ABSTRACT
Antimicrobial susceptibility testing (AST) assigns pathogens to “susceptible” or “resistant” clinical categories based on clinical breakpoints (CBPs) derived from MICs or inhibition zone diameters and indicates the likelihood for therapeutic success. AST reports do not provide quantitative measures for the reliability of such categorization. Thus, it is currently impossible for clinicians to estimate the technical forecast uncertainty of an AST result regarding clinical categorization. AST error rates depend on the localization of pathogen populations in relation to CBPs. Bacterial species are, however, not homogeneous, and subpopulations behave differently with respect to AST results. We addressed how AST reporting errors differ between isolates with and without acquired drug resistance determinants. Using as an example the beta-lactams and their most important resistance mechanisms, we analyzed different pathogen populations for their individual reporting error probabilities. Categorization error rates were significantly higher for bacterial populations harboring resistance mechanisms than for the wild-type population. Reporting errors for amoxicillin-clavulanic acid and piperacillin-tazobactam in Escherichia coli infection cases were almost exclusively due to the presence of broad-spectrum- and extended-spectrum-beta-lactamase (ESBL)-producing microorganisms (79% and 20% of all errors, respectively). Clinicians should be aware of the significantly increased risk of erroneous AST reports for isolates producing beta-lactamases, particularly ESBL and AmpC. Including probability indicators for interpretation would improve AST reports.
INTRODUCTION
Reports of antimicrobial susceptibility testing (AST) data have been used for decades to translate in vitro laboratory results into predictions of clinical outcome by interpretative categorization of pathogens as “susceptible” or “resistant” based on clinical breakpoints (CBPs) (1, 2). The “susceptible” category implies a high likelihood for therapeutic success if standard dosing is applied, whereas the “resistant” category implies probable therapeutic failure (3, 4). In contrast to other laboratory tests, such as quantitative PCRs or those employing serologic parameters, for which variation coefficients indicating measurement precision are usually reported, or can at least be requested, AST reports do not provide such indicators (5, 6). As a consequence, it is currently impossible for the clinician to estimate the forecast uncertainty of an AST categorization and the related reliability of the predicted clinical outcome. AST categorizations based on MIC or inhibition zone measurements close to the CBPs are reported as “susceptible” or “resistant” without further comment, although the statistical-error probability increases the closer a MIC or zone diameter measurement is to the CBP (7). In consequence, providing levels of reliability for AST categorization would improve clinical decisions in antimicrobial therapy.
For a long time, AST categorization was based not only on MIC and zone diameter measurements but also on the detection of individual resistance mechanisms, i.e., interpretative reading (8 to 10). Even if in vitro results indicated susceptibility to a drug, the reported category was edited to “resistant” if the presence of a resistance mechanism was confirmed, e.g., in the case of extended-spectrum beta-lactamases (ESBLs). While this strategy is retained, e.g., for methicillin resistance in Staphylococcus aureus, CLSI and EUCAST recently abandoned editing of AST reports based on the detection of ESBLs (3, 11 to 13). AST categorization now depends solely on MIC and/or zone diameter measurement. However, AST methods show considerable technical and biological variations (2, 7, 14 to 16). The rate of major, very major, or minor clinical categorization errors for a given antibiotic depends on (i) the presence, width, or absence of a “gray” (or intermediate) zone between the susceptible and resistant categories, (ii) the relative position of a population with respect to the CBP, and (iii) the sum of the values corresponding to methodological imprecision and biological variations (7). Since it is reasonable to assume that subpopulations may behave differently with respect to the latter two aspects, we addressed to what extent AST reporting error rates differ between wild-type isolates (no acquired drug resistance determinants) and isolates encoding a resistance mechanism (termed “resistotypes” in this work). Beta-lactam resistance in the three most prevalent Enterobacteriaceae species was chosen as a model, since the treatment of ESBL producers with penicillin-inhibitor combinations and/or newer cephalosporins is the subject of an ongoing debate (17 to 21).
MATERIALS AND METHODS
Bacterial strains.In total, 7,148 nonduplicate clinical isolates recovered in our laboratory between 2010 and 2013 were included in the study, comprising 4,287 Escherichia coli, 1,886 Klebsiella pneumoniae, and 975 Enterobacter cloacae isolates. Isolates of the same species were considered duplicates if they (i) originated from the same patient and (ii) showed at most one major (susceptible versus resistant) and two minor (susceptible versus intermediate and/or intermediate versus resistant) differences in AST interpretations. The numbers of the various phenotypes for a species can be retrieved from Table 1. For some isolates, not all zone diameters were available, resulting in lower numbers of data for certain drug/species combinations (Table 1).
