Rapid Detection of Vancomycin-Intermediate Staphylococcus aureus by Matrix-Assisted Laser Desorption Ionization–Time of Flight Mass Spectrometry

Abstract

Vancomycin is the standard of care for the treatment of invasive methicillin-resistant Staphylococcus aureus (MRSA) infections. Infections with vancomycin-nonsusceptible MRSA, including vancomycin-intermediate S. aureus (VISA) and heterogeneous VISA (hVISA), are clinically challenging and are associated with poor patient outcomes. The identification of VISA in the clinical laboratory depends on standard susceptibility testing, which takes at least 24 h to complete after isolate subculture, whereas hVISA is not routinely detected in clinical labs. We therefore sought to determine whether VISA and hVISA can be differentiated from vancomycin-susceptible S. aureus (VSSA) using the spectra produced by matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS). Strains of MRSA were characterized for vancomycin susceptibility phenotype by broth microdilution and modified population analysis. We tested 21 VISA, 21 hVISA, and 38 VSSA isolates by MALDI-TOF MS. Susceptibility phenotypes were separated by using a support vector machine (SVM) machine learning algorithm. The resulting model was validated by leave-one-out cross validation. Models were developed and validated by using spectral profiles generated under various subculture conditions, as well as with and without hVISA strains. Using SVM, we correctly identified 100% of the VISA and 97% of the VSSA isolates with an overall classification accuracy of 98%. Addition of hVISA to the model resulted in 76% hVISA identification, 100% VISA identification, and 89% VSSA identification, for an overall classification accuracy of 89%. We conclude that VISA/hVISA and VSSA isolates are separable by MALDI-TOF MS with SVM analysis.

INTRODUCTION

Delayed initiation of appropriate antibiotic therapy results in increased mortality rates for patients with sepsis (1–6). Staphylococcus aureus is a pathogen frequently isolated from patients with sepsis, with worse clinical outcomes noted for patients with methicillin-resistant S. aureus (MRSA) (7). Vancomycin remains the standard of care for the treatment of invasive infections with MRSA (8, 9). Infections with vancomycin-nonsusceptible isolates are associated with prolonged bacteremia, longer hospital stays, and greater rates of clinical treatment failure than infections with vancomycin-susceptible S. aureus (VSSA) (10, 11).

Vancomycin nonsusceptibility falls into two related phenotypes: vancomycin-intermediate S. aureus (VISA) and heterogeneous VISA (hVISA). The VISA phenotype is reliably detected by broth microdilution, and these strains are characterized by a vancomycin MIC of 4 or 8 μg/ml. Reliance on susceptibility testing to identify the VISA phenotype therefore delays the identification of these strains and thus the initiation of appropriate antimicrobial therapy. Because only a subpopulation of cells of hVISA strains (≤10⁻⁵ to 10⁻⁶) have vancomycin MICs in the intermediate range, hVISA isolates frequently test susceptible to vancomycin (i.e., MICs of ≤2 μg/ml) by broth microdilution methods, which typically test only approximately 10⁴ cells. Consequently, patients with hVISA infections may never be identified, despite having higher vancomycin treatment failure rates than patients with VSSA bloodstream infections (12). Unlike methicillin resistance and frank vancomycin resistance, which arise through the acquisition of discrete genetic elements (mecA and vanA, respectively), the genetic changes that lead to manifestation of the VISA and hVISA phenotypes are varied and not easily detectible by a single gene assay (summarized in reference 13). While these genetic alterations are variable, they tend to occur in regulatory genes that control cell metabolism and cell wall structure and function. Consequently, most VISA and hVISA isolates share common phenotypic features such as increased cell wall thickness and lowered growth rate (13).

Matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS) is increasingly used by clinical microbiologists in the identification of bacterial and yeast isolates (14, 15). Analysis of microorganisms by MALDI-TOF MS results in the production of a protein signature that is characteristic of the species or genus of the microorganism being tested. While initial investigations using MALDI-TOF MS suggested the presence of proteomic differences between VSSA and VISA strains (16), this work focused on small molecules (≤2,000 Da) and did not use MALDI-TOF MS instruments or analysis parameters that are routinely used in clinical microbiology laboratories. The primary goals of this study were therefore to determine whether proteomic differences between VISA and VSSA can be detected by MALDI-TOF MS and, if so, to develop a MALDI-TOF MS-based tool that can be used to rapidly identify VISA strains.

