Mortality related to methicillin-resistant Staphylococcus aureus (MRSA) bloodstream infection (BSI) remains high, despite changes in the epidemiology. To analyze the current predictive factors for mortality we conducted a prospective study in a large cohort of patients with MRSA-BSI from 21 Spanish hospitals. Epidemiology, clinical data, therapy and outcome were recorded. All MRSA strains were analysed, including susceptibility to antibiotics and molecular characterization. Vancomycin MICs (V-MIC) were tested by the E-test and microdilution methods. Time until death was the dependent variable in a Cox regression analysis. Overall, 579 episodes were included. Acquisition was nosocomial in 59% and vascular catheter was the most frequent source (38%). A dominant PFGE genotype was found in 368 (67%) isolates, which belonged to Clonal Complex (CC)5 and carried SCCmecIV and agr2. Microdilution V-MIC50 and V-MIC90 were 0.7 and 1.0 mg/L, respectively. Initial therapy was appropriate in 66% of episodes. Overall mortality was observed in 179 (32%) episodes. The Cox-regression analysis identified age >70 years (HR 1.88), previous fatal disease (HR 2.16), Pitt score >1 (HR 3.45), high-risk source (HR 1.85) and inappropriate initial treatment (HR 1.39) as independent predictive factors for mortality. CC5 and CC22 (HR 0.52 and 0.45) were associated with significantly lower mortality rates than CC8. V-MIC ≥1.5 did not have a significant impact on mortality, regardless of the method used to assess it.
Methicillin-resistant Staphylococcus aureus (MRSA) bloodstream infection (BSI) has been a cause of concern in healthcare systems around the world in recent decades, due to an increase in its incidence and undesirable related outcomes . Despite the recent epidemiological changes, MRSA clonal replacement, clinical use of new antibiotics and improvement in supportive therapies, mortality related to MRSA-BSI remains close to 30% .
A variety of factors have been associated with worse outcomes among MRSA-BSI patients, including host characteristics, bacterial genetic background and therapeutic management . At present, however, no predictors for mortality have been definitively established, because only a few recent large prospective multicentre studies have analysed all these factors together.
Vancomycin, which is the standard therapy for MRSA-BSI, has recently been the focus of attention. Its suboptimal in vitro killing activity for S. aureus compared with betalactams  and the observation of MIC creep in many hospitals around the world  have raised questions about its suitability for use against MRSA invasive infections [6, 7]. Special concern has been raised during the last decade regarding the impact on mortality of MRSA strains with reduced vancomycin susceptibility [3, 7-9].
This study aimed to assess the prognostic factors for mortality among MRSA-BSI patients, taking into account the current epidemiological data and carefully examining the clinical impact of vancomycin susceptibility on mortality.
Study period and patients
The study was conducted from June 2008 to December 2009 at 21 Spanish hospitals. Four centres had <500 beds, nine had 500–1000 beds and eight had >1000 beds. An infectious disease specialist prospectively followed adult patients (>16 years old) with MRSA blood cultures previously detected at the Microbiology Laboratory, and excluded the cases that did not meet the inclusion criteria, such as lack of signs and symptoms consistent with sepsis. A standardized protocol with demographic and clinical information, including age, sex, co-morbid conditions, source and acquisition, diagnostic explorations, antibiotic treatment, follow-up and outcome was followed. Strains were sent to a central laboratory for further analysis.
All the episodes included in the study were used to assess the prognostic factors for mortality, which was measured 30 days after the first blood culture. Two different models that only differed in the method used to assess MRSA vancomycin susceptibility (E-test or microdilution) were used. MICs were stratified as follows: <1.5 and ≥1.5 μg/mL. Time until death was the dependent variable in a Cox regression analysis and was censored 30 days after the end of therapy or end of follow-up. The prognostic factors for non-appropriate initial antibiotic therapy were analysed in a logistic regression model. Patients who received appropriate initial therapy were included in an adjusted multivariate analysis designed to assess the existence of differences in mortality related to the specific antibiotic administered.
