on in vitro susceptibility data, separate consideration of empirical and definitive therapy, and appropriate statistical adjustment for the baseline severity of illness of the patient. The need for morerapid diagnostic tests was emphasized. Finally, until such bedside diagnostic technologies are available, additional studies to identify patients at risk for colonization or infection with MDR pathogens especially the fungi are required to best balance the dual needs for judicious and effective empirical antimicrobial therapy for patients with BSIs. STATEMENT 5: INITIAL APPROPRIATE ANTIMICROBIAL THERAPY AND SOURCE CONTROL ARE THE MOST IMPORTANT DETERMINANTS OF OUTCOME IN SEVERE SEPSIS AND SEPTIC SHOCK Rationale and Definition of Statement Severe sepsis and septic shock are commonly encountered consequences of severe infection, both community acquired and hospital acquired [97]. Sepsis and its adverse sequelae, shock and organ dysfunction, are currently the 10th leading cause of death in the United States and one of the most common causes of death in the noncoronary ICU [97, 98]. Martin et al. [93] found that the incidence of sepsis in the United States has S74 CID 2008:47 (Suppl 2) Kollef et al.

21 dramatically increased over a 22-year period ( ), from 82.7 cases per 100,000 population to cases per 100,000 population, for an annualized increase of 8.7%. The incidence of severe sepsis is predicted to continue to rise in the next 2 decades [99]. To prevent the progression to organ dysfunction and other adverse sequelae of severe sepsis and septic shock, it is important to identify the patient with sepsis as early as possible and to institute effective therapy [97, 98, 100]. Representatives from 11 societies proposed a consensus guideline for severe sepsis and septic shock management [101], and they recommended the following actions: the early administration of effective antibiotics, source control when appropriate, provision of early goal-directed fluid resuscitation and vasopressor support when required, maintenance of adequate oxygenation and ventilation as necessary, use of physiologic steroid-replacement therapy for vasopressor-dependent patients with relative adrenal insufficiency, antithrombotic therapy when warranted and not contraindicated, and prevention of the various complications of critical illness [101]. The Surviving Sepsis Campaign recently published an update to these recommendations [102] that advises institution of early, effective antibiotic therapy; evaluation of the need for source control; early, goal-directed fluid resuscitation with either crystalloid or colloid norepinephrine or dopamine as needed to maintain a mean arterial blood pressure 65 mm Hg; dobutamine for patients with myocardial depression; packed RBCs to treat hemoglobin levels of!7.0 g/dl; use of lung-protective ventilatory support strategies when needed; weaning from ventilatory support by protocol and spontaneous breathing trials; prevention of complications of critical illness (insulin to maintain blood sugar levels at!150 mg/dl); deep vein thrombosis prophylaxis; stress ulcer prophylaxis; and use of sedation protocols [102]. The use of early, effective antibiotic therapy and source control, when indicated, has been considered the cornerstone of sepsis therapy. Conventional wisdom would suggest that appropriate (effective) antibiotic therapy is essential for an improved outcome. This hypothesis will never be clinically tested because it would be considered unethical to deprive a patient of timely or effective antibiotic treatment. Most of our current evidence in support of early, effective antibiotic therapy is based on retrospective outcome analysis comparing early, effective antibiotic therapy with initially ineffective antibiotic agents. Because the definition of sepsis has varied over time and, in the older literature, often required the presence of bacteremia and hypotension, it is difficult to make comparisons with some of the current data [103]. However, it is still apparent that, regardless of the definition, mortality rates improve with timely use of effective, appropriate antibiotic therapy. Other important aspects of antibiotic effectiveness relate to the timing of antibiotic therapy (early vs. delayed administration). Studies that are amenable to prospective, randomized clinical trials involve the use of single versus multiple antibiotic agents and the effects on outcome parameters. The adequacy of source control is also a difficult topic to evaluate with the published literature [104]. Few studies have commented on the adequacy of drainage or debridement. The adequacy of source control is evaluated primarily by retrospective review of the clinical course or by a post hoc adjudication committee. As in investigations evaluating the effectiveness of antibiotic therapy, ethical concerns preclude a study of adequate versus inadequate source control. Therefore, to answer both of these important questions, we must turn our attention to an analysis of the limited available literature, with emphasis on clinical trials that have evaluated or commented on the adequacy of either antibiotic therapy or source control, as it relates to outcome for patients with sepsis. Methods A PubMed search to identify studies related to appropriate antibiotic use and source control in sepsis outcome was performed on 14 September The search was limited to 1996 through September 2007, and all searches were restricted to the English language, adult humans, and full-text articles. The text words adequate antibiotics yielded 18,240 articles, sepsis/ septic shock yielded 14,423 articles, and survival/mortality yielded 93,638 articles. When the search for adequate antibiotics was combined with that for sepsis/septic shock, the result was 1209 articles. The combination of the search for adequate antibiotics and the search for survival/mortality yielded 7744 articles. When the search was limited to adequate antibiotics AND sepsis/septic shock AND survival/mortality, the result was 680 articles. A review of the titles and abstracts of the 680 articles resulted in 16 articles that were relevant to the statement. In addition, an evidence-based review from the Surviving Sepsis Campaign on the topics of adequate antimicrobial therapy and adequate source control was examined along with the references from these articles [100, 104]. Evidence To evaluate evidence concerning appropriate antibiotic therapy for severe sepsis and septic shock, we must first acknowledge that no clinician would intentionally give a patient ineffective antibiotic therapy. It is convention to classify appropriate or effective antibiotic therapy on the basis of the culture and susceptibility results; however, not all patients with sepsis will have positive culture results. Most large clinical trials report that approximately one-third of patients have positive blood culture results and approximately one-quarter of patients will have no positive culture result of any type [105]. It is also disappointing to find that several large epidemiologic studies and clinical trials involving patients with sepsis do not give any data concerning HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S75

