At Mount Sinai Medical Center, MRSA organisms identified from blood culture are routinely stored frozen at −70°C. We retrieved frozen blood culture isolates (previously not thawed or subcultured) for all adult patients hospitalized with MRSA bacteremia from January 2002 through May 2007. We excluded episodes of polymicrobial bacteremia if MRSA was isolated in a single blood culture bottle or if the patient received inappropriate empirical treatment for the co-pathogen. We included in our analysis the first organism isolated from blood culture during any episode of MRSA bacteremia.

A computerized data system identified 748 eligible isolates, among which we were able to retrieve 699 (93.4%). These isolates had originally been tested for drug susceptibility by use of an automated instrument, the Microscan (Siemens Healthcare, Sacramento, CA, USA); for all isolates, the vancomycin MICs were <2 µg/mL; during 2002–2007, we used the Positive Breakpoint Combo 20 (Siemens Healthcare), which might not accurately detect VISA isolates (16). We sent some isolates—8 (15%) VISA strains, 88 (15%) non-VISA strains, and 10 VISA control strains (from the Network on Antimicrobial Resistance in Staphylococcus aureus, www.narsa.net)—to an outside laboratory for blinded testing by using the Vitek 2 (bioMérieux, Durham, NC, USA) with the AST-GP-67 card, and we sent 8 (15%) hVISA isolates to an outside laboratory for retesting by using time-killing profiles. No discrepancies were noted.

Retesting of isolates was performed with no access (blinded) to clinical data. Mueller-Hinton agar plates (study and control strains) were inoculated with 0.5 McFarland inoculum (10⁸ CFU/mL), and antimicrobial drug susceptibility to vancomycin was assessed by using Etest (AB Biodisk, Solna, Sweden), which has excellent sensitivity and specificity for this purpose (17). Isolates for which vancomycin MICs were >1 µg/mL (95.1%) were tested for the presence of hVISA by using vancomycin and teicoplanin Etest strips (the Etest macromethod). In brief, study and control inocula equivalent to 2.0 McFarland turbidity standard were plated on brain–heart infusion and Mueller-Hinton agars and incubated at 35°C–37°C for 24 and 48 hours, respectively. Isolates were interpreted as being hVISA strains when vancomycin and teicoplanin MICs were >8 µg/mL or teicoplanin-only MIC was >12 µg/mL (16). Quality control testing was performed weekly by using American Type Culture Collection (Manassas, VA, USA) organisms 29213, 29212, 700698 (hVISA), and 700699 (VISA) per Clinical and Laboratory Standards Institute guidelines (9). Actual Etest values were used for MIC₅₀ (the value below which 50% of MIC values for MRSA isolates tested fell), MIC₉₀ (the value below which 90% of MICs fell), and geometric mean MIC calculations.

We abstracted electronic and paper charts for each of the 603 patients corresponding to the 699 isolates; 65 of these patients were hospitalized >2 times for MRSA bacteremia. Chart abstraction was performed by 2 independent reviewers with no access (blinded) to laboratory data. Each reviewer separately examined 10% of charts; the κ statistic for coding of exemplar key variables was 0.87, indicating excellent agreement (18). Each new hospital admission was categorized as a new episode of bacteremia; the first positive blood culture was used as the index infection. We abstracted patient information regarding demographics, concurrent illnesses, patient’s residence before hospitalization (facility vs. community), bacteremia severity, and previous health care exposures.

We categorized each MRSA infection into 1 of 3 groups according to Etest result for vancomycin MICs as follows: VISA (MIC 4–12 µg/mL), hVISA, or non–VISA/hVISA MRSA (MIC <2 µg/mL). We also assessed MIC₅₀ and MIC₉₀ for vancomycin.