Methicillin-resistant Staphylococcus aureus-BSI is defined as the presence of at least one positive blood culture for MRSA in a blood sample from a patient with clinical findings consistent with infection. Co-morbidity was measured by the Charlson score, which stratifies the associated diseases into an ordinal scale. Patients were classified into three categories on the McCabe scale according to their prognosis of survival before the MRSA-BSI: non-fatal if death was expected within a period longer than 5 years; ultimately fatal if death was expected between 1 and 5 years; and rapidly fatal if it was expected in the following year. Severity of disease was assessed by the Pitt score. Three acquisition categories were considered according to Friedman criteria: (i) nosocomial-BSI if the episode was diagnosed at least 48 hours after the hospital admission, either to the ICU or to a conventional hospital ward (non-ICU), and if there were no signs or symptoms of infection at admission; (ii) healthcare related-bacteraemia if the patient had had contact with the healthcare system within the previous 3 months; and (iii) community-acquired otherwise. Source of the BSI was defined according to the CDC criteria . BSI of unknown source (primary) was defined when its origin was uncertain after careful examination of the clinical and microbiological data. It was considered high risk if the source was lower respiratory tract, endocarditis or unknown . Distant extension was diagnosed in the presence of at least one distant infection secondary to blood spread seeding. Persistence was defined as growth of MRSA in blood cultures after more than 48 h of appropriate antibiotic therapy. The initial antibiotic was defined as the antibiotics administered in the first 48 h after BSI onset, regardless of the microbiological information. Definitive treatment was considered as the antimicrobials administered after conducting microbiological sensitivity tests. Antibiotic treatment was considered appropriate if the strain was susceptible to at least one of the administered antibiotics, with the exception of aminoglycosides, which were considered inappropriate, regardless of the sensitivity tests. Source eradication was considered if the catheter or foreign body was removed, or if a surgical intervention or drainage of BSI source was performed.
Susceptibility testing and molecular epidemiology of MRSA isolates
Each hospital identified the strain and performed preliminary susceptibility tests. Isolates were sent to a central reference laboratory. All S. aureus were identified by latex agglutination (Pastorex Staph-plus, Bio-Rad Laboratories, Madrid, Spain) and DNase production (DNas E-test Agar, BioMérieux, Marcy l'Étoile, France). Antimicrobial susceptibility of all MRSA isolates was tested by the disk-diffusion method according to the Clinical Laboratory Standard Institute (CLSI) guidelines . The antimicrobial agents tested were penicillin, oxacillin, cefoxitin, erythromycin, clindamycin, gentamicin, tobramycin, ciprofloxacin, rifampicin, trimethoprim-sulphamethoxazole, tetracycline, fosfomycin, vancomycin, teicoplanin, chloramphenicol, daptomycin and linezolid. Minimum inhibitory concentration (MIC) was determined by the microdilution method in accordance with CLSI criteria by using commercial panels (ESTEN 2009, Sensititre™, Izasa, Barcelona, Spain) read visually. Vancomycin MICs were also studied by E-test (BioMérieux) on Mueller-Hinton agar, using a turbidity of 0.5 on the McFarland scale.
Pulsed-field gel electrophoresis (PFGE) was performed after SmaI restriction of chromosomal DNA. Restriction patterns were interpreted in accordance with criteria published elsewhere . To define PFGE types the FINGERPRINTING™ II software, version 3.0 (BioRad Laboratories, Inc., Madrid, Spain) was applied.
A dendrogram was generated by the unweighted-pair group method with arithmetic mean based on Dice coefficients. To define PFGE types the similarity coefficient cut-off was set at 80%. Optimization and band position tolerance were both set at 0.6%.Representative isolates of each PFGE type and subtype were studied to determine the Multilocus Sequence Type (MLST)  and the Staphylococcal Chromosome Cassette mec (SCCmec) types . MLSTs and SCCmec types were further inferred for all the strains. The agr polymorphism and the presence of genes encoding class S (lukS-PV) and class F (lukF-PV) proteins for Panton–Valentine Leucocidine (PVL) were studied by PCR in all the isolates, following the methodology described elsewhere [16, 17].