22 the administration of effective antibiotic therapy or the adequacy of source control [106, 107]. Other important questions (related to the use of culture and susceptibility results) surround the ability of in vitro culture results to reflect in vivo effects of antibiotics, as well as the debate about the thresholds differentiating a true pathogen from a colonizing organism. A retrospective review of 612 patients with bacteremia caused by gram-negative bacteria demonstrated a significant reduction in mortality when appropriate antibiotic therapy was administered [108]. Fish [109] reviewed mortality differences between patients receiving appropriate antimicrobial therapy and patients receiving inappropriate antimicrobial therapy in 11 studies and demonstrated an association between a significant reduction in mortality and appropriate antibiotic treatment. Furthermore, in the Surviving Sepsis Campaign, the 11 societies used a modified Delphi method to define the role of effective antimicrobial therapy in the management of severe sepsis and septic shock [100]. The consensus committee concluded that prompt institution of effective antimicrobial therapy is one of the most important predictors of outcome; unfortunately, most of the evidence in support of their recommendations reflects category III, IV, or V evidence (table 3). Using the same modified Delphi method to assess source control in the management of severe sepsis and septic shock, the same group concluded that source control represents a key component for successful sepsis management and should be used when indicated [104]. Source control includes drainage of infected fluids, debridement of infected soft tissues, and removal of infected devices or foreign bodies. Source control should correct anatomic derangements resulting in ongoing contamination and restore function [104]. In Spain, Garnacho-Montero et al. [110] conducted a prospective cohort study of 406 critically ill patients with sepsis in a tertiary care hospital, to determine the impact of effective, empirical antibiotic therapy on early, 28-day, and 60-day mortality. The administration of inadequate antibiotic therapy was associated with an RR of 1.55 (95% CI, ) for increased mortality, compared with effective antibiotic therapy. An observational, prospective cohort study of 3413 patients with BSI also demonstrated an increased mortality RR (1.6; 95% CI, ) associated with the administration of inadequate antibiotic therapy [111]. Ineffective antibiotic therapy was found to increase all-cause mortality (52.1% vs. 23.5%; RR, 2.22; 95% CI, ; P!.001) and infection-related mortality (42% vs. 17.7%; RR, 2.37; 95% CI, ; P!.001) in 2000 consecutive ICU patients included in a prospective, observational, cohort study [3]. In this study, the use of ineffective antibiotics was greater in the setting of nosocomial infection with or without prior antibiotic therapy. A prospective study of 707 patients with bacteremia and/or fungemia evaluated the impact of effective versus ineffective antimicrobial therapy administered initially, after results of cultures were obtained, and after the susceptibility results were available; this study noted an increase of up to 3.18 in the RR for death, when effective therapy was compared with ineffective therapy at all time points [112]. The investigation of innovative therapies to improve the outcome of severe sepsis and septic shock seems a perfect opportunity to evaluate the relationship between effective antibiotic therapy and outcome. Unfortunately, this important variable is often not assessed as part of a trial. Typically, the evaluation of effective antibiotic and/or source control therapy involves the use of a post hoc adjudication committee to evaluate the culture and susceptibility results in relation to the antimicrobial agents administered. A multicenter, prospective, randomized, double-blind, controlled trial of high-dose intravenous immunoglobulin (IVIG), in addition to antibiotic therapy, given to patients with sepsis undergoing surgery, found that overall mortality increased from 20.4% with effective antibiotic therapy to 87.5% with ineffective therapy [113]. In the PROWESS trial, drotrecogin alfa (activated) significantly reduced mortality among patients with severe sepsis, but 148 patients (8.8%) in the intention-to-treat population (patients given drotrecogin alfa [activated], n p 74; patients given placebo, n p 74) received inadequate antibiotic therapy [114]. In a subsequent analysis, it was noted that the observed mortality rates for patients given drotrecogin alfa (activated) and patients given placebo who did not receive adequate antibiotic therapy (29.7% and 43.2%, respectively) were higher the observed mortality rates for those who received adequate antiinfective therapy (24.2% and 29.6%, respectively) [115]. The provision of adequate antimicrobial treatment, when evaluated, has been quite variable, with reported values ranging from 75% [116] to 88% [117]. The timing of antibiotic administration has also been found to be an extremely important factor in outcome. A retrospective cohort study of 2731 patients with hypotension and sepsis found that mortality increased 7.6% for each 1-hour delay in antibiotic administration after the onset of hypotension [4]. When antibiotics were administered within 30 min after the onset of hypotension, the survival rate was 82.7%, but it fell to 42% when the antibiotics were delayed 6 h after the onset of hypotension. To improve the timely administration of antibiotics and other diagnostic and therapeutic aspects of sepsis management, clinicians have incorporated recommended therapies into bundles of care [118]. Nguyen et al. [118] conducted a 2-year, prospective, observational cohort study of 330 patients who presented to the emergency department with severe sepsis or septic shock, and they found a significant reduction (from 20.8% to 39.5%; P!.01) in mortality when bundles were used and, specifically, when antibiotics were administered within 4 h after presentation (OR, 0.38; 95% CI, ; P p.03). Adequacy of source control is evaluated even less frequently S76 CID 2008:47 (Suppl 2) Kollef et al.

23 than is antimicrobial therapy. Among patients requiring surgery for source control in a multicenter, prospective, randomized, double-blind, placebo-controlled trial of platelet-activating factor acetylhydrolase for treatment of severe sepsis and septic shock, adequate source control occurred in 190% of both treatment groups [117]. In contrast, in the PROWESS trial of drotrecogin alfa (activated) for severe sepsis, the initial sourcecontrol procedure was judged to be adequate in only 90 (50.8%) of 177 patients receiving drotrecogin alfa (activated) and 86 (47.3%) of 182 patients receiving placebo. Initial source control was inadequate for 38 (21.5%) of 177 patients receiving drotrecogin alfa (activated) and 51 (28.0%) of 182 patients receiving placebo and was indeterminate for 49 (27.7%) and 45 (24.7%), respectively. In patients with adequate or inadequate source control, no reduction in mortality was noted for drotrecogin alfa (activated). In patients with indeterminate source control, drotecogin alfa (activated) was associated with a reduction in mortality (37.0% vs. 56.6%; OR, 0.65; 95% CI, ; absolute risk reduction, 19.6%) [119]. Grading of Evidence On the basis of a review of the 12 studies cited above, 67% of the workshop members agreed that the nature of the evidence available to support this statement was category II, whereas 17% voted category III, and 17% voted category V (table 3). Level of Support When voting on the support for this statement, 9% of the summit participants voted to accept the statement completely, 82% voted to accept the statement with some reservations, 9% voted to accept the statement with major reservations, and 0% voted to reject the statement. In comparison, of the 744 IDSA members who participated in the online survey, 50% voted to accept the statement completely, 35% voted to accept the statement with some reservations, 10% voted to accept the statement with major reservations, 5% voted to reject the statement with reservations, and 0% voted to reject the statement completely (figure 5). Discussion Consideration of the statement concerning the importance of effective antibiotics and source control in outcomes for patients with severe sepsis and septic shock caused the vast majority (180%) of the summit participants to support the statement either completely or with some reservations. This conclusion was reached by the participants despite the lack of multicenter, prospective, randomized, placebo-controlled, double-blind clinical trials substantiating the importance of effective antibiotic therapy in outcomes for patients with severe sepsis and septic shock. Despite this lack of evidence, most clinicians believe that administration of early, effective antimicrobial therapy and source control, when indicated, are among the key components of management of severe sepsis and septic shock. Other key components of effective sepsis management include fluid resuscitation, restoration and maintenance of hemodynamic function, support of oxygenation and ventilation as necessary, and prevention of the complications of critical illness [97, 98, 101, 102]. Ethical considerations and common sense prohibit conducting clinical trials to establish the key role for effective antibiotics and source control in the management of severe sepsis. It is also impossible to establish the relative value of one key component compared with another. The various components of effective sepsis management may be viewed as links