MATERIALS AND METHODS

Bacterial strains and growth conditions.

All of the strains used in this study were MRSA isolates from frozen stocks that had been stored in 50% glycerol and/or Mueller-Hinton (MH) broth at −80°C. All of the isolates were originally obtained from clinical specimens (predominantly from blood cultures) and had been tested by standard susceptibility methods prior to freezing. These strains were obtained from two geographically separate clinical laboratories, as well as the Network on Antimicrobial Resistance in Staphylococcus aureus (NARSA), which collects strains from laboratories around the world.

The culture media used in this study were sheep blood (tryptic soy agar with 5% sheep blood), MH, and brain heart infusion (BHI) agars (all from Remel, Lenexa, KS). Isolates were incubated at 35°C with 5% CO₂ for 20 to 24 h prior to extraction and testing. All of the isolates previously identified as having the VISA phenotype were confirmed to have maintained this phenotype after freezing by vancomycin broth microdilution. All of the isolates previously identified as having VSSA and hVISA phenotypes by the modified population analysis method (described below) were retested by this method after freezing. Only those isolates that maintained the hVISA phenotype were included in this study. To minimize the potential for loss of the vancomycin-nonsusceptible phenotype, isolates were also analyzed by MALDI-TOF MS after the initial subculture from frozen stocks.

PAP-AUC testing.

All strains with a vancomycin MIC of ≤2 μg/ml by broth microdilution were subjected to testing for hVISA by the modified population analysis method as described previously (17). Briefly, colonies were suspended in sterile saline to achieve a 2 McFarland standard suspension (∼5 × 10⁸ CFU/ml) and spiral diluted (EasySpiral; Interscience, France) onto BHI agar supplemented with 0.5 to 8 μg/ml vancomycin. After 48 h of incubation at 35°C, surviving colonies were enumerated with an automated plate reader (Scan500; Interscience, France). Surviving colonies were plotted against vancomycin concentrations to generate population analysis profiles (PAPs), and the area under the curve (AUC) for each strain was calculated by the trapezoidal method. AUC ratios were determined by dividing the AUC of the sample strain by the AUC of the archetype hVISA strain, Mu3, which was run with each experiment. Any strain with an AUC ratio of ≥0.9 was considered to be positive for the hVISA phenotype.

MALDI-TOF MS analysis.

A 1-μl loopful of bacteria was used for extraction with ethanol-formic acid-acetonitrile as described previously (18), with the modification that 10 μl each of formic acid and acetonitrile was used for the final pellet resuspension. One microliter of the final supernatant was spotted in duplicate onto a 96-well polished stainless steel target plate and overlaid with 1 μl of a saturated solution of α-cyano-4-hydroxycinnamic acid in 50% acetonitrile and 2.5% trifluoroacetic acid (both from Bruker Daltonics, Billerica, MA). Spectra were acquired in a linear positive-ion mode at a laser frequency of 60 Hz across a mass/charge (m/z) ratio of 2,000 to 20,000 with a MicroFlex LT mass spectrometer (Bruker Daltonics) as described previously (19). For each spectrum, 240 laser shots in 40-shot steps from different areas of the sample spot were accumulated and analyzed (automatic mode, default settings). A total of five individual spectra were acquired for each spot analyzed. All experiments were repeated in biological duplicate, for a total of 20 spectra per isolate.

Spectral analysis.

Spectra were analyzed with the MALDIQuant program in R (20, 21). Spectra were trimmed to 2,000 to 16,000 Da, with intensities transformed using the square root function. Spectra were smoothed with the Savitzky-Golay filter (22), and baseline noise was subtracted using statistics-sensitive nonlinear iterative peak clipping (23). Intensity was calibrated by the total ion current method so that individual spectra could be compared even when the absolute peak heights were dissimilar.

Peaks were detected with a signal to noise ratio of 2 to maximally identify peaks while decreasing the noise of the spectra. Only peaks present in 80% of the spectra for a given isolate (≥16/20) were considered for further analysis. For each isolate, all peaks meeting this threshold were merged into one representative spectrum. Because of the inherent inaccuracy of precise mass detection by MALDI-TOF MS (24), all spectra were warped by using the included warpingFunctions command of MALDIQuant prior to peak binning to ensure that identical peaks with slightly different mass ratios in different spectra were binned together.

