Haemophilus influenzae with Non-Beta-Lactamase-Mediated Beta-Lactam Resistance: Easy To Find but Hard To Categorize

Abstract

Haemophilus influenzae is a major pathogen, and beta-lactams are first-line drugs. Resistance due to altered penicillin-binding protein 3 (rPBP3) is frequent, and susceptibility testing of such strains is challenging. A collection of 154 beta-lactamase-negative isolates with a large proportion of rPBP3 (67.5%) was used to evaluate and compare Etest (Haemophilus test medium [HTM]) and disk diffusion (EUCAST method) for categorization of susceptibility to aminopenicillins and cefuroxime, using MICs generated with broth (HTM) microdilution and clinical breakpoints from CLSI and EUCAST as the gold standards. In addition, the proficiency of nine disks in screening for the rPBP3 genotype (N526K positive) was evaluated. By Etest, both essential and categorical agreement were generally poor (<70%), with high very major errors (VME) (CLSI, 13.0%; EUCAST, 34.3%) and falsely susceptible rates (FSR) (CLSI, 87.0%; EUCAST, 88.3%) for ampicillin. Ampicillin (2 μg) with adjusted (+2 mm) zone breakpoints was superior to Etest for categorization of susceptibility to ampicillin (agreement, 74.0%; VME, 11.0%; FSR, 28.3%). Conversely, Etest was superior to 30 μg cefuroxime for categorization of susceptibility to cefuroxime (agreement, 57.1% versus 60.4%; VME, 2.6% versus 9.7%; FSR, 7.1% versus 26.8%). Benzylpenicillin (1 unit) (EUCAST screening disk) and cefuroxime (5 μg) identified rPBP3 isolates with highest accuracies (95.5% and 92.2%, respectively). In conclusion, disk screening reliably detects rPBP3 H. influenzae, but false ampicillin susceptibility is frequent with routine methods. We suggest adding a comment recommending high-dose aminopenicillin therapy or the use of other agents for severe infections with screening-positive isolates that are susceptible to aminopenicillins by gradient or disk diffusion.

INTRODUCTION

Nontypeable Haemophilus influenzae (NTHi) frequently causes acute otitis media, conjunctivitis, sinusitis, and respiratory tract infections, including pneumonia and exacerbations in chronic obstructive pulmonary disease and cystic fibrosis, and may also cause invasive disease (1). Beta-lactams are first-line drugs when systemic therapy is indicated, but resistance due to transferable beta-lactamase genes (bla_TEM or bla_ROB) and/or substitutions in penicillin-binding protein 3 (PBP3), encoded by the ftsI gene, is common (1, 2). Strains with low-level PBP3-mediated resistance (low-rPBP3) possess the R517H (group I) or the N526K (group II) substitution, whereas strains with high-level resistance (high-rPBP3) are characterized by the additional substitution S385T (2, 3). The distinction is clinically important because high-rPBP3 strains express higher resistance to extended-spectrum cephalosporins (1, 2, 4–6). Group II low-rPBP3 is the predominating genotype in Australia (7), Europe (8, 9), and North America (6, 10), whereas high-rPBP3 strains predominate in Japan and Korea (6, 11).