in a chain, and the chain is only as strong as its weakest link. Therefore, without provision of all necessary aspects of sepsis management, the outcome will be less than optimal. Unfortunately, technology has not progressed to the point where the clinician can rapidly diagnose the microbial cause of the infection leading to sepsis or determine the susceptibility of the organism at an early point in time and thus enable provision of effective antibiotics at the time of diagnosis. Even the diagnosis of sepsis is often not confirmed until the results of cultures are available for review. This process typically takes hours to days, and, to date, is not available during the golden hour of initial management when the best outcome can be expected [4]. The use of markers for sepsis (e.g., procalcitonin levels, soluble triggering receptors expressed on myeloid cells, and peptide nucleic acid fluorescence in situ hybridization) and other techniques for early diagnosis may improve the diagnosis of sepsis, but we still lack an early indicator of organism susceptibility to various antibiotics [ ]. For now, clinicians have directed their attention to the implementation of sepsis bundles, to ensure the early administration of effective sepsis therapy, including antibiotics, to achieve the best possible outcomes for patients [118]. Future Directions Although not necessarily the major determinant of outcome in severe sepsis and septic shock, early, effective antibiotic therapy and source control, when indicated, are among the key components necessary for optimal outcome. Future efforts will likely focus on uses of advanced diagnostic techniques to identify specific bacterial and fungal pathogens, such as fluorescence in situ hybridization and PCR. These tests will help facilitate early diagnosis of sepsis and will direct more-appropriate early therapy. In addition, the efficacy of antibiotic treatment should be more definitively investigated in terms of the MIC of the microbe and the blood level and, potentially, the mechanism of action of the antibiotic. Relying only on the categories susceptible or resistant may not adequately define appropriate antibiotic treatment. This has particular relevance to antibiotics such as vancomycin and aminoglycosides. This strategy HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S77

24 Figure 5. Voting comparison for statement 5 ( Initial appropriate antimicrobial therapy and source control are the most important determinants of outcome in severe sepsis and septic shock ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Webbased survey; Summit refers to the Health Care Associated Infection Summit panel. not only will result in better sepsis outcome but also will assist with antibiotic stewardship and potentially minimize the development of bacterial resistance. STATEMENT 6: VANCOMYCIN IS OBSOLETE FOR TREATING MRSA INFECTIONS Rationale and Definition of Statement Vancomycin has been the workhorse antimicrobial for the treatment of MRSA infections for 140 years. In the past decade, the prevalence of hospital-associated MRSA infections has reached 64% in most US hospitals [123]. In addition, there has been a virtual explosion of community-onset MRSA infections among young, healthy individuals in a wide variety of situations, including high school, college, and professional football teams; prisons; and so forth. Although vancomycin was prescribed sporadically and infrequently years ago, its use has increased exponentially over the past decade. As a consequence, there is increasing evidence that vancomycin is not currently as effective as it once was; this evidence results from frank treatment failures as well as growing concern related to the emergence of various types of vancomycin resistance. The current epidemics of hospital-associated MRSA and community-onset MRSA infections have developed rapidly, and there are no concrete guidelines addressing the current problems associated with treatment of MRSA infections. The purpose of the current investigation is to examine the mounting evidence regarding treatment failures and reduced in vitro activity of vancomycin against MRSA. Methods A PubMed database search to identify studies related to vancomycin was concluded on 26 September The search term vancomycin yielded 13,064 articles. Vancomycin limited to the English language resulted in 11,528 articles and, when combined with last ten years, yielded 4181 articles. When these elements were combined with human, 3166 articles were found. Further narrowing of the field was accomplished by combining the elements failure (162 articles), Staphylococcus aureus (74 articles), and MRSA (30 articles). Excluding case reports yielded 10 articles. Finally, 3 abstracts on in vitro susceptibility were added from abstracts from the annual meetings of the IDSA and the Interscience Conference on Antimicrobial Agents and Chemotherapy. Evidence Changes in the susceptibility of MRSA to vancomycin. There have been 3 studies performed in the United States that evaluated the in vitro susceptibility of MRSA strains to vancomycin over time, with the objective of identifying trends in the susceptibility of MRSA to vancomycin. In the first study, the MIC 90 values for vancomycin among MRSA strains were compared in vitro at M. D. Anderson Hospital across a 20- year period; 25 strains from 1985 and 28 strains from 2004 were examined. This study demonstrated that the MIC 90 increased from 0.2 mg/ml to 2.0 mg/ml during this nearly 20- year time span [124]. The second study compared the in vitro susceptibility of blood isolates of MRSA in 2002 with that of blood isolates of MRSA in 2005 at the New England Medical Center (Boston, MA) and demonstrated a dramatic increase in the MICs for vancomycin (figure 6) [125]. The third study was an in vitro investigation of MICs for vancomycin among 945 strains of S. aureus from 2000 and 1418 strains of S. aureus from In 2000, 79.9% of strains had vancomycin MICs of 0.5 mg/ml, and 19.9% ( P!.01) had vancomycin MICs of 1.0 mg/ml. In contrast, in 2004, 28.8% had vancomycin MICs of 0.5 mg/ml, whereas 70.4% ( P!.01) had vancomycin MICs of 1.0 mg/ml [126]. According to this study, a marked increase in vancomycin MICs was apparent for MRSA isolates in Los Angeles from 2000 through Clinical failure of vancomycin in patients with bacteremia caused by MRSA strains with MICs of 4 mg/ml. An observational series of 14 case reports of clinical failures of vancomycin for treatment of bacteremia caused by MRSA strains with MICs of vancomycin 4 mg/ml were compiled [127]. These S78 CID 2008:47 (Suppl 2) Kollef et al.

25 Figure 6. In vitro comparison of vancomycin MICs at the New England Medical Center (Boston, MA), in 2002 and Adapted from [125]. failures were the primary evidence compelling the Clinical and Laboratory Standards Institute to lower the vancomycin-susceptible MRSA MIC breakpoint to 2 mg/ml. Vancomycin-resistant MRSA. Absolute resistance of MRSA strains has been described [128]. These strains are called vancomycin-resistant MRSA and have been defined as strains with MIC of 116 mg/ml for vancomycin. Thus far, 7 vancomycin-resistant MRSA strains have been described in Japan and the United States [129]. Vancomycin-intermediate MRSA (VISA) and heteroresistant VISA (hvisa). The first descriptions of emergence of vancomycin-intermediate strains of MRSA [130] or glycopeptide-intermediate S. aureus [131] were from Japan, and these are defined as having MICs of 4 8 mg/ml [130]. Soon thereafter, clinical failures of vancomycin for patients with MRSA bacteremia caused by vancomycin-intermediate strains were reported. In a retrospective study in Australia, 76% of 25 patients with MRSA bacteremia who were given treatment with vancomycin experienced failed therapy, as defined as persistence of bacteremia for 17 days. Interestingly, these MRSA strains were relatively susceptible to vancomycin, with MICs of 2 4 mg/ml [132]. Although these strains were defined as being VISA strains, they contained populations of microbes that were resistant to vancomycin. Because of the heterogenous population of MRSA, these strains are called hvisa. In the laboratory, demonstration of heteroresistance requires a large inoculum of 10 7 MRSA organims per ml because 1 in 100,000 bacteria is, in fact, resistant to vancomycin. It likely has clinical relevance for cases in which the load of MRSA is high, as one might expect in a large