Continuous variables were compared using the Student's t-test or the Mann–Whitney U-test as appropriate. Qualitative and stratified continuous variables were compared using Fisher's exact test or Pearson's chi-squared test. Relative risks were calculated with 95% confidence intervals in a univariate analysis. Time until death was the dependent variable in a Cox regression analysis, and was censored 30 days after the end of therapy or the end of follow-up. Administration of appropriate antibiotic therapy was the dependent variable in a logistic regression model. All the variables with theoretical clinical significance and those that achieved a p value <0.10 in the univariate analysis were included in the multivariate analysis, and adjusted odds ratios (ORs) were calculated with 95% CI. Analyses were performed using SPSS v15 (Microsoft, USA).
The study was approved by the Spanish Network for Research in Infectious Diseases (REIPI) as well as the Institutional Review Board of each participating centre. Because no direct patient contact was planned, the requirement for informed consent was waived. The data were de-identified in each centre and only then transferred for analysis.
Five hundred and ninety episodes of MRSA-BSI were confirmed. Eleven were excluded from the analysis due to lack of information. Therefore, 579 episodes were finally included, which are summarized in Table 1.
Table 1. Summary of all episodes of methicillin-resistant Staphylococcus aureus bloodstream infection characteristics
Twenty-five PFGE types were found among 552 available isolates (Fig. 1). A dominant PFGE genotype (pulse-type 2) was found in 368 (67%), all of which belonged to Clonal Complex (CC) 5 (ST125 and ST146). They carried a SCCmec element type IV and agr type 2. Forty-seven (9%) isolates of clonal type ST228-agr2, a single locus variant of ST5, were considered a separate clone based on its SCCmec polymorphism type I. ST22-SCCmecIV-agr1 and ST8-SCCmecIV-agr1 represented 9% (48/552) and 8% (44/552), respectively, of all studied strains. PVL was found to be positive in 15 isolates; 11 of them belonged to ST8, three to ST125 and one to ST714. Among PVL-positive isolates of ST8, six strains out of 11 belonged to the USA300 clone. Vancomycin MIC (microdilution and E-test) distribution according to the molecular characterization is summarized in Table 2. Vancomycin MIC ≥ 1.5 μg/mL measured by microdilution and E-test was observed in 3.3% and 42.9% of the isolates, respectively. Linezolid MIC50 and MIC90 were 2 and 2 μg/mL and daptomycin MIC50 and MIC90 were <0.5 and 1 μg/mL.
Table 2. Distribution of vancomycin MIC (μg/mL) tested by microdilution and E-test methods among the three major clonal complexes in the cohort of methicillin-resistant Staphylococcus aureus bloodstream infections
Initial antibiotic therapy was appropriate in 66% of episodes. Factors related to inappropriate initial antibiotic therapy were certain sources of infection (unknown source (OR 4.22), lower respiratory tract (OR 3.55) and skin and soft tissues (OR 2.58)) and Pitt score ≤3 (OR 1.68; Table 3). The median delay in starting appropriate antibiotic therapy was 1.8 days (IQR 0–3). Antibiotics were administered for a median of 18 days (IQR 15–27). The source was eradicated or drained in the first 2 days in 35% of cases.
Table 3. Predictive factors for inappropriate initial therapy for methicillin-resistant Staphylococcus aureus bloodstream infection
Appropriate initial antibiotic n (%)
Inappropriate initial antibiotic n (%)
OR (95% CI)
Non-nosocomial acquisition includes healthcare-related and community acquisitions.
Overall mortality was observed in 179 (32%) episodes. The Cox-regression analysis identified age >70 years (HR 1.88), ultimately fatal or rapidly fatal disease (HR 2.16), Pitt >3 (HR 3.45), high-risk source (HR 1.85) and inappropriate initial treatment (HR 1.39) as independent predictive factors for mortality. CC5 (HR 0.52) and CC22 (HR 0.45) were identified as protective factors compared with CC8 (Table 4). Vancomycin MIC ≥ 1.5 μg/mL, measured either by the microdilution method (HR 1.71) or by E-test (HR 0.78), did not show a significant impact on mortality. Neither was vancomycin MIC ≥ 2 μg/mL identified as an independent predictor for mortality, regardless of the method used to assess it (microdilution, HR 0.83 (95% CI 0.11–6.05); E-test, HR 0.94 (95% CI 0.61–1.46)).