To evaluate peaks that are characteristic of VISA, hVISA, and VSSA, peak selection was accomplished by forward stepwise multiple regression with built-in functions of R (reference peaks). Peaks present in ≥75% of the spectra that had a coefficient of variation (CV) of <50% within a given class (i.e., VISA, VSSA, or hVISA) were included for consideration (modified from reference 25). Peaks meeting these criteria were initially evaluated by principal-component analysis (PCA) with the vegan package in R (26). Clustering of VISA, hVISA, and VSSA groups were plotted against the first and second principal components.

To ensure comparability between different runs, peak values of each run were compared to those of the reference peaks, and the value of any test peak within 5 Da of a reference peak was replaced with that of the reference peak with the tidyr program in R (27).

Development of a machine learning approach to distinguish VISA and VSSA strains.

We developed a MALDI-TOF MS-based machine learning approach to discriminate VISA and VSSA strains by using the peaks that met the selection criteria described in the previous section. In total, 21 VISA and 38 VSSA strains were used for the initial support vector machine (SVM) model (i.e., VISA-versus-VSSA SVM), with a second machine learning model subsequently built that also incorporated spectra from 21 hVISA strains (VISA/hVISA-versus-VSSA SVM). SVM was run with the kernlab program in R (28) with the radial basis function (rbfdot) kernel to allow nonlinear mapping of the data and a cost of 10 to avoid overfitting the model. The SVM predictions were output as probabilities and exported to a comma-separated file with the MASS program in R (29).

Strain typing.

Because MALDI-TOF MS has been used successfully for strain typing (30), we wanted to ensure that classification of VISA, hVISA, and VSSA by our machine-based learning approach was not simply due to the detection of differences in the strain type. For nine VSSA and nine hVISA isolates, amplification of the spa gene X region was performed as described previously (31) and amplicons were sequenced with a 3130xl genetic analyzer (Applied Biosystems, Foster City, CA). The spa types were assigned with the Ridom SpaServer (32). For the nine VISA isolates and one hVISA isolate from the NARSA collection, the spa types were obtained from NARSA and repeats were defined with the Ridom SpaServer. Repeat regions were aligned with ClustalX (33). The spa types with similar or identical repeat profiles were grouped into clusters and mapped onto a dendrogram with the APE program in R (34).

RESULTS

Pilot study to identify the optimal medium type for analysis of VISA and VSSA by MALDI-TOF MS.

To identify the optimal medium that maximizes potential mass spectral differences between VISA and VSSA strains, we selected 16 VISA and 12 VSSA isolates for analysis following growth on sheep blood, MH, and BHI agars, as well as MH and BHI agars supplemented with 0.5 and 1.0 μg/ml vancomycin (MH0.5, MH1.0, BHI0.5, and BHI1.0, respectively). However, because only 7 (58%) of the 12 VSSA isolates were capable of growth on all six media, these 7 isolates were used for subsequent analyses. As shown in Table 1, between 213 and 247 total peaks were identified on the different medium types.

TABLE 1.

Comparison of VISA and VSSA isolates on six culture medium types^a

Parameter	BHI0.5	BHI1.0	Sheep blood	MH	MH0.5	MH1.0
Total no. of peaks	213	247	220	237	216	231
No. of 75CV peaksb	60	59	68	47	59	69

^aSpectral data from 12 VISA and 7 VSSA isolates were used to perform stepwise forward multiple regression analysis to identify the peaks that best separated the two groups.

^bPeaks present in ≥75% of VISA and/or VSSA isolates with a CV of <50% within each group.

To evaluate peaks that are representative of VISA and VSSA, the subsequent analysis was limited to those peaks present in ≥75% of either group (i.e., VISA or VSSA) with a CV of <50% within that group (Table 1). Using forward stepwise multiple linear regression, we identified 22 peaks that maximized the separation of VISA and VSSA isolates. PCA of VISA and VSSA isolates with these 22 peaks produced separate but overlapping groups on all of the agar types used, although these groups were most cleanly separated on MH agar (Fig. 1). The inclusion of vancomycin did not appear to improve the separation of these groups by MALDI-TOF MS (Fig. 1). As MH agar is readily available in many clinical microbiology laboratories, we chose to use this medium in subsequent analyses.

FIG 1.