We defined the number of days to clearance of bacteremia as the date of first positive MRSA culture subtracted from the date of first negative culture for all patients for whom this information was available. Episodes of bacteremia ended on the date of death or on the date of the first negative blood culture that was not followed by a positive culture within 7 days. We did not study subsequent episodes during the same hospitalization. Vancomycin trough levels were measured after 3 doses of vancomycin; if multiple levels were measured, the modal level was used and classified as either >15 µg/mL or <15 µg/mL. We defined prior vancomycin exposure as receipt of >3 doses of vancomycin at least 7 days and <12 months before MRSA bacteremia. Empirically prescribed therapy was defined as appropriate if MRSA were susceptible to the antimicrobial drug used (according to in vitro susceptibility testing) and if therapy was started within 48 hours of the blood culture result. History of MRSA infection was defined as hospitalization with MRSA during the prior 12 months. We defined an episode as community associated if the patient was bacteremic within 48 hours of hospitalization and lacked health care–associated risk factors such as dialysis, nursing home residence, or history of MRSA infection. We divided health care–associated MRSA cases into community onset or hospital onset. Cases were health care–associated community onset if the patient had such risk factors and was found to be bacteremic within 48 hours of hospital admission; cases were of health care–associated hospital onset if the bacteremia occurred after 48 hours of hospitalization, consistent with the schema of Klevens et al. (19).

We defined renal insufficiency as serum creatinine level >2 mg/dL or glomerular filtration rate <50 mL/min/1.73 m² according to the Cockcroft-Gault equation. The source of the bacteremia was determined by a combination of positive MRSA culture growth from a site other than blood, radiologic evidence, or an attending physician’s statement in the medical record. We used the Duke criteria to define endocarditis (20). Severity of bacteremia at onset of infection was determined by use of vasopressors, elevation of serum creatinine levels from baseline (renal insufficiency), and admission to an intensive care unit after positive MRSA blood culture result. The major patient outcome measure was 90-day all-cause mortality rate. Mortality rate was determined by calculating deaths from the date of positive MRSA blood culture result up to 90 days while hospitalized, divided by the number of patients in the study (n = 603). Data regarding death after hospital discharge were not available for analysis. MRSA-attributable deaths were not included because of the difficulty in assessing exact cause of death.

We used SAS version 9.1 (SAS Institute, Cary, NC, USA) for statistical analyses (21). We used standard methods to describe univariate data and to calculate t-tests, and we used χ² for bivariate associations. We used generalized linear models (SAS Proc GLM) to assess associations between the 3 groups of MRSA infection and quantitative variables. We conducted multivariable analyses by using logistic regression (SAS Proc Logistic). We developed models for 3 variables: infection with VISA, infection with hVISA, and death. Model building was guided first by conceptual models of likely effect and informed by our bivariate analysis results. We assessed correlation coefficients between pairs of potential predictor variables by using appropriate parametric or nonparametric methods and included only 1 of any pair of variables with an r² of >0.25 in any model. Guided by the rule for stability of estimates established by Peduzzi et al., we limited the total number of predictor variables in any model (22). The significance of the models (Table 1, Table 2) are demonstrated by likelihood ratios, Wald <0.0001. When choosing between similar or comparable models, we selected the model associated with the smallest Akaike information criterion (<582); for example, in our analysis of predictors of risk for death, we rejected variables indicating HIV, malignancy, transplant, recent surgery, and presence of a medical device because they increased the Akaike information criterion. The predictive and discriminative performance of our models is shown in Tables 1 and 2. Our model predicting death has a c-score of 0.872; among 99,198 pairs, 87.1% were concordant, 12.7% were discordant, and 0.2% had the same scores. Although we also present the more familiar adjusted odds ratios, our primary measures of impact are adjusted risk measures (adjusted risk ratio and adjusted risk difference), which we derived from regression risk analysis, an enhancement over the usual presentation of logistic regression (23).