Mean zone diameters of wild-type and resistant populations, distance from mean zone diameters to clinical breakpoints, and associated error ratesa
Susceptibility testing.Susceptibility testing was done by disk diffusion according to EUCAST recommendations (22). Antibiotic disks and Mueller-Hinton agar were obtained from Becton Dickinson, Franklin Lakes, NJ. Inhibition zone diameters were recorded using the Sirweb/Sirscan system (i2a, Montpellier, France) to standardize reading precision (23). ESBL and AmpC detection was done as previously described (24, 25). The reading precision of the Sirscan instrument was determined by 100 repeat measurements of the same plate and inhibition zone (see Table 2).
Measurement precision of disk diffusion testing displayed as median diameter values and 1-fold standard deviations of 100 repeat measurements for clinical wild-type and BSBL-, ESBL-, and AmpC-producing E. coli strains and 120 repeat measurements of E. coli ATCC 25922
Interpretation errors.Differences in clinical categorization, referred to as “discrepancies,” are split according to therapeutic implications. Those resulting in erratic assignment of bacterial isolates to adjacent interpretative categories (susceptible to intermediate, intermediate to susceptible, intermediate to resistant, resistant to intermediate) are referred to as “minor errors.” Erroneous categorization of true-susceptible isolates as resistant are referred to as “major errors” leading to unnecessary restriction of therapeutic options. The most serious clinical implications result from “very major errors,” i.e., categorization of true-resistant isolates as susceptible, as there is a high likelihood of therapeutic failure.
Phenotype/resistotype definition.E. coli isolates were considered wild type, i.e., devoid of mechanisms of resistance to beta-lactams, if the isolates were susceptible to ampicillin (according to EUCAST CBPs) to exclude the presence of broad-spectrum beta-lactamases (BSBLs) such as TEM-1/2 or SHV-1 and if the isolates were susceptible to cefoxitin (according to the EUCAST CBPs) to exclude porin deficiency and/or overexpression of efflux systems. To exclude cephalosporinases, isolates were required to be wild type with respect to cephalothin per the in-house epidemiological cutoff (ECOFF) value of 10 mm (30-μg disk; data not shown). K. pneumoniae isolates were considered wild type (i.e., chromosomally encoded SHV-1 beta-lactamase producers only) if ESBL, AmpC, and carbapenemase production had been excluded (24–27) and if they were susceptible to cefoxitin as a marker for porin deficiency according to EUCAST CBPs. E. cloacae was considered wild type if ESBL and carbapenemase production had been excluded (24–26, 28) and if they were susceptible to cefpodoxime according to EUCAST CBPs to exclude isolates with derepressed chromosomal AmpC production. E. cloacae isolates susceptible to cefpodoxime were checked for resistance to ceftriaxone, cefotaxime, ceftazidime, and cefepime and found susceptible. Any colony within the inhibition zone of a third-generation cephalosporin was counted for zone diameter measurement. BSBL-producing E. coli isolates were defined as ampicillin resistant, non-ESBL, non-AmpC, carbapenemase negative, and cefoxitin susceptible. E. cloacae isolates with derepressed AmpC production were defined as cefpodoxime resistant, non-ESBL, and carbapenemase negative.
Statistical analysis.Error probabilities were calculated separately for all diameter values adjacent to the interpretative category borders as described previously (7). In brief, theoretical upper and lower tails of the standard normal distribution were used to calculate probability distributions for the actual diameter measurements, which can be assumed to be normally distributed around the mean diameters obtained by disk diffusion with respect to different cutoff values (29, 30). Standard deviations of 120 independent measurements of E. coli ATCC 25922 (EUCAST quality control [QC] strain) are listed in Table 2. Standard deviations of E. coli ATCC 25922 were in general agreement with those derived from EUCAST QC tables and were thus used as the basis for probability calculations (31). In addition, 100 independent measurements of nine clinical E. coli strains (three wild-type isolates, two BSBL-producing isolates, three ESBL isolates, and one AmpC-producing isolate) were included to test for the influence of the genotype on measurement precision. To test for statistical significance, the Mann-Whitney U test was applied.