Susceptibility testing of H. influenzae has been characterized as “a tricky business” (12). Categorization of susceptibility to aminopenicillins (with or without bla inhibitors) is particularly challenging (2, 13). First, the rPBP3 population overlaps the wild-type population in terms of MICs: bla-negative rPBP3 strains have ampicillin MICs of 0.5 to 16 mg/liter (1, 2, 13), whereas the epidemiological cutoff (ECOFF) value is ≤1 mg/liter (www.eucast.org/mic_distributions_ecoffs). Second, the clinical relevance of current breakpoints is debated (1, 2, 13). Ampicillin breakpoints from the Clinical and Laboratory Standards Institute (CLSI) were originally set to distinguish between bla-positive and bla-negative strains (13). Non-species-related (pharmacokinetic-pharmacodynamic [PK/PD]) breakpoints calculated by the European Committee on Antimicrobial Susceptibility Testing (EUCAST) suggest that isolates with ampicillin MICs of ≤2 mg/liter are susceptible to the standard dosage and that isolates with MICs of ≤8 mg/liter are susceptible to high-dose therapy (14). However, CLSI (15) and EUCAST (14) both use ECOFF to define susceptibility and divide the rPBP3 population by categorizing strains with ampicillin MICs of >1 mg/liter (EUCAST) or >2 mg/liter (CLSI) as resistant irrespective of dosage. In addition, susceptibility testing is associated with uncertainty due to technical and biological variation. For broth microdilution (BMD), a precision of ±1 dilution is accepted (16). Thus, categorization of H. influenzae by MIC-based susceptibility to ampicillin will in some cases lead to categorization errors, even with reference methods.

Gradient tests and disk diffusion are commonly used for routine susceptibility testing (16). Several studies have indicated that gradient tests may be unreliable for H. influenzae and aminopenicillins (17–21). Irrespective of method, different media for susceptibility testing of H. influenzae to ampicillin may affect the results (17). CLSI recommends Haemophilus test medium (HTM), composed of Mueller-Hinton agar (MHA) supplemented with 5 g/liter yeast extract, 15 μg/ml beta-NAD, and 15 μg/ml hematin (22), whereas EUCAST recommends Mueller-Hinton fastidious (MH-F) agar, consisting of MHA supplemented with 5% defibrinated horse blood and 20 mg/liter beta-NAD (23). Different ampicillin disk potencies are recommended for H. influenzae (CLSI, 10 μg; EUCAST, 2 μg), but most studies indicate that the low-potency disk correlates best to broth dilution (19, 24, 25). EUCAST has defined screening breakpoints for the 1-unit benzylpenicillin disk (14). The test limits agent-directed testing to strains with beta-lactam resistance mechanisms but may not be used for detection of rPBP3 in bla-positive H. influenzae.

The aims of this study were (i) to compare disk and gradient diffusion for categorization of susceptibility to aminopenicillins and cefuroxime, with BMD MICs interpreted according to CLSI (15) and EUCAST (14) breakpoints as the gold standards, and (ii) to evaluate selected beta-lactam disks as screening for the rPBP3 genotype in H. influenzae.

(These data were presented in part [PG1, PV10, and CEC30] at the 21st European Congress of Clinical Microbiology and Infectious Diseases [ECCMID], Milan, Italy, 2011 [26]).

MATERIALS AND METHODS

Bacterial isolates.

The test population consisted of 154 bla-negative H. influenzae isolates from a previous investigation (9) and comprised 50 (32%) R517H-, N526K-, and S385T-negative (sPBP3) isolates and 104 (68%) group II low-rPBP3 isolates (N526K positive and S385T negative). All sPBP3 isolates were susceptible to ampicillin (MIC range, 0.06 to 1 mg/liter), amoxicillin (MIC range, 0.125 to 2 mg/liter), and cefuroxime (MIC range, 0.125 to 1 mg/liter) by previous HTM (Oxoid, Basingstoke, United Kingdom) broth microdilution (BMD) MICs interpreted with breakpoints from CLSI (15) and EUCAST (14). The 104 rPBP3 isolates had ampicillin MIC ranges of 0.5 to 8 mg/liter, amoxicillin MIC ranges of 0.125 to ≥32 mg/liter, and cefuroxime MIC ranges of 0.25 to ≥32 mg/liter; the proportions categorized as susceptible (S), intermediate (I), and resistant (R) were as follows: ampicillin, 42%/36%/22%; cefuroxime, 49%/35%/16% (CLSI); and ampicillin, 42%/58%; amoxicillin, 42%/58%; cefuroxime, 36%/11%/54% (EUCAST).

Gradient MIC (Etest).