abscess, necrotizing fasciitis, consolidative pneumonia, bacteremia, and endocarditis. Because most clinical laboratories are standardized to use inocula of 10 5 MRSA organisms for susceptibility testing, newer methods must be developed to alert clinicians of this phenomenon. Failure of vancomycin in bacteremia caused by hvisa. A retrospective study of isolates from all patients with MRSA bacteremia ( n p 53) in a hospital in Australia evaluated a 12- month period (July 2001 June 2002), with the objective of identifying the prevalence of hvisa and the outcomes for these patients given treatment with vancomycin [133]. No VISA isolates were recovered; however, 5 (9.4%) of 53 MRSA isolates were heteroresistant to vancomycin. Patients infected with hvisa were more likely to have high bacterial loads (P p.001), compared with patients infected with vancomycin-susceptible MRSA, and patients with hvisa infections were more likely to experience a failure of vancomycin treatment (P!.001), compared with patients infected with vancomycin-susceptible MRSA [133]. Failure rate of vancomycin treatment as a function of rising MICs in patients with infection caused by vancomycin-susceptible MRSA. A total of 122 S. aureus isolates, 63 of which were MRSA with vancomycin MICs of mg/ml, from 87 patients given treatment with vancomycin were analyzed. Of the 87 patients, 45 had no clinical or bacteriological response to vancomycin. Among the 36 clinically evaluable patients infected with S. aureus strains that had the accessory gene regulator (agr) group II polymorphism, 31 had an infection that failed to respond to vancomycin, whereas only 5 had an infection that responded successfully to vancomycin. There was a significant association between vancomycin treatment failure (45 of 63) and MIC increase ( P p.004) (figure 7) [134]. Failure of vancomycin as a function of the rapidity of bacterial killing. This investigation analyzed isolates ( n p 30) from patients with bacteremia in 24 US hospitals, with the objective of correlating clinical failure with in vitro vancomycin susceptibility and bactericidal activity. For MRSA isolates with vancomycin MICs 0.5 mg/ml, vancomycin was 55.6% successful in the treatment of bacteremia, whereas vancomycin was only 9.5% ( P p.02) effective in cases in which MRSA MICs for vancomycin were 1 2 mg/ml. In addition, the failure rate for vancomycin was 100% if!4.71 log of bacteria were killed in 72 h ( n p 9); 77% if log of bacteria were killed in 72 h ( n p 13); and 50% if log of bacteria were killed in a 24- Figure 7. Clinical failures of vancomycin treatment for vancomycinsusceptible Staphylococcus aureus: the role of MICs. Data are from [134]. HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S79

26 h period ( n p 8). The differences between treatment groups were statistically significant ( P p.05) [135]. Vancomycin tolerance among MRSA isolates. Another cause of failure of antimicrobial treatment is the phenomenon of bacterial tolerance, which is defined as a minimum bactericidal concentration (MBC)/MIC ratio of 32 or an MBC of 16. Of 207 evaluated strains of S. aureus, 102 were MRSA; 14.7% of wild-type MRSA demonstrated tolerance, whereas 69.3% of hvisa and 100% of VISA isolates demonstrated tolerance. Thus, even large doses of vancomycin may not reach bactericidal blood and tissue levels sufficient to kill tolerant strains of MRSA [136]. Does increasing the dose of vancomycin to achieve serum trough levels of 115 mg/ml increase efficacy, or does it increase nephrotoxicity? The rationale for this study was a perceived increase in vancomycin treatment failures for infections caused by vancomycin-susceptible MRSA strains with high MICs and the general practice to recommend a higher vancomycin target trough level of mg/ml, in an effort to increase efficacy. However, there are no data regarding potentially increased renal toxicity associated with these higher doses. In a prospective cohort study of patients with MRSA infections ( n p 95), investigators sought to correlate the distribu- tion of vancomycin MICs and treatment outcomes with trough levels at least 4 times the MIC. There was no nephrotoxicity when trough levels were!15 mg/ml. However, 11 (12%) of 63 patients developed nephrotoxicity with trough levels 15 mg/ ml. Multivariate analysis implicated concomitant nephrotoxins, such as aminoglycosides and amphotericin B. In the hightrough-level group, only 2% (in the absence of nephrotoxins) developed nephrotoxicity [137]. Of the 95 patients in the study, 51 (54%) were infected with high-mic strains and had pneumonia (77%) and/or bacteremia. An initial response rate of 74% was achieved when the target trough level was attained, irrespective of MIC. However, despite achieving the target trough level, the group infected with high-mic strains had fewer end-of-treatment responses (24 [62%] of 39 vs. 34 [85%] of 40; P p.02) and higher infection-related mortality (11 [24%] of 51 vs. 4 [10%] of 44; P p.16), compared with the group infected with low-mic strains. Infection with a high- MIC strain ( P p.03) and high APACHE II score ( P p.009) were independent predictors of poor response in multivariate analysis. Nephrotoxicity occurred only in the high-trough-level group (11 [12%] of 63); this was significantly predicted by concomitant therapy with other nephrotoxic agents. A high prevalence of clinical MRSA strains with elevated vancomycin MICs (2 mg/ml) requires aggressive empirical vancomycin dosing to achieve a trough level of 115 mg/ml. The rationale for this recommendation is obvious, yet there is little clinical experience with high vancomycin dosages, and toxicity becomes an important issue, as discussed below. Combination or alternative therapy should be considered for invasive infections caused by these strains. Will higher trough levels of vancomycin be associated with a higher incidence of renal toxicity? In a prospective review of patients with HAP, 43 patients were followed up for changes in creatinine clearance ( n p 43). Overall, there was a 25% de- crease in creatinine clearance among all patients receiving vancomycin. There was a 30% decrease in creatinine clearance among patients with a low trough level (5 15 mg/ml), compared with a 60% decrease in creatinine clearance among patients with high trough levels of 115 mg/ml ( P p.006) [65]. Grading of Evidence Of the workshop participants, 83% voted that the evidence to support the statement was category III, and 17% voted that it was category II (table 3). Level of Support Interestingly, 36% of the summit participants voted to accept the statement with some reservations, 36% voted to accept the statement with major reservations, 18% voted to reject the statement with reservations, and 9% voted to reject the statement completely. In comparison, of the 744 IDSA members who participated in the online survey, 1% accepted the statement completely, 9% accepted the statement with some reservations, 7% accepted the statement with major reservations, 38% rejected the statement with reservations, and 45% rejected the statement completely (figure 8). Discussion This statement is of key importance, given the emerging data regarding reduced susceptibility of MRSA to vancomycin and the increasing reports of failure of vancomycin in the treatment of clinical infections. Recognition of this problem is not yet widespread, as evidenced by the IDSA membership s diverse responses to this statement that vancomycin is obsolete in the treatment of MRSA infections. Summit participants responses were diverse yet more accepting of the statement. The quality of the evidence supporting emerging resistance to vancomycin and so-called MIC creep is not robust, largely because it reflects regional differences and is not yet on a national scale. Still, these small studies are compelling because the studies were done in several different geographical regions by independent investigators. There may be some bias, since many of the sites are large hospitals in densely populated regions of the United States where vancomycin use may be greater. In addition, some studies are retrospective and start with patients who experienced failure of vancomycin treatment. Not surprisingly, some of the isolates from these cases have reduced susceptibility to vancomycin. Still, all studies presented here support the contention that strains of MRSA are emerging S80 CID 2008:47 (Suppl 2) Kollef et al.