Table 4. Predictive factors for mortality due to methicillin-resistant Staphylococcus aureus bloodstream infection. Univariate and multivariate analyses
30-day deaths n (%)
30-day survivors n (%)
Multivariate analysis with microdilution vancomycin MIC ≥ 1.5
Multivariate analysis with E-test vancomycin MIC ≥ 1.5
This multicentre cohort study reinforces the concept that recent trends towards aging, non-nosocomial acquisition, extended use of new antibiotics and advances in medical support [1, 2] have not significantly modified the high mortality rates associated with MRSA-BSI, which remain close to 30% . We analysed the current prognostic factors for mortality, taking into account potential clinical, microbiological and therapeutic predictors. In our opinion, a number of our results merit further discussion.
From a clinical perspective, age, co-morbidities, high-risk sources and severity of sepsis were identified as independent host predictors for mortality. All of these had been identified in previous studies of MRSA-BSI [3, 8, 19]. A recent review  showed age to be the strongest and most consistent predictor for mortality in S. aureus bacteraemia. In contrast to other studies , healthcare-related acquisition was not identified as a risk factor for worse outcome, probably because of the low mortality rate among episodes acquired during haemodialysis (14%), which represented 19% of the non-nosocomial cohort.
There is little information in the literature regarding the influence of genetic background on the outcome of MRSA infections. It remains unknown whether some MRSA clones have a greater ability to cause invasive disease: some studies have found significant differences between them, but others have not [20, 21]. A recent study observed a greater number of haematogenous complications in episodes caused by strains belonging to CC5 and CC30 . Interestingly, in our study CC5 (ST125, ST146 and ST228) and ST22 (EMRSA15) were identified as protective factors compared with CC8. Regarding the PVL gene, in agreement with other authors , we found no relationship with worse outcomes. This gene was only harboured by some of the isolates in our study (3%), highlighting the current low prevalence of typical MRSA community clones in Spain .
We were particularly interested in assessing the clinical impact of vancomycin susceptibility on mortality. We did not find an independent association with vancomycin MIC ≥ 1.5 μg/mL, measured either by microdilution or by E-test. These results contrast with those of other studies that reported an association between higher vancomycin MIC measured by E-test and mortality  or treatment failure  among patients treated with vancomycin. Striking results in two recent studies suggested that decreased vancomycin susceptibility might be a marker of certain unidentified host or strain factors, related to worse outcomes: Aguado et al.  observed that E-test vancomycin MIC ≥ 1.5 was the only independent predictive factor for complicated MSSA bacteraemia treated with betalactams, while Holmes et al.  found increased mortality among episodes of S. aureus BSI with MIC ≥ 1.5, regardless of the specific antibiotic therapy administered. Like us, however, other authors did not find a significant impact of vancomycin susceptibility on mortality [8, 9]. Interestingly, Rojas et al.  did not observe any association using either susceptibility method.
In our opinion, these contradictory results between studies might have two complementary explantions: first, according to our results and the findings of Holmes et al. , certain clones might have an independent association with mortality. Therefore, because vancomycin MIC differs from one clone to another, it is reasonable to hypothesize that the distinct clonal distribution in countries can influence the impact of vancomycin susceptibility observed in different studies. Secondly, the low correlation between E-test and microdilution, which is illustred in Table 2 , might also lead to these different observed results. In our opinion, as Rojas et al.  stressed, there is lack of evidence in favour of using the E-test instead of the recommended microdilution method to assess MRSA vancomycin MIC. As far as we know, only one previous study found an association between vancomycin MIC ≥ 2 μg/mL and mortality, using the microdilution method . A recent meta-analysis of MRSA BSI studies that assessed vancomycin MICs by E-test concluded that while MIC ≥ 1.5 μg/mL was not associated with higher mortality, MIC ≥ 2 was . In contrast, our study and others  did not find an independent association with E-test vancomycin MICs ≥2. In summary, we believe that more studies will be needed to resolve this controversy.