Software.All calculations were done using IBM SPSS statistics software version 20 (IBM Corporation, Armonk, NY) and Microsoft Excel 2010 software (Microsoft Corporation, Redmond, WA).
RESULTS
Disk diffusion diameters for a total of 7,148 nonduplicate clinical isolates were used in the study. Statistical probabilities for antibiogram misclassifications were determined for individual subpopulations, i.e., resistotypes (wild type, BSBL, ESBL, and AmpC), as previously described (7).
The total number of interpretation errors for a given species/drug combination depends on four variables: (i) technical measurement precision, (ii) range of population distributions, (iii) distance of the population's lowest diameter or MIC value from the next CBP, and (iv) relative resistotype prevalence. A paradigmatic example is depicted in Fig. 1: higher measurement precision lowers the probability of assigning a population's tail area to the false category (Fig. 1A); larger diameter ranges of a population lead to higher numbers of isolates in the area at risk (Fig. 1B); a low distance of a population from the CBP increases the probability of assigning a population's tail area to the false category (Fig. 1C); and the higher the prevalence of a resistotype clustering around the CBP, the higher the number of isolates in the area at risk for interpretation errors (Fig. 1D).
Schematic representation of variables influencing interpretation error rates. Variables influencing interpretation errors are indicated as follows: technical measurement precision (A), population distributions (B), distance of a population's median diameter from the next CBP (C), and relative resistotype prevalences (D1 and D2). (A) Higher measurement precision, i.e., a lower standard deviation, will result in a smaller area at risk for errors around the CBP (blue dotted curve) compared to lower measurement precision (red dotted curve). (B) Higher population variation (red bars) will result in a higher error risk than lower population variation (blue bars) as more isolates are located in the area at risk for errors (dotted black curve). (C) The closer a population to the CBP, the more isolates are at risk for errors. CBP 1 (blue) will result in fewer errors than CBP 2 (red). (D1 and D2) Influence of the prevalence of subpopulations/resistotypes on error rates, ESBL-producing organisms (red bars), and wild-type organisms (blue bars). The black vertical line indicates the EUCAST clinical breakpoint for 2010 to 2013 for amoxicillin-clavulanic acid (≥17-mm diameter, susceptible; <17-mm diameter, resistant). The dotted curve indicates the error probability around the clinical EUCAST breakpoint. (D1) E. coli ESBL prevalence, 9%; E. coli wild-type strain prevalence, 32%; data for other resistotypes (BSBL and AmpC; total prevalence, 59%) are not shown. The total error rate for all E. coli isolates with respect to amoxicillin-clavulanic acid was 2.5% (see Table 1). (D2) The ESBL prevalence was modeled at 60% and the wild-type prevalence at 1% (data for other resistotypes, i.e., BSBL and AmpC [total prevalence, 39%], are not shown). The total error rate increased to 4.8%.
Technical measurement precision values were similar for all resistotypes as reflected in the average standard deviations of the resistotypes (see Table 2). In the critical range for interpretation errors, i.e., diameter values flanking CBPs (15 to 25 mm), 1-fold standard deviations of all resistotype/drug combinations ranged from 1.0 to 1.5 mm. This value range is consistent with EUCAST methodological precision requirements (QC ranges of 4 to 6 reflecting a zone of a 2-fold standard deviation range around the target value, i.e., a 1-fold standard deviation of 1 to 1.5 mm) (31). Therefore, the standard deviations for the EUCAST QC E. coli ATCC 25922 strain were considered a reasonable basis for error calculations.
Figure 2 depicts the above-mentioned factors for the E. coli study population and amoxicillin-clavulanic acid: the closer the median diameter of a resistotype to the CBP, the more isolates in the zone of increased error risk. While few wild-type isolates (mean diameter of 25 mm, distance to CBP of 8 mm) are located in the area at risk for AST errors (Fig. 2A), significantly more BSBL isolates (mean diameter of 20 mm, distance to CBP of 3 mm) (Fig. 2B) and ESBL isolates (mean diameter of 17 mm, distance to CBP of 0 mm) (Fig. 2C) are found in the area at risk. Finally, the lower mean diameter of the AmpC resistotype (10 mm) (Fig. 2D) increased the distance to the CBP to −7 mm, decreasing the number of isolates in the zone of highest risk for misclassification. The influence of distances of population mean diameters from CBPs was also reflected in our analysis: the distances from the median diameter of the wild-type E. coli population to the next EUCAST CBP ranged from 7 mm for piperacillin-tazobactam and ceftazidime to 12 mm for cefotaxime. The corresponding rates of expected interpretation errors were low (0.3% minor errors for ceftazidime, <0.1% errors for the other drugs) (Table 1). In contrast, the distances from the median diameter of the E. coli ESBL resistotype to the next EUCAST CBP were significantly different for different beta-lactam drugs, ranging from 0 mm for amoxicillin-clavulanic acid (i.e., the median diameter for ESBL-positive E. coli was equal to the EUCAST CBP) to −11 mm from the resistant EUCAST CBP for cefotaxime and ceftriaxone. Thus, the resulting interpretation error rates for ESBL-producing E. coli were significantly higher, ranging from 2.4% and 1.3% minor errors for cefotaxime and ceftriaxone to 5.4% major and very major errors for amoxicillin-clavulanic acid (Table 1).