Ampicillin, amoxicillin, and cefuroxime MICs were determined by Etest (bioMérieux, Marcy-l'Étoile, France) according to the manufacturer's recommendations (in-house HTM; MHA from Oxoid). Essential agreement (±1 dilution), categorical agreement (S, I, and R), error rates (very major error [VME], major error [ME], and minor error [mE]) and falsely susceptible rates (FSR; proportions of R isolates categorized as S) were calculated with previously determined BMD MICs (9) interpreted according to CLSI (15) and EUCAST (14) clinical MIC breakpoints as the gold standards. For aminopenicillins, “essential correlation” (defined as an amoxicillin MIC within ±1 dilution of the ampicillin MIC + 1 dilution) and “categorical correlation” (identical categorization of ampicillin and amoxicillin according to MIC breakpoints) were also calculated. Because CLSI has not defined breakpoints for amoxicillin (15), categorical correlation was calculated only with EUCAST breakpoints (14).

Disk diffusion (EUCAST).

Disk diffusion (in-house MH-F; MHA from Oxoid) was performed with EUCAST methods (23) and nine disks (Oxoid): 1 unit benzylpenicillin (PG1), 5 units benzylpenicillin (PG5), 10 μg phenoxymethylpenicillin (PV10), 2 μg ampicillin (AMP2), 2/1 μg amoxicillin-clavulanic acid (AMC3), 20/10 μg amoxicillin-clavulanic acid (AMC30), 30 μg cefaclor (CEC30), 5 μg cefuroxime (CXM5), and 30 μg cefuroxime (CXM30).

All disks were evaluated for their ability to identify isolates with the rPBP3 genotype (N526K positive). Sensitivity and specificity at different zone breakpoints were used to construct receiver operating characteristic (ROC) curves. The disks were compared by sensitivity, specificity, and positive and negative predictive values with optimized screening breakpoints, i.e., the zone breakpoints resulting in the highest proportions of correct results.

Categorization was performed according to current EUCAST recommendations (14), using standard disks (AMP2 and CXM30) and zone breakpoints, and susceptibility to amoxicillin inferred from ampicillin. Categorization of susceptibility to ampicillin, amoxicillin, and cefuroxime was also performed by using alternative disks (AMC3 and CXM5) and zone breakpoints and by using the PG1 disk and EUCAST screening breakpoints (14). Categorical agreement, error rates, and FSR for disk diffusion were calculated with previously determined BMD MICs (9) interpreted according to EUCAST clinical MIC breakpoints (14) as the gold standard (CLSI, not applicable). The results were compared with those obtained by Etest, and significance levels were calculated using the chi-square test and software at www.medcalc.net (MedCalc Software, Ostend, Belgium).

Quality control.

All results for the reference strains H. influenzae ATCC 49247 and H. influenzae ATCC 49766 (MICs) and H. influenzae NCTC 8468 (zones) were within accepted ranges (15, 27).

RESULTS

MIC determination and susceptibility categorization (Etest).

By Etest, 94.2% (145/154) of the isolates had ampicillin MICs of ≤1 mg/liter and were categorized as susceptible with both guidelines; 5.2% (8/154) had ampicillin MICs of 2 mg/liter and were categorized as intermediate (CLSI) or resistant (EUCAST); one isolate had a MIC of >2 mg/liter and was categorized as resistant. The corresponding proportions with BMD were 61.0%, 24.0%, and 14.9%. Essential correlation between ampicillin and amoxicillin MIC was 93.5% (144/154) by Etest and 86.4% (133/154) by BMD. With EUCAST breakpoints for ampicillin and amoxicillin (CLSI, not applicable), categorical correlation between the two agents was 88.3% (136/154) by Etest, compared to 89.6% (138/154) by BMD.

Table 1 shows correlations between Etest and BMD. Etest generally overestimated MICs at lower ranges and underestimated MIC at higher ranges, and agreement rates were low for all three agents. On average, Etest underestimated ampicillin MIC by one dilution for isolates with BMD MICs of 2 mg/liter, by two dilutions for isolates with BMD MICs of 4 mg/liter, and by three dilutions for isolates with BMD MICs of 8 mg/liter, leading to high VME and FSR with both MIC guidelines. In contrast to ampicillin and amoxicillin, low categorical agreement rates with cefuroxime Etest (both guidelines) were caused mainly by high proportions of minor errors; however, FSR were high with CLSI breakpoints.