27 over time with reduced susceptibility to vancomycin and that this phenomenon is associated with or causes treatment failures. Thus, although the quality and size of these studies are not robust, it is clear that vancomycin susceptibility among MRSA isolates is changing rapidly, and these preliminary studies are providing early warning of larger problems to come. It is also true that very large doses of vancomycin may be necessary to achieve an area under the concentration-time curve/mic ratio for MRSA infections caused by strains with vancomycin MICs of 1 2 mg/ml [138]. Future Directions Future directions discussed by the summit members clearly involve additional studies. Prospective studies that evaluate the in vitro susceptibility of MRSA strains to vancomycin on regional and national scales are sorely needed. New assays to detect heteroresistance among clinical isolates of MRSA need to be developed and correlated with clinical outcomes. Finally, prospective studies that evaluate clinical responses of MRSA infections, in terms of susceptibility issues and vancomycin trough levels, should be done immediately. Until then, if vancomycin is to be used, these data mandate increased knowledge of the MICs and vancomycin trough levels, to ensure that patients are given appropriate treatment. Specifically, clinical laboratories need to provide clinicians with the actual vancomycin MIC of the MRSA strain, because clinical failure increases proportionally to the MIC, even among susceptible strains. STATEMENT 7: SERIOUS HAIs DUE TO SUSPECTED GRAM-NEGATIVE BACTERIA SHOULD BE TREATED EMPIRICALLY WITH DUAL COVERAGE THAT INCLUDES AN AMINOGLYCOSIDE Rationale and Definition of Statement The terminology of HAIs is rapidly permeating the classification of various infection types. Classification schemes for both BSIs and pneumonia have already been adopted. These classifications identify specific patients at risk, as well as treatment recommendations. HAIs can describe a wide variety of infection types; therefore, it is important to focus recommendations on the basis of patient-specific circumstances. The evaluation of this statement was insupportable in its entirety; therefore, the review of the literature centered on the use of dual empirical coverage as well as the use of an aminoglycoside in combination therapy. The use of dual empirical coverage is well supported in the literature; however, the selection of an aminoglycoside is problematic because it is not necessarily appropriate in all clinical situations. There are 5 inherent issues that will be addressed in the evaluation of this statement: the role of adequate empirical therapy for serious HAI in the determination of outcome [139], the potential value of combination antimicrobial therapy in the determination of outcome [140], the potential efficacy of aminoglycosides as a component of combination antimicrobial therapy for serious HAI [3], the potential efficacy of quinolones as a component of combination therapy for serious HAI [141], and the influence of antibiotic-resistance surveillance on selection of therapeutic agents [46]. Each of these issues will be discussed with respect to evidence in favor of or against acceptance of the statement. Methods A PubMed literature search was conducted on 4 September 2007 to identify studies related to dual empirical coverage of infections with gram-negative pathogens. The search term cross infection/drug therapy or cross infection therapy was combined using the AND function with antibiotic therapy and gram-negative. Results were limited to the English language and studies published within the past 5 years. The search yielded 204 articles, 2 of which were relevant to the statement. A second PubMed search was conducted using the search terms gram-negative bacterial infections/drug therapy and cross infection, combined using the AND function. This search yielded 200 articles, 10 of which were relevant to the statement. Twelve additional articles were also reviewed from previous searches. Evidence The role of adequate empirical therapy for serious HAI in the determination of outcome. The outcome of serious HAI is improved by selection of adequate empirical antibiotic therapy, as defined by susceptibility of the infecting organism(s) to the agent(s) selected. Several retrospective and prospective clinical studies since the mid-1990s have provided statistical evidence of the positive effect of adequate empirical antibiotic therapy on clinical outcome. These studies have also concluded that adjustment of therapy when susceptibility data become available does not reverse the unfavorable effect of inadequate empirical therapy. In 1997, Luna et al. [139] described a prospective cohort study of 132 patients with VAP to determine the impact on outcome of a change in antibiotic therapy based on the results of culture of specimens collected by early bronchoalveolar lavage (BAL). Among patients from whom a pathogen was recovered by BAL, mortality was 91% after inadequate initial therapy and 38% after adequate initial therapy ( P!.001). When therapy was changed according to BAL culture results, mortality was comparable to that among patients who continued to receive inadequate therapy. Kollef and Ward [140] reported the results of a similar study in 1998 to determine the influence of mini-bal cultures on subsequent changes in antibiotic therapy and outcomes in 130 patients with suspected VAP. Mortality among patients for whom therapy was either begun or changed HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S81

28 Figure 8. Voting comparison for statement 6 ( Vancomycin is obsolete for treating MRSA infections ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Web-based survey; Summit refers to the Health Care Associated Infection Summit panel. MRSA, methicillin-resistant Staphylococcus aureus. at the time of BAL culture results was 60.8%, compared with 33.3% among patients requiring no change in initial antibiotic therapy ( P!.001). Thus, a delay in initiation of adequate therapy was associated with greater mortality. A prospective cohort study by Kollef et al. [3] was subsequently reported in 1999; the report described 655 critically ill infected patients admitted to the ICU. The overall mortality was 15.6%. Mortality among patients receiving inadequate initial antimicrobial treatment was 52.1%, compared with 12.2% among patients who received adequate initial treatment (P!.001). The effects of inappropriate initial antimicrobial therapy on outcomes for 286 patients with bacteremia due to antibioticresistant organisms were reported by Kang et al. in 2005 [141]. In a study of patients with a high-risk source of bacteremia, inappropriate initial antibiotic therapy was independently associated with increased mortality (mortality among those given appropriate therapy, 27.4%; mortality among those given inappropriate therapy, 38.4%; P p.049) (OR, 3.64; 95% CI, ; P p.030). Fraser et al. [46] reported similar results in a study published in 2006 involving 920 patients with microbiologically documented infections. Thirty-day all-cause mortality was 20.1% among those who received inappropriate initial empirical antibiotic therapy and was 11.8% among those who received appropriate therapy ( P p.001). In a study of patients with bacteremia published in 2007 by Tumbarello et al. [84], 186 patients infected with ESBL-producing organisms had 21-day mortality of 59% after inadequate initial antimicrobial therapy, compared with 18.5% among those who received adequate initial therapy ( P!.001). The potential value of combination antimicrobial therapy in the determination of outcome. The potential benefit of combination antibiotic therapy, compared with effective singledrug therapy, remains ill defined. However, in a retrospective study of 115 patients with P. aeruginosa bacteremia, Chamot et al. [142] found that adequate empirical combination therapy yielded lower 30-day mortality than did adequate empirical monotherapy, inadequate empirical monotherapy, or inadequate empirical combination therapy. In addition, adequate definitive combination therapy given when susceptibility results became available did not improve survival, compared with adequate definitive monotherapy. These results support the benefit of adequate empirical monotherapy or combination therapy, compared with delayed definitive therapy for bacteremia due to P. aeruginosa. Adequate combination empirical therapy was also more effective than adequate empirical monotherapy. Neutropenic patients accounted for 30% (34 of 115 patients) of the study population. A prospective, observational study of 230 patients with Klebsiella bacteremia showed no difference (20% vs. 18%; P 1.05) in 14-day mortality between those given monotherapy and those given combination therapy (b-lactam plus aminoglycoside) [143]. However, for the subgroup of patients who experienced hypotension (systolic blood pressure, 90 mm Hg) within 72 h before or on the day of the positive blood culture, those who received combination therapy experienced significantly lower mortality (24%) than did those who received monotherapy (50%). A meta-analysis published in 2004 reviewed 17 studies that compared combination therapy and monotherapy for bacteremia caused by gram-negative organisms [144]. The authors found no mortality benefit with combination therapy. However, analysis of only P. aeruginosa bacteremia showed a significant mortality benefit (OR, 0.50; 95% CI, ). In the above-mentioned studies, most of the effective combination therapies used b-lactam and aminoglycoside agents to which common isolates were susceptible. In the more recent era of MDR gram-negative bacilli, use of novel empirical antibiotic combinations may be dictated by advanced local resistance patterns. In certain areas of New York City and Morocco, effective empirical combination therapy requires use of a polymyxin alone or in combination with other agents, according to special in vitro susceptibility tests [ ]. The potential efficacy of aminoglycosides as a component of combination antimicrobial therapy for serious HAIs. S82 CID 2008:47 (Suppl 2) Kollef et al.