Error probabilities, diameter distributions, and distance to clinical breakpoints for E. coli and amoxicillin-clavulanic acid. Zone diameter distributions of amoxicillin-clavulanic acid (gray bars) for the E. coli wild type (A), BSBL resistotype (B), ESBL resistotype (C), and AmpC resistotype (D) are shown. Median diameter values are indicated by vertical red lines. The error probability around the clinical EUCAST breakpoint (≥17-mm diameter, susceptible; <17-mm diameter, resistant [vertical black lines]) corresponding to methodological imprecision (Table 2) is indicated by the dotted curve. The overlapping area (black shaded) of error probability and moving diameter average (solid black line) are proportional to the cumulated relative error risk, which is dependent on the distance from median CBP diameters (double-headed arrows). (A) The wild-type E. coli results show only a small overlap of curves (low error risk). (B) The overlap of curves increases for the BSBL resistotype. (B and C) The overlap of curves is maximal for the ESBL resistotype (C) and decreases for the AmpC resistotype (D) in agreement with the relative distances from the mean population diameters to the clinical breakpoint. Note that the AmpC population data are not unimodal but show two peaks at 6 mm and 10 mm. The corresponding subpopulations are most probably related to plasmid-encoded AmpC strains showing lower mean diameters and to AmpC promoter/attenuator mutants that usually display higher mean diameters (24).
K. pneumoniae and E. cloacae, which naturally produce chromosomal beta-lactamases, showed lower median wild-type diameters than E. coli, resulting in higher expected interpretation errors rates, as the same CBP for susceptibility applies to all Enterobacteriaceae species (Table 1). Amoxicillin-clavulanic acid, piperacillin-tazobactam, and ceftazidime median zone diameters were 20 mm, 22 mm, and 26 mm for wild-type K. pneumoniae (harboring a chromosomal SHV-type beta-lactamase) but were 25 mm, 27 mm, and 28 mm for wild-type E. coli, respectively. Consequently, the corresponding distances of median K. pneumoniae diameters from CBPs were lower than those for E. coli, leading to higher error rates (0.8%, 9.2%, and 1.7% versus <0.1%, <0.1%, and 0.3% for amoxicillin-clavulanic acid, piperacillin-tazobactam, and ceftazidime, respectively) (Table 1). The highest wild-type strain error rates were found for E. cloacae, which harbors a chromosomal AmpC type beta-lactamase. E. cloacae wild-type error rates ranged from 1.6% for cefepime to 8.6% for ceftazidime (Table 1).
Categorization error rates were significantly higher for some resistotypes than for the wild type of the same species (Table 3). The relative risks of interpretation errors for the E. coli ESBL resistotype compared to the E. coli wild-type strain significantly increased for ceftriaxone (13-fold, P = 0.002), piperacillin-tazobactam (>107-fold, P < 0.001), and amoxicillin-clavulanic acid (>54-fold, P < 0.001). Similarly, the interpretation error risk was significantly increased for E. coli expressing an AmpC beta-lactamase (e.g., >154-fold increase for piperacillin-tazobactam, P < 0.001) (Table 3). In addition, the K. pneumoniae and E. cloacae ESBL and AmpC resistotypes showed a significant increase in expected interpretation errors compared to their wild-type populations (Table 3).