TABLE 1.

Categorization of susceptibility of beta-lactamase-negative isolates (n = 154) to ampicillin, amoxicillin, and cefuroxime by Etest (HTM)^a

^aHTM broth microdilution (BMD) MICs were interpreted according to breakpoints from CLSI (15) and EUCAST (14) as gold standards.

^bDifference between Etest and BMD MIC, expressed as the number of 2-fold dilutions by which the Etest MIC is higher (positive values) or lower (negative values) than BMD MIC. CA, categorical agreement between Etest MIC and BMD MIC (breakpoints indicated by vertical lines). mE, minor error (intermediate by Etest and susceptible/resistant by BMD MIC or susceptible/resistant by Etest and intermediate by BMD MIC); ME, major error (resistant by Etest and susceptible by BMD MIC). VME, very major error (susceptible by Etest and resistant by BMD MIC).

^cEA, essential agreement (proportion of Etest MICs within ±1 dilution of BMD MIC) calculated by summarizing the values shown in bold in the adjacent column.

^dFSR, falsely susceptible rate: proportion of isolates resistant by BMD MIC categorized as susceptible by Etest. FSRs are calculated by dividing the number of VMEs (“n” column) by the number of isolates with BMD MICs above the resistance breakpoint (indicated by vertical lines).

^eIncludes Etest MICs deviating from BMD MIC by more than three dilutions.

Susceptibility categorization (disk diffusion).

Zone-MIC correlations showed considerable overlapping of susceptible and resistant isolates with AMP2, AMC3, CXM30, and CXM5 (see Fig. S1 in the supplemental material). Table 2 summarizes the performances of these four disks and PG1 for categorization of susceptibility to ampicillin, amoxicillin, and cefuroxime.

TABLE 2.

Categorization of susceptibility of beta-lactamase-negative isolates (n = 154) to ampicillin, amoxicillin, and cefuroxime by disk diffusion (EUCAST method)^a

Agent	Diskb	Breakpointsc		FSRd			VMEe			MEf		mEg		CAh
		MIC (mg/liter)	Zone (mm)	Fraction	%	Comparisoni	n	%	Comparisoni	n	%	n	%	n	%	Comparisoni
Ampicillin	AMP2	1/1j	16/16j	46/60	76.7	(↓)	46	29.9	(↓)	6	3.9	NA	NA	103	66.9	(↑)
	AMP2	1/1j	18/18	17/60	28.3	↓↓	17	11.0	↓↓	24	15.6	NA	NA	114	74.0	(↑)
	AMC3	1/1j	15/15l	24/60	40.0	↓↓	24	15.6	↓	15	9.7	NA	NA	116	75.3	↑
	AMC3	1/1j	16/16	17/60	28.3	↓↓	17	11.0	↓↓	20	13.0	NA	NA	118	76.6	↑
	PG1	1/1j	12/12k	2/60	3.3	↓↓	2	1.3	↓↓	45	29.2	NA	NA	107	69.8	(↑)
Amoxicillin	AMP2	2/2j	16/16j	47/60	78.3	(↑)	47	30.5	(↑)	7	4.5	NA	NA	101	65.6	(↓)
	AMP2	2/2j	18/18	16/60	26.7	↓↓	16	10.4	↓	22	14.3	NA	NA	117	76.0	(↑)
	AMC3	2/2j	15/15l	25/60	41.7	↓	25	16.2	↓	16	10.4	NA	NA	114	74.0	(↑)
	AMC3	2/2j	16/16	17/60	28.3	↓↓	17	11.0	↓	20	13.0	NA	NA	118	76.6	(↑)
	PG1	2/2j	12/12k	3/60	5.0	↓↓	3	1.9	↓↓	46	29.9	NA	NA	105	68.2	(↓)
Cefuroxime	CXM30	1/2j	26/25j	15/56	26.8	↑	15	9.7	↑	27	17.5	20	13.0	93	60.4	(↑)
	CXM5	1/2j	20/19	7/56	12.5	(↓)	7	4.5	(↑)	32	20.8	17	11.0	99	64.3	(↑)
	PG1	1/2j	12/12k	4/56	7.1	—	4	2.6	—	40	26.0	11	7.1	99	64.3	(↑)