PCA comparison of VISA and VSSA isolates grown on six different culture medium types. PCA of 12 VISA and 7 VSSA isolates grown on six culture medium types (BHI0.5, BHI1.0, sheep blood, MH, MH0.5, and MH1.0) was performed. Ellipses denote the 90% confidence intervals around the groups.

Analysis of VISA and VSSA proteomic profiles by MALDI-TOF MS analysis.

In order to compare the proteomic profiles of VISA and VSSA on the basis of MALDI-TOF MS, we extended our analysis to include a total of 21 VISA and 38 VSSA isolates. The vancomycin susceptibility phenotype of these isolates was confirmed pre- and poststorage at −80°C. We identified 60 peaks that were present in ≥75% of a given group with an intragroup CV of <50%. None of these peaks were exclusively unique to VISA or VSSA, although one peak (8,258 Da) was present in 81% of the VSSA isolates and only 10% of the VISA isolates, and one peak (4,540 Da) was present in 84% of the VISA isolates and only 18% of the VSSA isolates (Table 2). Nineteen of these 60 peaks, however, showed highly significant differences in height between VISA and VSSA isolates (P < 0.0001; data not shown). Forward stepwise regression analysis of these 60 peaks identified 14 that best separated the VISA and VSSA isolates (Table S1 in the supplemental material). As shown in Fig. 2, PCA of the VISA and VSSA isolates with these peaks showed two separate but overlapping groups.

TABLE 2.

Comparison of spectra from VISA and VSSA isolates

Parameter	Value
Total no. of peaks	449
No. of 75CVa peaks	60
Molecular size (Da) of peak differentially present in:
VISA	4,540b
VSSA	8,258c
No. of peaks with P < 0.0001 between VISA and VSSA	19

^aPeaks present in ≥75% of VISA and/or VSSA with CV <50% within each group.

^b84% VISA, 18% VSSA.

^c10% VISA, 81% VSSA.

FIG 2.

PCA of VISA and VSSA isolates definitively identified on MH agar. PCA of 21 VISA and 38 VSSA isolates with reproducible phenotypes separated by using the 14 peaks identified by forward stepwise multiple regression was performed. Ellipses denote the 90% confidence intervals around the groups.

VISA and VSSA can be reliably differentiated by MALDI-TOF MS.

SVM analysis with the 14 peaks that optimally separated VISA from VSSA (VISA-versus-VSSA SVM) provided 100% separation of VISA and VSSA isolates. Leave-one-out cross validation (LOOCV), in which each isolate is serially removed from the SVM analysis to test the robustness of the model, resulted in the correct classification of 100% of the VISA isolates (21/21) and 97% of the VSSA isolates (37/38), for an overall classification accuracy of 98% (Table 3).

TABLE 3.

Classification based on VISA versus VSSA and VISA/hVISA versus VSSA SVM models

Test condition	Classification of isolates [no. correct/total (%)]
	VISA	VSSA	hVISA
VISA vs VSSA SVMa without hVISA	21/21 (100)	38/38 (100)	NAb
LOOCV VISA vs VSSA SVM without hVISA	21/21 (100)	37/38 (97)	NA
hVISA identification with VISA vs VSSA SVM	NA	NA	13/21 (62)
hVISA/VISA vs VSSA SVMc	21/21 (100)	38/38 (100)	21/21 (100)
LOOCV hVISA/VISA/VSSA SVM	21/21 (100)	34/38 (89)	16/21 (76)

^aSVM performed with the 14 peaks identified by forward stepwise multiple regression to maximally separate VISA from VSSA.

^bNA, not applicable.

^cSVM performed with the 25 peaks identified by forward stepwise multiple regression to maximally separate hVISA/VISA from VSSA.

In order to assess whether MALDI-TOF MS can identify hVISA isolates, we analyzed 21 hVISA isolates with the VISA-versus-VSSA SVM. With this model, 62% of the hVISA isolates (13/21) were classified as VISA. We hypothesized that hVISA may have been suboptimally classified by the VISA-versus-VSSA regression data because hVISA isolates were not specifically grouped with VISA isolates during the regression analysis. We therefore repeated the forward stepwise regression with a binary hVISA/VISA-versus-VSSA separation. This identified 25 peaks that optimally separated the two phenotypes (see Table S1 in the supplemental material). As with the VISA-versus-VSSA classifier, we achieved 100% separation of hVISA/VISA and VSSA when all of the isolates were included in the analysis. LOOCV resulted in the correct identification of 76% of the hVISA isolates (16/21), with 100% identification of the VISA isolates (21/21) and 89% identification of the VSSA isolates (34/38), for an overall classification accuracy of 89% (71/80) (Table 3).