Relative increases and decreases in error rates for resistotypes compared to wild-type populationsa
For K. pneumoniae, the wild-type population accounted for the majority of errors regarding all beta-lactams except for cefotaxime (Table 1). A comparable situation was found for E. cloacae (Table 1). Total errors for Enterobacteriaceae species (i.e., the wild-type strain plus all resistotypes) were, to a considerable extent, dependent on the number of interpretation errors seen with non-wild-type populations (Table 1). For instance, in E. coli the ESBL resistotype was the primary source of interpretation errors for all cephalosporins and ertapenem (47% to 82% of all errors, respectively; Table 1). Errors for amoxicillin-clavulanic acid and piperacillin-tazobactam in E. coli were almost exclusively due to the BSBL and ESBL resistotypes (79% and 20% of all errors, respectively; Table 1).
Interestingly, the interpretation error rates for ESBL and AmpC resistotypes were also significantly increased for ertapenem (Table 1), despite the fact that ertapenem is considered a therapy of choice for ESBL and AmpC producers (32). For E. coli, 79% of all expected ertapenem interpretation errors were caused by the ESBL resistotype (Table 1). Expected error rates for ertapenem ranged from <0.1% for wild-type E. coli to >7% for the ESBL and AmpC resistotypes; for wild-type K. pneumoniae (natural BSBL producer) and E. cloacae (natural AmpC producer), the expected ertapenem error rates were 2.5% and 6%, respectively (Table 1).
DISCUSSION
The suitability of amoxicillin-clavulanic acid, piperacillin-tazobactam, or newer cephalosporins for the treatment of ESBL-producing Enterobacteriaceae infections is still controversial. Clinical data for treatment outcome reveal an ambiguous picture (19, 33–35). Until 2009, CLSI and EUCAST recommended reporting in vitro susceptible and intermediate AST results for penicillins, cephalosporins, and monobactams against ESBL producers as resistant results (CLSI) or to modify interpretations of susceptible to intermediate and of intermediate to resistant (EUCAST) (36, 37). Such editing (i.e., “interpretative reading”) for beta-lactams and ESBL-producing isolates has been abandoned (3, 38). However, EUCAST expert rules contain a warning on uncertain therapeutic outcomes for ESBL-producing isolates and inhibitor combinations (rule 9.1) or newer cephalosporins, and monotherapy treatment of Enterobacter spp. with newer cephalosporins is discouraged (rule 9.2) (13). The new CLSI and EUCAST “report as found” strategy for ESBL producers and beta-lactams emphasizes the role of inhibition zone diameter or MIC value relative to a CBP as the single parameter for clinical categorization, i.e., for the prediction of the clinical outcome.
In a previous work, we demonstrated that CBP setting and measurement precision significantly influence the rate of reporting errors for individual species/drug combinations (7). Accepted error rates in AST classification systems are <5% for minor and <1% for major and very major errors (2). Such error rates for species/drug combinations are calculated from mixed wild-type and non-wild-type populations, so they represent mean error rates for all isolates of an individual species. Species are, however, not homogeneous but rather represent subpopulations of wild-type isolates and one or more resistotypes. Thus, average error rates calculated from species distributions represent a statistical mean but do not apply to an individual isolate.
Following the revised CLSI/EUCAST reporting strategy, we analyzed reporting error probabilities for various resistotypes. Subpopulations carrying ESBL, BSBL, or AmpC beta-lactamases displayed significantly different probabilities of erroneous clinical categorization (Table 1). Most importantly, the average categorization reliability was significantly lower for non-wild-type populations, in particular, for ESBL and AmpC-producing isolates (Table 3), which accounted for the majority of expected errors (Table 1). This may explain, in part, ambiguous reports on the therapeutic applicability of certain resistotype/drug combinations, e.g., ESBL-positive isolates and amoxicillin-clavulanic acid and piperacillin-tazobactam (21, 39).
In its 2014 update of AST guidelines, EUCAST altered the amoxicillin-clavulanic acid susceptible and resistant CBPs to ≥16 and <16 mm for uncomplicated urinary tract infections and to ≥19 mm and <19 mm for all other cases (40). This CBP modification hardly changes the picture of expected errors; e.g., applying the new CBPs, the expected error rates for ESBL and E. coli would be at 5.3% for urinary tract isolates and at 6% for other isolates versus 5.4% for all isolates, applying the former uniform CBPs of ≥17 mm and <17 mm. This example indicates that the problem of resistotype-dependent reporting errors cannot be solved by CBP changes alone. This study analyzed data from disk diffusion testing. However, by logical consequence, the problem of interpretation errors and overlapping or adjacent wild-type and non-wild-type populations may be extrapolated also to MIC-based methods, as disk diffusion diameters and MICs correlate.