29 The potential efficacy of aminoglycosides as a component of combination antibacterial therapy is illustrated by the studies described above. Guidelines for the management of HAP, VAP, and HCAP in adults were published jointly by the ATS and the IDSA in 2005 [1]. Suggested antibiotic combinations included either an aminoglycoside or quinolone, chosen on the basis of local susceptibility data. Early evidence that aminoglycoside therapy for serious pneumonia caused by gram-negative pathogens is relatively ineffective because of poor tissue penetration is contradicted by more-recent pharmacokinetic/pharmacodynamic studies indicating that optimal aminoglycoside therapy is achieved by once-daily administration, not divided daily doses [148]. The potential efficacy of quinolones as a component of combination therapy for serious HAIs. The potential efficacy of quinolones as a component of adequate empirical combination therapy depends on the intensity of local use and likely the degree of resistance among invading gram-negative pathogens. A surveillance study published by Neuhauser et al. [149] examined fluoroquinolone resistance in They found that overall susceptibility to ciprofloxacin decreased from 86% in 1994 to 76% in 2000 and was significantly associated with increased use of fluoroquinolones. Other studies published during the past decade have documented the rising use of fluoroquinolones in various areas of the United States and its association with increasing resistance among gram-negative bacilli [ ]. The influence of antibiotic resistance surveillance on selection of therapeutic agents. Increasing antibiotic resistance among gram-negative bacilli in the United States and internationally has been well documented in the past few decades. Its local incidence should influence the selection of empirical therapy for serious infections with gram-negative bacilli. National surveillance studies in the United States have indicated a greater degree of quinolone resistance than aminoglycoside resistance among P. aeruginosa and Acinetobacter isolates, particularly from ICUs [ ]. Routine colonization surveillance in an ICU has demonstrated that knowledge of colonization status before infection is associated with higher rates of appropriate therapy for patients with bacteremia caused by antibiotic-resistant gram-negative bacilli [155]. A retrospective cohort study notes the clinical implications of resistance and the value of accurate susceptibility information. In that study, Tam et al. [156] examined 34 bacteremia episodes involving P. aeruginosa isolates with reduced susceptibility to piperacillin-tazobactam, which was given empirically for 7 episodes. Thirtyday mortality was found to be 85.7% in the group receiving piperacillin-tazobactam, compared with 22.2% in the group receiving other antipseudomonal agents ( P p.004). Tam et al. [156] observed an increase in mortality among patients infected with an isolate that had increased resistance, despite the fact that these patients had received appropriate therapy. Currently, there is no real clinical data supporting a synergistic effect of dual coverage for P. aeruginosa or any other gram-negative bacilli. The same is true for resistance. The main rationale for dual coverage of gram-negative bacilli is to increase the likelihood of the administration of appropriate therapy. On another note, in 2007, Livermore and Pearson [157] analyzed the utility of international, national, and local resistance surveys. They concisely summarized the essence of their findings in the title Antibiotic resistance: location, location, location, emphasizing that for patient management, good local data are essential [157, p. 7]. They highlighted the complexity of issues and large variances in resistance rates according to country, patient characteristics, and unit of care (i.e., nursing home vs. ICU). The authors concluded that, although these surveys help to illustrate trends, local susceptibility data are essential to good clinical management. Grading of Evidence On the basis of a review of the studies cited above, 33% of the workshop members voted that the evidence to support the statement was category II, 50% voted that it was category III, and 17% voted that it was category V (table 3). Level of Support Overall, 27% of the workshop members voted to accept the statement with some reservations, 64% voted to accept the statement with major reservations, and 9% voted to reject the statement with reservations. None of the summit members voted to accept or reject the statement completely. In comparison, of the 744 IDSA members who participated in the online survey, 30% voted to accept the statement completely, 38% voted to accept the statement with some reservations, 11% voted to accept the statement with major reservations, 18% voted to reject the statement with reservations, and 3% voted to reject the statement completely (figure 9). Discussion and Future Directions In conclusion, this statement can be supported by evidence that microbiologically adequate empirical treatment of serious HAIs due to gram-negative bacteria provides optimal clinical outcome. Evidence also supports the use of dual therapy to provide broad empirical coverage, as well as improved mortality for patients with bacteremia caused by P. aeruginosa. National surveillance data from the United States provide evidence that aminoglycosides retain greater susceptibility than do quinolones as potential second agents in combination therapy. However, in selected and expanding geographic areas, antimicrobial resistance has progressed to include all aminoglycosides and quinolones, as well as all b-lactams. This phenomenon precludes the use of a definitive general statement that includes a HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S83

30 Figure 9. Voting comparison for statement 7 ( Serious HAIs due to suspected gram-negative bacteria should be treated empirically with dual coverage that includes an aminoglycoside ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Webbased survey; Summit refers to the Health Care Associated Infection (HAI) Summit panel. single agent or class of agents as appropriate therapy in all locations. STATEMENT 8: PATIENTS WITH SERIOUS HAIs WHO HAVE RISK FACTORS FOR FUNGAL INFECTIONS REQUIRE EARLY EMPIRIC ANTIFUNGAL THERAPY TO REDUCE MORTALITY Rationale and Definition of Statement Traditionally, patients presenting to the hospital with suspected BSI or severe sepsis have been considered at risk for infections with selected pathogens, including MSSA, Streptococcus pneumoniae, and gram-negative organisms such as E. coli. Recognition that risk factors for infection with antibiotic-resistant pathogens include factors beyond the hospital setting has led to the evolution of the concept of HAIs. Briefly, this concept explained in detail elsewhere in this supplement attempts to capture the fact that many patients regularly interact with the health care system and are routinely exposed to extensive antimicrobial therapy outside the hospital. As such, they may become infected with a broad range of pathogens, including organisms traditionally classified as community associated illness or with bacteria previously thought to arise only in hospitalized persons who develop nosocomial syndromes. One class of nonbacterial organisms has recently emerged as an important pathogen causing nosocomial BSIs [158, 159]. Yeast represents an increasingly common cause of serious hospital-acquired BSI. More specifically, Candida species are the third or fourth most common cause of hospital-acquired BSIs, depending on the epidemiologic literature reviewed [158, 159]. This observation begs the question as to whether this finding applies to patients with health care associated BSI. In other words, does yeast now cause BSI in persons presenting to the emergency department with a syndrome that resembles BSI or severe sepsis? To validate the proposed statement, it is necessary to explore 3 specific issues: What is the prevalence of Candida as a cause of BSI in patients presenting to the emergency department? Does failure to treat candidemia result in adverse outcomes for patients? Do patients with such candidal BSI have risk factors for HAI? Methods A literature search of the PubMed database was conducted on 4 September The search was not limited to the English language. The purpose of the search was to identify articles addressing the epidemiology of candidemia, the distribution of the specific species of Candida that may cause BSI, the prevalence of candidemia among patients presenting to the emergency department, and treatment strategies for candidemia. Specific search terms used included Candida, candidemia, fungus, fungemia, bloodstream infection, and sepsis. The term candidemia OR fungemia resulted in the identification of 2689 articles. Although searching for the terms healthcare associated and healthcare-associated resulted in 150,000 potential articles, the combination of either of these phrases with candidemia OR fungemia OR BSI yielded only 12 publications. To expand the potential number of studies to be reviewed, the search strategy was subsequently modified to incorporate these phrases: inappropriate therapy, risk factors, and presumptive therapy. These selections were pooled in a Boolean fashion with the original search terms attempting to capture BSI infection with yeast. Despite broadening the search, only 4 additional articles potentially relevant to the statement were located. The paucity of published literature suggests that the concept of health care associated candidemia has not been well studied. This may reflect that either the concept is relatively new or this condition is not of clinical concern. In either case, the limited number of studies necessarily precludes definitive and strongly worded conclusions about the statement and suggests that readers of this literature must be cautious as they explore this area. Additionally, the small number of analyses automatically must make one skeptical as to the generalizability of the observations described in reports of these studies. S84 CID 2008:47 (Suppl 2) Kollef et al.