Strain type does not account for the proteomic differences observed between VISA/hVISA and VSSA isolates.

We performed spa typing of a subset of randomly selected hVISA and VSSA strains used in our study; spa typing of one hVISA and nine of the VISA isolates used had previously been performed. As shown in Fig. 3 (also see Table S2 in the supplemental material), VISA, VSSA, and hVISA isolates were well represented in each of the clades we observed upon spa typing. Thus, classification of the vancomycin susceptibility phenotype by our machine learning model does not appear to be due to differences in the strain type.

FIG 3.

Dendrogram of the spa types of 10 hVISA, 9 VISA, and 9 VSSA isolates randomly selected from the strains used in this study. As shown by the dendrogram, the spa types of the different isolate types (VISA, hVISA, and VSSA) spanned multiple clusters and VISA, hVISA, and VSSA isolates were found within each major cluster.

DISCUSSION

In this study, we have shown that there are significant proteomic differences between VISA and VSSA isolates detected by MALDI-TOF MS. Furthermore, we were able to achieve 98% classification accuracy of VISA and VSSA isolates by MALDI-TOF MS using a proteomics-based machine learning approach. That we were able to achieve this classification accuracy even in the absence of selective pressure is consistent with the previously published finding that VISA and VSSA isolates exhibit differences in gene expression in the absence of vancomycin (35). It is interesting that we observed the cleanest separation of VISA and VSSA phenotypes when using MH agar. The reasons for this are unclear; however, we speculate that because this agar is less rich than the BHI and sheep blood agars, it is possible that the different strains may revert to a “basal” level of gene expression that is fundamentally different between VISA and VSSA strains.

Many previous attempts to identify antibiotic-resistant bacteria by MALDI-TOF MS have focused on the presence or absence of specific spectral peaks (16, 36, 37). Only two peaks were differentially present in VISA and VSSA isolates in our study, and these were not exclusive to either group. This is similar to the findings of Lu and colleagues, who previously reported the presence of two intense peptide peaks that were present in 88% of the VISA strains analyzed but also present in up to 20% of the VSSA strains analyzed (16). Instead, we evaluated the presence and/or absence of any given peak, as well as the peak height, standardized between spectra using the total ion current calibration. This allowed us to evaluate differences in protein quantity between the two groups, even when the actual proteins expressed were the same. By looking only at those peaks that maximized the distinction between VISA and VSSA isolates, even closely related spectra could be distinguished. We found 19 peaks with significantly different expression in VISA and VSSA strains, highlighting the differences in protein expression between these two groups. These data are consistent with previously published reports that showed increased abundance of numerous proteins by two-dimensional gel analysis of isogenic VISA and VSSA strains (38). The use of SVM for this purpose also allowed the sensitive discrimination of VISA and VSSA isolates even in the setting of complex proteomic signatures within the two groups, as nonlinear SVM can find optimal boundaries between two groups that would not be possible with a more rigid classifier.

To our knowledge, this is the first study to attempt to differentiate VISA and VSSA strains by MALDI-TOF MS with an instrument commonly used for routine microorganism identification. Furthermore, our study used a commercially available culture medium routinely found in many clinical microbiology laboratories (MH agar). Assay and analysis can be completed in as little as 60 min for less than $1/sample, which is substantially faster, less expensive, and more accurate than the recently reported multigene molecular assay designed to separate VISA from VSSA (98% versus 84% accuracy, respectively) (39). For clinical laboratories performing identification of MRSA directly from positive blood cultures (e.g., microarray, etc.), subculture of the positive blood culture broth to MH agar could be performed and subjected to analysis upon growth of the colonies, resulting in potential detection of the VISA phenotype 1 day prior to the availability of MIC testing results.

The identification of hVISA is known to be problematic because this phenotype is not generally detected by standard susceptibility testing methods. The gold standard method of identification, the PAP-AUC, is time-consuming, expensive, and impractical for implementation in clinical microbiology laboratories. However, up to 50% of isolates with a vancomycin MIC of 2 µg/ml (susceptible by CLSI standards) can harbor hVISA clones by PAP-AUC (40). Thus, an assay that can reliably detect these isolates in the clinical microbiology laboratory is of considerable interest. While a number of screening methods for hVISA have been published (e.g., BHI agar supplemented with 3 μg/ml vancomycin), these typically require prolonged incubation (48 h) (41, 42). We have shown that our assay can detect between two-thirds and three-quarters of the PAP-AUC-confirmed hVISA isolates. This is comparable to or better than other methods used to screen for this phenotype, all of which require additional growth of the isolate in vitro (41–43).