This study found only minor errors for drug/species combinations with an intermediate zone (see Table 1). However, minor errors may nonetheless influence clinical decision making as they will either prompt the selection of another drug, limiting therapeutic options (susceptible-to-intermediate error), or prevent the use of high-dose or combination therapy (intermediate-to-susceptible error). The majority of minor errors expected were “susceptible to intermediate,” especially for the E. coli wild type, the BSBL-producing E. coli, and the K. pneumoniae wild type (Table 1). The ESBL genotypes showed a balanced proportion of susceptible/intermediate and intermediate/resistant minor errors, whereas AmpC-producing isolates tended to produce more “intermediate-to-resistant” and “resistant-to-intermediate” errors. These findings are consistent with the distance from the median diameters of the resistotypes to the susceptible/intermediate and intermediate/resistant CBPs.
The CBPs of the drugs analyzed in this study are identical for all Enterobacteriaceae (22). However, many authors point out the importance of setting species-specific CBPs to avoid erroneous AST reports due to different diameter/MIC distributions of species (2, 7, 41). Our results reinforce these statements, as the same CBPs applied to wild-type E. coli, K. pneumoniae, and E. cloacae lead to significantly different error rates for most beta-lactams (Tables 1 and 3). Error rates for wild-type E. coli were close to 0%, whereas those for wild-type K. pneumoniae and E. cloacae reached 9.2% (K. pneumoniae with piperacillin-tazobactam). Interpretation error rates were dependent on resistotypes rather than being species related, i.e., error rates for species with natural resistance mechanisms were analogous to those for species with acquired similar resistance mechanisms; e.g., the K. pneumoniae wild type harboring an SHV-1 beta-lactamase (BSBL type) and the E. coli BSBL resistotype show similar error rates (Tables 1 and 3). Furthermore, the AST error characteristics of K. pneumoniae and E. coli AmpC resistotypes equaled those of the E. cloacae wild type, which produces a chromosomal AmpC. From the perspective of antibiogram reliability, knowledge of the resistotype may, therefore, be at least as relevant as correct species identification.
Our results underline that total error rates in antibiograms for species/drug combinations depend on the prevalences of individual resistotypes, e.g., the number of ESBL-producing organisms in relation to the wild type, as illustrated by Fig. 1D. As ESBL producers present a higher likelihood for reporting errors, error rates increase in parallel to ESBL prevalence. In our study population, E. coli ESBL had a prevalence of 9% (Table 1), which is in line with the rate in many countries in Central Europe, and the resulting total error rate for all E. coli isolates with respect to amoxicillin-clavulanic acid was 2.5% (Table 1) (42). If the ESBL prevalence were as high as 60%, as reported for many regions in Asia, the total error rate would increase to 4.8% (43).
What are the practical implications of our findings for AST reporting and therapeutic decision making? To ensure an adequate therapy by improving antibiograms, three possible options exist. (i) For the short term, a simple and practical approach would be to pursue detection and reporting of ESBL and AmpC production (currently recommended by CLSI and EUCAST for epidemiological purposes only). Clinicians should be aware of the significantly increased risk of erroneous AST reports for isolates harboring beta-lactamases, in particular, ESBL and AmpC. This also applies to drugs that are recommended for the treatment of ESBL and AmpC producers such as ertapenem. (ii) AST reports may indicate MIC and diameter measurements in combination with reference ranges and laboratory measurement precision data, enabling clinicians to assess the reliability of antibiotic susceptibility classification. (iii) AST reports could benefit from inclusion of an indicator of interpretation reliability, e.g., percentages of probability for correct clinical classification based on CLSI and/or EUCAST CBPs. These probabilities would reflect the resistotype/drug combination-dependent AST forecast reliability for clinical outcome, facilitating the selection of the most adequate drug for treatment.
ACKNOWLEDGMENTS
We are grateful to the laboratory technicians of the Institute of Medical Microbiology, University of Zurich, for their dedicated help, to Forouhar Mouttet for performing proper statistical analyses, and to Reinhard Zbinden and Guido Bloemberg for valuable discussions.
We declare that we have no conflicts of interest.
This work was supported by the University of Zurich.
FOOTNOTES
- Received 11 June 2014.
- Returned for modification 9 July 2014.
- Accepted 31 July 2014.
- Accepted manuscript posted online 6 August 2014.
- Copyright © 2014, American Society for Microbiology. All Rights Reserved.