31 Evidence Epidemiology of health care associated candidemia. Two reports either directly or indirectly explored this question [160, 161]. In general, these studies suggest that health care associated candidemia exists as a distinct entity. In a large surveillance project focused on patients with candidemia presenting to the emergency department, Sofair et al. [161] prospectively evaluated all cases of candidal BSI in several hospitals in various regions of the United States. This project was sponsored by the CDC and represented a specific effort to grapple with the notion of community-onset candidemia. These investigators had clear criteria for defining a BSI caused by Candida as community onset in origin. Of 1143 cases of candidemia evaluated, the authors determined that 356 (31%) were community-onset infections [161]. More importantly, these investigators determined the prevalence of select risk factors for candidemia in persons with community-onset disease. A review of the distribution of these risk factors reveals that the vast majority of these communityonset cases of candidemia did, in fact, represent HAIs. For example, 1 in 5 subjects had underlying malignancy, and more than a quarter of the 365 persons were receiving immunosuppressive therapy [161]. More strikingly, approximately half of patients with community-onset infection had central venous catheters in place. With respect to the distribution of specific species causing candidemia, patients with community-onset infection were less likely than were persons with traditional nosocomial infection to have Candida albicans implicated. Additionally, 25% of community-onset infections were due to Candida glabrata a rate no different from the one seen in the traditional nosocomial candidemia cases. Exploring the question of health care associated candidemia from a different perspective, Shorr et al. [160] reviewed a large administrative database to determine the general prevalence of health care associated BSIs. They defined health care associated BSI as a BSI case diagnosed within 2 days after infection that had any of the following conditions: the patient was admitted from a nursing home, had been hospitalized in the past 30 days, was being given treatment with immunosuppressive therapy, had active malignancy, or was receiving chronic hemodialysis. Among nearly 7000 blood-culture confirmed cases of BSI, health care associated processes accounted for nearly 55% [160]. Nearly 2% of health care associated BSIs were due to yeast [160]. This rate of fungal BSI was lower than the rate of fungemia noted in nosocomial BSI. However, since the rate was not zero, it indicated that the proposed definition for health care associated fungemia does capture a unique population of patients. Conversely, these data underscore the relative infrequency of this condition, given the huge number of BSI cases seen annually in emergency departments in the United States. Implications of failure to treat candidemia. Over the past 5 years, multiple analyses have documented that failure to promptly treat serious infections increases a patient s probability of death [70, 79, 82, 162]. This finding has been confirmed for multiple disease states, from VAP to severe sepsis and septic shock [70, 79, 82, 162]. The relationship between mortality and either a delay in initial antibiotic therapy or the administration of inadequate therapy also applies if one focuses on specific pathogens (e.g., MRSA), rather than on clinical syndromes [5]. The definitions used for inadequate therapy generally categorize this as the administration of an anti-infective agent to which the culprit pathogen is resistant in vitro. For candidemia, only 2 reports attempted to address the relationship between inadequate or delayed antifungal therapy and survival [91, 92]. A potential explanation for the existence of so few reports dealing with this topic is the fact that in vitro susceptibilities for antifungal agents are not well described and that controversy exists regarding what represents in vitro resistance. In a retrospective analysis, Morrell et al. [92] reviewed 157 cases of candidemia. Their aims were to determine predictors of outcome in this disease and to describe the relationship between survival and both delays in antifungal therapy and inadequate antifungal treatment. Two analyses were conducted; the primary analysis included all antifungals, and a secondary analysis evaluated patients infected with C. albicans, Candida parapsilosis, or Candida tropicalis generally susceptible to fluconazole. The results were not stratified by organism type. Approximately half of the patients were infected with a non C. albicans species, and nearly 1 in 5 cases was attributed to either Candida krusei or C. glabrata [92]. They defined delayed therapy as administration of an antifungal agent 112 h after the patient s initial blood culture specimen was drawn. Inadequate therapy represented use of fluconazole for infections due to C. krusei. Specific MIC 90 breakpoints were not determined. Overall mortality approached 30%, which is similar to the death rate for candidemia described in other reports [92]. For example, in the analysis by Sofair et al. [161] noted above, the death rate for candidemia exceeded 25%. Of the 157 patients, only 5 received timely and adequate antifungal therapy. Strikingly, the death rate among patients given adequate treatment within 12 h after the blood culture specimens were drawn was only 10% [92]. Among persons given antifungal therapy beyond this 12-h window, the mortality rate increased to 33% ( P p.169) [92]. More importantly, when stratifying the time to therapy into the periods of h, h, and 148 h after blood culture specimens were drawn, these authors saw no difference in mortality. In multivariate analysis, a delay in antifungal treatment independently doubled a patient s risk of death. Confirming these observations, Garey et al. [91] reviewed 230 cases of candidemia at 4 different centers. The crude mor- HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S85

32 tality rate in this cohort was 30%. Of note, all subjects were given treatment with fluconazole. These authors determined that persons given fluconazole on the day the culture specimen was obtained faced a mortality risk of 15%. They also observed a stepwise increase in probability of death as time progressed ( P p.0009). Specifically, persons given treatment on the day after the culture specimen was drawn had a mortality rate of 25%, whereas those who were finally given fluconazole 3days after the culture specimen was drawn had a 40% unadjusted chance for in-hospital death [91]. In their logistic regression, delay in therapy heightened the potential for death by 50% (AOR, 1.50; 95% CI, ) [91]. This relationship persisted even after exclusion of persons for whom fluconazole may have been inadequate on the basis of a definition similar to the one employed by Morrell et al. [92]. Is risk stratification possible? Numerous reports detail potential risk factors for fungemia [158, 159]. These range from patient variables, such as a history of recent abdominal surgery and underlying malignancy, to process of care issues, including presence of a central venous catheter or receipt of parenteral nutrition [158, 159]. Unfortunately, efforts to develop a specific risk score that identifies persons with fungemia as the likely cause of their syndrome have been fraught with limitations. Often 2 approaches are employed one relying on the presence of certain risk factors, and the other using surveillance for Candida colonization. Use of risk scores tends to compromise specificity for the sake of sensitivity. In other words, although a proposed score may identify a cohort of persons more likely to have candidemia, the rate of candidemia remains sufficiently low, and it can be presumed that clinicians would need to give treatment to many patients without candidal infection to ensure that they were capturing cases of candidemia. Alternatively, surveillance-based strategies are necessarily cumbersome and are unlikely to be of value for treatment of health care associated candidemia, because the patient is, by definition, presenting to the emergency department and has not been in the hospital long enough to have had surveillance cultures performed. A report by Leon et al. [163] represents a recent attempt to refine the risk-score paradigm. In a multicenter trial in Spain, these investigators studied 1669 persons who stayed in the ICU for at least 7 days. The