^aHTM broth microdilution (BMD) MICs interpreted according to EUCAST breakpoints (14) were used as the gold standard.

^bAMP2, 2 μg ampicillin; AMC3, 2/1 μg amoxicillin-clavulanic acid; CXM5, 5 μg cefuroxime; CXM30, 30 μg cefuroxime; PG1, 1 unit benzylpenicillin.

^cBreakpoints are presented as S≤/R> (MIC) and S≥/R< (zone). The reference for all breakpoints is this study unless otherwise indicated.

^dFalsely susceptible rate, i.e., proportion of isolates resistant by broth microdilution (BMD) MIC categorized as susceptible by disk diffusion.

^eVery major error, i.e., susceptible by disk diffusion and resistant by BMD MIC.

^fMajor error, i.e., resistant by disk diffusion and susceptible by BMD MIC.

^gMinor error, i.e., susceptible/resistant by disk diffusion and intermediate by BMD MIC or intermediate by disk diffusion and susceptible/resistant by BMD MIC. NA, not applicable.

^hCategorical agreement between disk diffusion and BMD MIC.

ⁱComparison with results obtained by Etest (Table 1) and calculation of significance levels using chi-square test. ↑, higher; ↓, lower; —, identical; ↑↑ and ↓↓, P ≤ 0.0001; ↑ and ↓, P ≤ 0.05; (↑) and (↓), not significant (P > 0.05).

^jEUCAST clinical breakpoint (14).

^kEUCAST screening breakpoint for detection of isolates with beta-lactam resistance mechanisms (14).

^lEUCAST clinical breakpoint for categorization of susceptibility to amoxicillin-clavulanic acid (14).

With current EUCAST zone breakpoints, categorical agreement was slightly higher with AMC3 than AMP2 (P = 0.1081), with significantly lower VME (P = 0.0047) and FSR (P = 0.0001). AMC3 was also superior to ampicillin Etest for categorization of susceptibility to ampicillin. A minor zone breakpoint adjustment (+2 mm) for AMP2 slightly increased agreement with the BMD MIC for categorization of susceptibility to ampicillin (P = 0.1593) and amoxicillin (P = 0.0464). With adjusted zone breakpoints for AMP2 and AMC3 (+1 mm), there were no significant differences between the disks, and both were superior to Etest for categorization of susceptibility to aminopenicillins. In addition, categorical correlation between AMP2 (ampicillin) and AMC3 (amoxicillin) increased from 79.9% (123/154) with current EUCAST breakpoints to 90.9% (140/154) with adjusted zone breakpoints (P = 0.0121) (see Fig. S2 in the supplemental material).

For cefuroxime, categorical agreement with both disk potencies (and the PG1 screening disk) was poor and not significantly different from that obtained with Etest (EUCAST breakpoints). FSR and VME were significantly higher with CXM30 than Etest (and PG1), whereas there were no significant differences between Etest and CXM5 or between CXM5 and CXM30.

Screening (disk diffusion).

The relative performance of the nine evaluated disks is shown in Fig. 1, and the correlation between zones and resistance genotype for selected disks is shown in Fig. S3 in the supplemental material. The optimized screening breakpoint for PG1 (S ≥ 12 mm) was identical to the breakpoint recommended by EUCAST. PG1 identified rPBP3 strains with the highest sensitivity and accuracy of all disks tested. CXM5 was superior to the other four disks with agents that are stable in the presence of beta-lactamase (Table 3).