The less reliable classification of hVISA isolates by MALDI-TOF MS may potentially be explained by the infrequent distribution of nonsusceptible organisms in the hVISA population (1 × 10⁻⁵ to 1 × 10⁻⁶) (44). As 10⁴ to 10⁶ CFU of the test microorganism are generally required for analysis by MALDI-TOF MS (45, 46), it is likely that the overwhelming majority of susceptible clones obscured the signal from the nonsusceptible ones. Although the hVISA phenotype is known to be unstable during freezing, we confirmed retention of the hVISA phenotype after freezing. Interestingly, the inclusion of low-level vancomycin (0.5 μg/ml) was insufficient to enrich these clones to a level at which they were easily detectable by SVM (data not shown). Similarly, the use of higher concentrations of vancomycin (up to 2.0 μg/ml) also failed to improve performance (data not shown). Despite this limitation, more than three-quarters of the PAP-AUC-confirmed hVISA isolates were classified in the nonsusceptible category when the hVISA/VISA-versus-VSSA SVM was used, suggesting that there are discernible proteomic differences between many hVISA and VSSA isolates.

Our study has several strengths. First, unlike other studies in which MALDI-TOF MS was used to predict antibiotic resistance (16, 36, 37), our study did not focus solely on peak presence or absence as a method of classification. Second, all of the strains used in our study were fully characterized for vancomycin susceptibility by the PAP-AUC method. Third, we acquired spectra with the same MALDI-TOF MS settings used for bacterial identification, which could allow this technique to be incorporated into clinical practice with minimum modification of the current protocols. We also used open-source software to manipulate the spectra and run the SVM assay, which decreases the cost of introducing this technique into the clinical laboratory. From the standpoint of the clinical laboratory scientist, use of the program requires minimal training, with the program being run as a script in R and spectra obtained in a clinical run being run against the established model. Finally, we also ensured that our findings were not simply due to strain discordance between the VISA and VSSA groups, which has previously proven to be a potential problem with identifying antibiotic resistance in S. aureus (36, 47).

One limitation of our study is that we did not thoroughly evaluate whether changing the vancomycin concentration or base agar would improve the detection of hVISA isolates, though preliminary data suggested that this would not affect the results (Fig. 1 and data not shown). We also did not evaluate whether the use of alternate criteria for peak inclusion would improve the performance of the SVM for detection of hVISA. A further limitation of our study is that we did not use direct cell analysis. Instead, we extracted all of our isolates, which modestly increases the time and expense of the assay. However, extraction is known to improve spectrum reproducibility and increase the number of peaks and mass range width for S. aureus isolates (48). Finally, we did not fully evaluate whether the use of a single spectrum per isolate tested could achieve similar results.

With the hVISA/VISA-versus-VSSA SVM, we were able to identify 76% of the hVISA, 100% of the VISA, and 89% of the definitively classified VSSA isolates tested. Though the actual incidence of hVISA and VISA among clinical specimens is difficult to ascertain and varies widely between studies, the prevalence of hVISA is generally thought to be between 2 and 20% and that of VISA to be between 0.5 and 2% (49–51). With an estimated hVISA prevalence of 5% and an estimated VISA prevalence of 1%, the algorithm we developed would result in a negative predictive value (NPV) of 98.6% and a positive predictive value of 31.8%. This means that any isolate classified as VSSA by our assay could be confidently treated with standard vancomycin therapy, but that any isolate called hVISA/VISA would require further testing. Even if the prevalence of hVISA were 20% and the prevalence of VISA were 2% (the top end of most estimates), then the NPV would still be 93.5%. Given that all VISA isolates would be detected regardless of prevalence, there would be no worry of misidentifying a VISA isolate as VSSA, and the detection of hVISA would likely be much greater than it currently is in most clinical microbiology laboratories. Earlier detection of the VISA phenotypes has the potential to improve clinical decision making and patient outcomes. However, additional studies are needed to clinically validate this test prior to integration into the standard workflow.