overall rate of candidemia was 6% [163]. Specific variables associated with subsequent ICU-onset candidemia included recent surgery, underlying severe sepsis, use of parenteral nutrition, and known Candida colonization [163]. Researchers developed a complex point-scoring tool based on logistic regression, which employed good screening characteristics for candidemia. Based on the plot of the receiver operating curve, their score had an area under the curve of 0.85 [163]. However, this score has not been independently validated in other settings or in other studies. Furthermore, for the purposes of determining who might be at greater risk for health care associated fungemia, their score may not be applicable, because it incorporates the findings from surveillance cultures. Addressing colonization in particular, Pairroux et al. [164] explored a role for the colonization index in determining the potential for candidemia. Again, admittedly, this strategy will not be helpful in the emergency department. However, for completeness, readers should familiarize themselves with this paradigm. These researchers completed a before-after study relying on the colonization index. They computed the colonization index as the number of sites on a patient that tested positive for Candida divided by the total number of sites swabbed. Swabbing was done biweekly. If the colonization index was 10.4, these patients were given preemptive therapy with fluconazole. With this technique and strategy, they were able to significantly reduce rates of proven ICU-acquired candidemia (from 2.2% to 0%; P!.001) [164]. There were no specific reports investigating risk stratification in health care associated or community-onset fungemia. This is perhaps not surprising, given the overall limited literature on this topic. Grading of Evidence On the basis of a review of the studies cited above, the workshop members agreed that the nature of the evidence available to support this statement was category II (table 3). Level of Support When voting on the support for this statement, 0% of the summit participants voted to accept the statement completely, 64% voted to accept the statement with some reservations, 27% voted to accept the statement with major reservations, and 9% voted to reject the statement with reservations. None rejected the statement completely. In comparison, of the 744 IDSA members who participated in the online survey, 27% voted to accept the statement completely, 44% voted to accept the statement with some reservations, 17% voted to accept the statement with major reservations, 10% voted to reject the statement with reservations, and 2% voted to reject the statement completely (figure 10). Discussion This statement should be viewed as complementary to the others in this supplement, addressing both particular pathogens and specific HAI syndromes. For the concept of HAI to prove meaningful, it must be internally consistent. Thus, if one limited the health care associated stratification to bacterial pathogens only, the entire notion might prove both difficult to apply and unhelpful. Therefore, recognition that, even for fungal BSI, the health care associated concept is unique reinforces the sup- S86 CID 2008:47 (Suppl 2) Kollef et al.

33 that the vast majority of patients are not given prompt treatment, it appears there is ample room for improvement. Figure 10. Voting comparison for statement 8 ( Patients with serious HAIs who have risk factors for fungal infections require early empiric antifungal therapy to reduce mortality ). IDSA refers to the members of the Infectious Diseases Society of America who responded to a Webbased survey; Summit refers to the Health Care Associated Infection (HAI) Summit panel. port for the need to adopt HAIs as distinct syndromes. Although there is certainly overlap between community-acquired, health care associated, and nosocomial processes, the evidence consistently underscores the need to break our traditional dichotomous classification scheme into 3 distinct components. Unfortunately, there are only 2 analyses that specifically address the epidemiology of health care associated candidal BSIs [160, 161]. These studies, however, were internally valid and well conducted. Thus, clinicians should at least recognize the potential for candidemia to be a cause of BSI in patients presenting to the emergency department. This statement is not meant to imply that physicians should prescribe antifungal treatment either routinely or reflexively. Instead, local epidemiologic information must be gathered to facilitate the development of local protocols to determine whether Candida species are an issue of concern. Readers should also note that there are no data suggesting that health care associated candidemia does not exist. In other words, there are no studies that specifically disprove this assertion. For risk stratification, one must rely on clinical judgment. No reliable tool exists to help determine which patients face an elevated potential for candidemia. Given the pathophysiology of the process, it appears that immunosuppression or presence of a central venous catheter is necessary, but neither is a sufficient condition for this disease. Perhaps, therefore, in giving treatment to persons presenting with a syndrome consistent with severe sepsis but not showing evidence of pulmonary infection (or other evident infection), clinicians should consider more formally candidal BSI in the differential diagnosis, particularly if multiple risk factors, including those noted above, are present. However, this recommendation represents opinion more than fact but does acknowledge that failure to promptly and adequately treat fungal BSI leads to substantial excess mortality. Conversely, one cannot hope to begin antifungal therapy promptly if one presupposes that yeast can never be a cause of health care associated BSI. Given that it seems Future Directions Certainly, more broadly designed prospective epidemiologic research is required. Such projects must include a range of institutions, rather than a focus exclusively on academic centers. With such information, geographic variations may become apparent. More importantly, these surveillance studies can simultaneously collect information that allows for the development and validation of risk-stratification tools. Finally, other diagnostic measures are needed. Since cultures for Candida may take several days to grow, clinicians require more-rapid diagnostic interventions to determine whether to continue or stop presumptive antifungal treatment. STATEMENT 9: ALL INFECTIONS IN IMMUNOCOMPROMISED PATIENTS SHOULD BE CONSIDERED HAIs UNTIL PROVEN OTHERWISE Rationale and Definition of Statement Infection due to a diverse spectrum of pathogens is the most common and well-recognized complication in patients with compromised immunologic host defenses as well as a native disease and/or iatrogenic interventions. The microbial etiology for such infections may vary across different specific immune defects, severity and duration, and other modifiers, including the patient s prior and present geographic location, prior exposure to anti-infectives for prophylaxis or treatment, and exogenous exposures (e.g., transfused blood products or donor organs). The orthodox view pertaining to the origin of infections in immunocompromised hosts recognizes that the inciting organism(s) may originate from (1) the patient s native endogenous flora or dormant organisms, which become reactivated with failed immune defenses; (2) endogenous flora, which has been modified principally by exposure to anti-infective agents or the animate and inanimate nosocomial environment; (3) exogenous reservoirs; and (4) as-yet-unknown sources. Among organ transplant recipients, the donor organ represents another reservoir for pathogen transmission. Although much of the important clinical management of immunocompromised patients occurs within the hospital unit or the critical care setting, some of the management has shifted to the parahospital or outpatient health care facilities. Prominent examples include outpatient chemotherapy for oncologic disease, management of HIV-associated illness, and long-term acute care facilities that receive organ transplant recipients for ventilator dependence or rehabilitation. HAIs within this expanded sphere might have a greater impact on the immunocompromised host than on immunocompetent patients and also may be caused by a different spectrum of pathogens. With HAI Summit Critical Appraisal CID 2008:47 (Suppl 2) S87