FIG 1.

Receiver operating characteristic (ROC) diagram showing the performance of nine beta-lactam disks for detection of penicillin-binding protein 3-mediated beta-lactam resistance (rPBP3) in beta-lactamase-negative isolates (n = 154). Optimized screening breakpoints (Table 3; also, see Fig. S3 in the supplemental material) are indicated for each disk. Solid lines, agents that are stable in the presence of beta-lactamase; dashed lines, agents that are susceptible to beta-lactamase. PG1, benzylpenicillin 1 unit; PG5, benzylpenicillin (5 units); PV10, phenoxymethylpenicillin (10 μg); AMP2, ampicillin (2 μg); AMC3, amoxicillin-clavulanic acid (2/1 μg); AMC30, amoxicillin-clavulanic acid (20/10 μg); CEC30, cefaclor (30 μg); CXM5; cefuroxime (5 μg); CXM30, cefuroxime (30 μg).

TABLE 3.

Screening for penicillin-binding protein 3-mediated beta-lactam resistance (rPBP3) by disk diffusion (EUCAST method) in beta-lactamase (bla)-negative isolates (n = 154)

Diska	Bla stable	Breakpoint (mm)b	Sensitivity (%)	Specificity (%)	PPVc (%)	NPV (%)d	Accuracye
							n	%
PG1	No	12f	96.2	94.0	97.1	92.2	147	95.5
PG5	No	20	90.4	94.0	96.9	82.5	141	91.6
PV10	No	20	95.2	82.0	91.7	89.1	140	90.9
AMP2	No	20	89.4	98.0	98.9	81.7	142	92.2
AMC3	Yes	18	81.7	86.0	92.4	69.4	128	83.1
AMC30	Yes	26	69.2	88.0	92.3	57.9	116	75.3
CEC30	Yes	23	77.9	92.0	95.3	66.7	127	82.5
CXM5	Yes	21	94.2	88.0	94.2	88.0	142	92.2
CXM30	Yes	27	85.6	88.0	93.7	74.6	133	86.4

^aPG1, 1 unit benzylpenicillin; PG5, 5 units benzylpenicillin; PV10, 10 μg phenoxymethylpenicillin; AMP2, 2 μg ampicillin; AMC3, 2/1 μg amoxicillin-clavulanic acid; AMC30, 20/10 μg amoxicillin-clavulanic acid; CEC30, 30 μg cefaclor; CXM5; 5 μg cefuroxime; CXM30, 30 μg cefuroxime.

^bOptimized screening breakpoints (S≥) with rPBP3 isolates (N526K positive) defined as screening targets (Fig. 1).

^cPositive predictive value.

^dNegative predictive value.

^eCorrect assignment to resistance genotype (N526K positive and screening positive, or N526K negative and screening negative).

^fIdentical to the screening breakpoint recommended by EUCAST (14).

Hazy growth within inhibition zones was frequently observed for rPBP3 isolates, in particular with PG1 and CEC30; sensitivity was substantially reduced when hazy growth was ignored.

DISCUSSION

In recent years, altered PBP3 has surpassed beta-lactamase as the most frequent beta-lactam resistance mechanism in H. influenzae in several geographical regions (6, 8–11), and reliable detection and susceptibility categorization of this organism have become increasingly important.

Commonly used acceptance criteria are >90% essential agreement for MIC determination and <1.5% VME and <3% ME for susceptibility categorization (16). Notably, VME and ME (and categorical agreement) are calculated with the complete test population as the denominator and thus strongly influenced by the proportion of resistant isolates (prevalence). These parameters may be used for comparison of methods for susceptibility testing with identical populations, but the usefulness of acceptance criteria based on defined categorical error rates without taking into account the representativeness of the test population is highly debatable. In contrast to VME, the falsely susceptible rate (FSR) is independent of prevalence and suitable for comparison of results obtained with different methods and test populations.

It should also be noted that categorical errors are dependent on MIC breakpoints and that rates achieved with different guidelines may not be compared.