As regards therapeutic management, the initial therapy was inappropriate in 34% of the episodes. It was also identified as an independent predictive factor for mortality, as in previous studies [3, 30]. This is an interesting observation because modifying this factor might reduce the number of related deaths. In our study, patients with lower Pitt scores and specific sources had a significantly higher risk of receiving initial inappropriate antibiotic therapy. Interestingly, we found no statistical differences in mortality in relation to the type of appropriate antibiotics used as initial or definitive therapy. However, we stress that this last analysis is limited because the choice of antibiotic was not controlled and there may have been some confounding factors in the variables regarding the therapy. Also, vancomycin serum levels were not assessed in this study. Therefore, conclusions regarding the impact of antibiotic therapy should be reached with caution.
In conclusion, we identified several prognostic factors for mortality in a large cohort of patients and provide support for the concept that host and bacterial characteristics, as well as initial therapy, have implications for MRSA-BSI outcomes. In contrast, higher vancomycin MIC had no impact on mortality, regardless of the method used to assess it. Special attention should be paid to patients with one or some of these predictors, in order to ensure the administration of correct initial therapy.
This work supported by Ministerio de Innovación, Instituto de Salud Carlos III (FIS 08/0335), and co-financed by European development Regional Fund “A way to achieve Europe” ERDF, Spanish Network for the Research in Infectious Diseases (REIPI RD06/0008). O.G had agrant Rio Hortega (CM08/228) from the Instituto de Salud Carlos III. A part of the results of this article was presented at the 51st Interscience Conference on Antimicrobial Agents and Chemotherapy, Chicago, 16–20 September, 2011. Presentation number: K-950.
B.A received funding for research from Pfizer, Novartis, Gilead y MSD and funds for advisory board membership Pfizer, Gilead, Novartis, Janssen, Astellas and MSD. N.B. received funding for speaking, consultancy, advisory board membership, travel from MSD, Pfizer, Gilead, Novartis, AstraZeneca. J.R.B. received funding for research from Novartis, has served as speaker for Astellas, Merck, AstraZeneca, and Pfizer, and has been consultant for Roche, Novartis, and Janssen. The rest of the authors have no conflicts of interest to declare.
REIPI/GEIH study groups: A. Jover, F. Barcenilla, M. Garcia (H Arnau de Vilanova, Lleida, Spain); M. Pujol, O. Gasch, MaA. Domínguez, M. Camoez (H Bellvitge, Universitat de Barcelona, IDIBELL, Barcelona, Spain); C. Dueñas, E. Ojeda (H Burgos, Burgos, Spain); J.A. Martinez, F. Marco (H Clínic, Barcelona, Spain); F. Chaves, M. Lagarde, F. López –Medrano (HU 12 de Octubre, Madrid, Spain); J.M. Montejo, E. Bereciartua, J.L. Hernández (H de Cruces, Bilbao, Spain); M.Á. Von Wichmann, A. Goenaga, J.M. García-Arenzana (H Donostia, Donostia, Spain); B. Padilla, C. Padilla, E. Cercenado (HGU Gregorio Marañón, Madrid, Spain); G. García-Pardo, J. Tapiol (HU Joan XXIII, Tarragona, Spain); J.P. Horcajada, M. Montero, M. Salvadó (HU del Mar, Barcelona, Spain); A. Arnáiz, C. Fernandez (HU Marques de Valdecilla, Santander, Spain); E .Calbo; M. Xercavins (HU Mutua de Terrassa, Terrassa, Spain); A. Granados, D. Fontanals (H del Parc Taulí, Sabadell, Spain); V. Pintado, E. Loza (HU Ramon y Cajal, Madrid, Spain); J. Torre-Cisneros, R. Lara, F. Rodríguez-López, M. Rodríguez, C. Natera (HU Reina Sofía, Córdoba, Spain); J.R. Blanco, I. Olarte (H San Pedro de la Rioja, Logroño, Spain); N. Benito, B. Mirelis (H de la Santa Creu i Sant Pau, Barcelona, Spain); J. Murillas, E. Ruiz de Gopegui (HU Son Espases, Mallorca, Spain); H. Espejo, MªA. Morera (H de Terrassa, Terrassa, Spain); J. Rodríguez-Baño, E. López, A. Pascual (HU Virgen Macarena, Sevilla, Spain); C .Martín, J.A. Lepe, J. Molina (HU Virgen del Rocio, Sevilla, Spain); R. Sordé, B. Almirante, N. Larrosa (HU Vall d'Hebrón, Barcelona, Spain).