Suspensions of 0.5 McFarland turbidity standard in brain heart infusion (BHI) broth were prepared from isolates after overnight incubation. Ten microliters of each suspension was injected onto BHI agar plates containing 4 mg/L vancomycin. Because of the reported inducibility of vancomycin heteroresistance in some strains of S. aureus by NaCl (13), each isolate was screened on agar with and without 4% NaCl supplementation.

Plates were incubated at 35°C. Results were recorded after 24 h and 48 h of incubation. For any growth in excess of a single pinpoint colony on screening media, either with or without 4% NaCl supplementation, a positive result was recorded.

Susceptibility to vancomycin was determined by agar dilution, according to NCCLS guidelines (17), with Mueller-Hinton agar. Testing was performed on all MRSA bloodstream isolates (parent isolates), as well as on those colonies that grew on screening media (subclones). MIC testing was performed on colonies taken from the screening agar on which they exhibited optimal growth (i.e., BHI agar containing 4 mg/L vancomycin, with or without 4% NaCl).

The identity of parent isolates and subclones was confirmed by Gram stain, catalase testing (20), and latex agglutination testing (Staphaurex, Murex Biotech, Ltd., Dartford, UK). Strains were grouped by type and subtype by using pulsed-field gel electrophoresis (PFGE) (21). Plugs were made by using standard techniques. Macrorestriction was performed with SmaI (22).

Suspensions of 3.0 McFarland turbidity standard in BHI broth were made from overnight cultures of selected parent isolates and subclones. Subclones were grown in 4 mg/L vancomycin before suspension in broth. Seven serial 10-fold dilutions of each suspension were prepared. Twenty-five microliters of each suspension and dilution was injected twice onto BHI agar containing 4% NaCl, at vancomycin concentrations of 0 mg/L, 1 mg/L, 2 mg/L, 4 mg/L, and 8 mg/L. NaCl was added due to the enhanced growth we observed on NaCl-containing media during screening. Colonies were counted after 48 h of incubation at 35°C, and sums of each inoculum pair were averaged.

We conducted a two-part retrospective analysis on patients with MRSA bacteremia: a case-control analysis, which considered the clinical features at the time of culture, and a cohort analysis, which evaluated outcomes after culturing.

For the case-control study, patients whose MRSA bloodstream isolates exhibited growth on screening media (case-patients) were compared with patients whose isolates exhibited no growth (controls), in terms of the following features: age, sex, coexisting chronic conditions, hospital events, and antibiotic exposures before culture. Particular attention was given to exposure to vancomycin, examined both as a dichotomous variable and for cumulative days of exposure.

In the cohort study, we compared the above two groups of participants for the following postculture outcomes: deaths, discharge disposition, and duration of hospital stay after culture. Possible confounding variables were evaluated through multivariate modeling.

Statistical analyses were performed by using SAS statistical software (version 8e, SAS Institute, Inc., Cary, NC). Continuous variables were compared by using the Student t test or the Wilcoxon rank sum test, depending on the normality of the distribution. Binary variables were compared by using the Fisher exact test. Nonbinary categorical variables were compared by using the chi-square test. Multivariate logistic regression models were used in the death and disposition analyses to control for confounding. A Cox proportional hazards model was used in the time-to-discharge analysis; observations were censored at patient’s death. For all statistical tests, a p value of <0.05 was considered significant.

We tested 173 MRSA bloodstream isolates from 154 patients. For the following reasons, we excluded 24 isolates from the analysis: 19 represented additional isolates from patients already included in the cohort, 4 came from outpatients, and 1 came from a patient whose hospital stay extended beyond the study period. Thus, we evaluated 149 isolates, each cultured from the blood of a unique inpatient, and will describe them here.

All isolates were susceptible to vancomycin (MIC₅₀, 1 mg/L; MIC₉₀, 1 mg/L; range 0.5–2 mg/L). Isolates from 61 patients (41% of the patient cohort) grew on screening media within 48 h. Subclones of 46 of these isolates (75%) exhibited a two- to fourfold increase in MIC compared with the parent strain. No subclones, however, exceeded the 4 mg/L NCCLS breakpoint for vancomycin susceptibility (the MIC of vancomycin for all but one subclone was <2 mg/L; for one subclone it was 4 mg/L) (1). We therefore did not characterize any of the isolates as hetero-VISA. We characterized isolates with positive results on our screening assay as having heterogeneously reduced susceptibility to vancomycin.

The isolates comprised 11 different PFGE types. We assigned letters and numbers to PFGE types and subtypes, respectively. One hundred and three (69%) were variants of type A, which had 35 different subtypes. Twenty-two (15%) were variants of type B, which had seven different subtypes. The remaining 24 isolates (16%) consisted of nine different types. Type A was associated with the absence of heterogeneously reduced susceptibility (odds ratio [OR] 0.35; p=0.004), whereas type B was associated with its presence (OR 4.86; p=0.002). The results of strain typing and the relationship between type and screening results are summarized in Table 1. Figure 1 shows the results of PFGE performed on several isolates.

We performed population analysis on eight isolates (four that exhibited growth on screening media and four with no growth) to determine the utility or population analysis in distinguishing between these two types of isolates. The results, represented graphically in Figure 2a, do not enable such a distinction to be made. A shift in the curve is apparent, however, for two isolates with positive screening results from the same patient cultured 14 months apart. For the earlier isolate, the MIC of vancomycin was 0.5 mg/L. For the later isolate, cultured after several interval admissions for MRSA bacteremia in which the patient received vancomycin, the MIC was 1.0 mg/L, and an upwardly shifted population curve (signifying a greater proportion of the population with higher MICs of vancomycin) was observed. A subsequent population analysis, run on selected parent-subclone pairs for isolates that grew on screening media, demonstrated an upward shift in the population curve for the subclone versus the parent when a corresponding increase in MIC existed. A representative population curve is shown in Figure 2b.

A comparison of patients with isolates with positive screening results (cases) to those with negative results (controls) demonstrated no significant differences in terms of age, sex distribution, coexisting chronic conditions, recent hospitalization, or hospital events before culture (Table 2). Similarly, no differences were found between the groups with respect to administration of any antibiotic, administration of vancomycin specifically, or number of days of vancomycin exposure before culture (Table 3). In the cohort analysis of the impact of the screening phenotype on patient outcomes, no differences were noted between case-patients and controls in deaths during hospitalization, discharge disposition, or time to discharge after culture for those who survived (Table 4).

Given the absence of standardized criteria for determining the screening phenotype, we further analyzed the data using a number of more stringent definitions. If we considered growth on screening media within 48 h with an associated increase in MIC among the subclones as the criteria for a positive screening result (n=46), again no significant predictors of the phenotype and no differences in outcomes were found. Twenty patients (13%) had isolates that grew on screening media within 24 h. Using this characteristic as the screening criterion, we found that the only significant predictor of this phenotype was intensive care unit stay before culture (OR 2.95; 95% confidence interval [CI] 1.07 to 8.15; p=0.05); no differences in outcomes were found (data not shown).

Finally, since the criteria for isolate collection were not identical in the two study institutions, we performed a subgroup analysis using data from Beth Israel Deaconess Medical Center patients alone (n=120). This analysis did not change the results of using the 48-h growth cutoff. Analysis of these data using the 24-h cutoff showed diabetes mellitus to be the only significant predictor of heterogeneously reduced susceptibility to vancomycin (OR 3.52; 95% confidence interval, 1.04 to 11.96; p=0.05), with no differences in outcomes (data not shown).