In the present study, using a test population with 67.5% rPBP3 isolates, neither essential nor categorical agreement rates with Etest exceeded 70% for any of the tested agents. VME were particularly frequent for ampicillin, and most resistant isolates were wrongly categorized as susceptible (both guidelines). Poor performance of gradient tests for susceptibility testing of H. influenzae to aminopenicillins has been reported previously (17–21, 28). Mushtaq et al. compared Etest and M.I.C. Evaluator (M.I.C.E., Oxoid) with agar dilution (IsoSensitest agar) for several agents and species (20). Overall essential agreement was poorest for the 56 Haemophilus isolates (Etest, 74.4%; M.I.C.E., 76.9%); in this group of organisms, essential agreement was particularly low for ampicillin and amoxicillin (Etest, 60.5% and 42.4%; M.I.C.E, 73.8% and 59.4%). Billal et al. reported even lower essential agreement for aminopenicillins (<50%) in a study comparing Etest to HTM BMD for categorization of 87 bla-negative rPBP3 strains; a large number of ampicillin Etest MICs were ≥6 dilutions higher than the BMD MIC (18). Overestimation of ampicillin resistance in Haemophilus by gradient tests was also observed by Rennie et al.; using HTM BMD as the gold standard, the authors found 8.2% (4/49) ME with M.I.C.E. and 6.1% (3/49) with Etest (21). In contrast, and consistent with the observations in the present study, Barry et al. tested 143 bla-negative H. influenzae isolates and found a higher ampicillin susceptibility rate with Etest (68%) than HTM BMD (59%) (17). Similarly, Garcia-Cobos et al. obtained a higher ampicillin susceptibility rate with Etest compared to HTM BMD in 34 bla-negative rPBP3 H. influenzae (88.2% versus 76.5%); Etest MIC was 1 to 2 dilutions lower than BMD MIC for 41% (14/34) of the isolates (19). Finally, Tristram tested 15 bla-negative rPBP3 H. influenzae isolates with Etest and M.I.C.E., and despite 90% essential agreement, 47% (14/30) of ampicillin gradient MICs were 1 to 2 dilutions lower than the HTM BMD MIC (28).

To our knowledge, this is the first study comparing gradient MIC and EUCAST disk diffusion with reference methods for susceptibility testing of H. influenzae. Using identical test populations, disk diffusion with current EUCAST breakpoints was noninferior (AMP2) or superior (AMC3) to Etest for categorization of susceptibility to aminopenicillins; with minor adjustments in zone breakpoints, both disks were superior to Etest. In a previous evaluation, Søndergaard et al. tested 135 bla-negative H. influenzae isolates (33% rPBP3) and reported that six isolates were ampicillin resistant by both EUCAST disk diffusion (MH-F medium) and HTM BMD (MIC > 1 mg/liter), whereas 10% (i.e., 13 or 14 isolates) were resistant by BMD only (29). Thus, more than half (7/13 or 8/14) of the ampicillin-resistant isolates were falsely categorized as susceptible with the AMP2 disk, compared to 76.7% (current breakpoints) and 28.3% (adjusted breakpoints) in the present study. In contrast, Zerva et al. used AMP2 and HTM (CLSI medium) for testing of 183 bla-negative H. influenzae isolates (including 22 non-ampicillin-susceptible isolates) and found 96% (176/183) categorical agreement with HTM BMD; no isolates with MICs of >1 mg/liter were categorized as susceptible by disk diffusion (25). The results may not be fully representative for the difference between HTM and MH-F. In a more recent multicenter study by Kärpänoja et al., two ampicillin-susceptible and three ampicillin-resistant (MIC > 1 mg/liter) bla-negative control strains were tested with AMP2 and HTM (24). Using the same interpretative criteria as Zerva et al. (S ≥ 16 mm and R < 16 mm) (25), the authors observed an overall FSR of 8% (sensitivity, 92%); however, the individual FSR for one of the ampicillin-resistant strains (reference MIC = 8 mg/liter) was 24%. Higher frequencies of errors would be expected with clinical strains.