Each year during 2002–2006, the annual number of hospital admissions in this study were 117, 77, 147, 121, and 161, respectively; through May 31, 2007, another 76 patients were hospitalized (equivalent to 184 annual hospitalizations). The original testing of strains by Microscan did not detect VISA; repeat testing using the Vitek 2 detected 2 (25%) of 8 study strains and 2 (20%) of 10 control strains. The rate at which polymicrobial bacteremia met inclusion criteria was 3% (20 episodes). The Figure demonstrates the proportion of VISA, hVISA, and non–VISA/hVISA MRSA by year and the increase in mean vancomycin MICs during the study period. The vancomycin MICs for most (87%) isolates were 1–2 µg/mL. For VISA, MICs were as high as 12 µg/mL, although for 60%, MICs were 4 µg/mL. For 94% of hVISA strains, vancomycin MICs were 1.5–2.0 µg/mL.

The geometric mean MIC of vancomycin was 1.7 µg/mL; modal MIC = 2 µg/mL. In 2002, the MIC₅₀ of vancomycin was 1.5 µg/mL; by 2007, it was 2 µg/mL. The MIC₉₀ was constant during the study period; vancomycin MIC₉₀ was 4 µg/mL in 2002 and 2004–2007.

Key patient characteristics are shown in Table 3. Recent medical care seemed to be associated with type of strain. Nearly 40% of infections were health care–community associated, almost all the rest were hospital associated. Average length of stay for patients with all infection types was 32.9 days. Many cases of MRSA bacteremia were in patients with renal insufficiency and/or cardiovascular disease; >40% had recently had a surgical procedure. Many (43%) cases of bacteremia were secondary to central venous catheter infections. The 90-day all-cause mortality rate was 31.5% for the 603 patients; rate was 27.2% when all 699 episodes of bacteremia were considered.

Among the 603 patients, 47% had been exposed to vancomycin and 60% of these had had prior MRSA infection. Prior vancomycin exposure was more likely for patients with VISA (62%, 95% CI 47.7%–65.3%) than for patients with hVISA (42%, 95% CI 29.4%–59%) or other MRSA strains (47%, 95% CI 41.3%–49.7%).

Among the 699 episodes of MRSA bacteremia, vancomycin was used to treat 566 (81%) episodes. Initial vancomycin treatment was switched to daptomycin or linezolid for 12% of MRSA (non-VISA, non-hVISA) infections and 15% of VISA infections. For the 699 episodes of bacteremia, mortality rates were 27.2% overall, 16% (95% CI 6.8%–24.8%) when antimicrobial drug treatment was changed, and 26% (95% CI 22.2%–29.8%) when not changed. Our data did not enable us to determine the extent to which switching, or not switching, antimicrobial drugs contributed to survival.

Multivariable analyses (Table 1) demonstrate associations between key clinical characteristics and VISA or hVISA infections. The adjusted risk difference represents the absolute difference in risk for that given characteristic, all else held equal. Cirrhosis of the liver and central venous catheter infections nearly doubled the risk for hVISA infection. Cirrhosis and active malignancy increased the absolute risk for VISA by 14% and 7%, respectively. History of vancomycin exposure within 1 year increased the risk for VISA (6%) but did not increase the risk for hVISA infection.

The effect of various clinical characteristics on risk for death within 90 days is summarized in Table 2. For patients with concomitant MRSA bacteremia, older age increased the risk of dying. Cirrhosis or renal insufficiency and having lived in a nursing home before hospitalization or having been admitted to an intensive care unit were each independently associated with death (after adjusting for covariates in the model); each increased the risk of dying by 7%–15%. Patients who required vasopressors had an absolute increase in risk for death of ≈50%, after covariates were adjusted for. Risk for death was independently associated with lower risk for death among those who had diabetes mellitus or who had had a vascular graft as the source of the infection. A consultation with an infectious diseases specialist decreased the risk for death by 11%. Neither a VISA nor hVISA strain was independently associated with all-cause death after covariates in the models were controlled for.

A subanalysis of vancomycin MICs for strains infecting patients who died in the hospital found that the mean MIC was 1.7 µg/mL. The current breakpoint of vancomycin susceptibility is 2 µg/mL.