The observed 90% correlation between ampicillin and amoxicillin BMD MIC supports current recommendations that ampicillin may be used to infer susceptibility to amoxicillin (14, 15) and that ampicillin-sulbactam may be inferred from amoxicillin-clavulanic acid (14). Nevertheless, categorization of susceptibility to aminopenicillins with and without bla inhibitors should ideally be based on the same agent irrespective of bla status. With minor zone breakpoint adjustments, we found excellent correlation between AMP2 and AMC3, and AMC3 was noninferior to AMP2 for categorization of susceptibility to aminopenicillins, consistent with previous observations with HTM (24). Thus, AMC3 may be used for categorization of bla-negative as well as bla-positive H. influenzae isolates. Although the low antibacterial activity of clavulanic acid against H. influenzae (MIC range, 25 to 125 mg/liter) (30) is unlikely to affect bla-negative isolates by broth dilution with a fixed concentration at 2 mg/liter (14), the question of whether separate zone breakpoints are needed for bla-positive versus bla-negative isolates should be investigated.

The poor categorical agreement and high frequency of falsely susceptible results by categorization of rPBP3 H. influenzae to aminopenicillins by disk diffusion and Etest observed in the present study are worrisome. The main reason for the discrepancy between AMP2 and the reference method was poor separation between rPBP3 isolates with ampicillin BMD MICs of 1 mg/liter (susceptible) and 2 mg/liter (nonsusceptible). This observation is not surprising, as these isolates belong to the same population. Slightly improved performance with minor zone breakpoint adjustments may possibly reflect that susceptibility testing is method dependent. Differences in disks and media from different manufacturers may greatly affect the results, but such variation was not investigated in the present study. Further investigations are needed to decide whether a change in breakpoints is advisable. The data suggest that a change in ampicillin MIC breakpoints to avoid division of the rPBP3 population would further improve categorical agreement with BMD MIC for both disk diffusion and Etest (data not shown). Definition of an intermediate category encompassing bla-negative isolates with ampicillin MICs up to 8 mg/liter is supported by PK/PD calculations (14); however, any changes in MIC breakpoints should be supported by clinical data, but such data are currently lacking (13).

With current breakpoints and methods for phenotypic susceptibility categorization, a useful approach to reduce errors is to screen for isolates with resistance mechanisms. A highly sensitive screening method practically eliminates ME, as wild-type isolates may be reported as being susceptible to beta-lactams without further testing. The PG1 disk recommended for screening by EUCAST (14) correctly categorized 95.5% of the isolates according to resistance genotype in the present study. Søndergaard et al. found similar accuracy (96%) but lower sensitivity (91%) using the same method (29). Diverging results may be due to different interpretation of hazy growth (24); this phenomenon may vary with MHA from different manufacturers (unpublished data). Notably, PG1 with current screening breakpoints was noninferior to Etest and superior to CXM30 for S/I/R categorization of susceptibility to cefuroxime (EUCAST breakpoints). As PG1 is unsuitable for rPBP3 screening in bla-positive strains, we evaluated five disks with agents that are stable in the presence of beta-lactamase. The CXM5 disk categorized isolates according to resistance genotype with the highest accuracy (92.2%) and was superior to the previously evaluated CEC30, CXM30, AMC3, and AMC30 disks (29, 31); however, CXM5 is currently not available from all manufacturers.

In accordance with guidelines from the Nordic Committee on Antimicrobial Susceptibility Testing (NordicAST) (32), we suggest that H. influenzae isolates that are rPBP3 positive by screening be categorized as cefuroxime resistant and always be reported as ampicillin resistant in cases of meningitis. In addition, to minimize the clinical consequences of falsely susceptible results, we suggest adding a comment recommending high-dose aminopenicillin therapy or the use of other agents in severe infections caused by screening-positive isolates categorized as susceptible to aminopenicillins by disk or gradient diffusion.