Comparison of Three Statistical Methods for Establishing Tentative Wild-Type Population and Epidemiological Cutoff Values for Echinocandins, Amphotericin B, Flucytosine, and Six Candida Species as Determined by the Colorimetric Sensititre YeastOne Method

Abstract

The Sensititre YeastOne (SYO) method is a widely used method to determine the susceptibility of Candida spp. to antifungal agents. CLSI clinical breakpoints (CBP) have been reported for antifungals, but not using this method. In the absence of CBP, epidemiological cutoff values (ECVs) are useful to separate wild-type (WT) isolates (those without mechanisms of resistance) from non-WT isolates (those that can harbor some resistance mechanisms), which is the goal of any susceptibility test. The ECVs for five agents, obtained using the MIC distributions determined by the SYO test, were calculated and contrasted with those for three statistical methods and the MIC₅₀ and modal MIC, both plus 2-fold dilutions. The median ECVs (in mg/liter) (% of isolates inhibited by MICs equal to or less than the ECV; number of isolates tested) of the five methods for anidulafungin, micafungin, caspofungin, amphotericin B, and flucytosine, respectively, were as follows: 0.25 (98.5%; 656), 0.06 (95.1%; 659), 0.25 (98.7%; 747), 2 (100%; 923), and 1 (98.5%; 915) for Candida albicans; 8 (100%; 352), 4 (99.2%; 392), 2 (99.2%; 480), 1 (99.8%; 603), and 0.5 (97.9%; 635) for C. parapsilosis; 1 (99.2%; 123), 0.12 (99.2%; 121), 0.25 (99.2%; 138), 2 (100%; 171), and 0.5 (97.2%; 175) for C. tropicalis; 0.12 (96.6%; 174), 0.06 (96%; 176), 0.25 (98.4%; 188), 2 (100%; 209), and 0.25 (97.6%; 208) for C. glabrata; 0.25 (97%; 33), 0.5 (93.9%; 33), 1 (91.9%; 37), 4 (100%; 51), and 32 (100%; 53) for C. krusei; and 4 (100%; 33), 2 (100%; 33), 2 (100%; 54), 1 (100%; 90), and 0.25 (93.4%; 91) for C. orthopsilosis. The three statistical methods gave similar ECVs (within one dilution) and included ≥95% of isolates. These tentative ECVs would be useful for monitoring the emergence of isolates with reduced susceptibility by use of the SYO method.

INTRODUCTION

Fungemia is an important cause of morbidity and mortality worldwide and is closely associated with high health care costs for hospitalized patients with serious underlying conditions (17, 24). Although Candida albicans is still the leading cause of fungemia, an epidemiological switch in etiology is being observed, with an increase in non-Candida albicans species and molds as causative agents. Rapid identification and antifungal susceptibility testing are very important for treatment. Echinocandins (anidulafungin [AND], caspofungin [CAS], and micafungin [MCF]) and amphotericin B (AMB) are the most frequently used antifungal agents for treatment of invasive Candida fungemia, as recommended in current guidelines (IDSA, FDA, etc.) (15).

Recently, Pfaller et al. used the CLSI methodology to study MIC distributions and define the epidemiological cutoff values (ECVs) of echinocandins for 11 Candida species, including the most common and some of the more uncommon species, as well as the ECVs of AMB and flucytosine (FC) (18, 20, 23). Species-specific clinical breakpoints (CBP) have been proposed for echinocandins (for C. albicans, Candida tropicalis, Candida glabrata, Candida parapsilosis, Candida guilliermondii, and Candida krusei), based on pharmacokinetic (PK) and pharmacodynamic (PD) parameters, clinical outcomes related to MIC values, MIC distributions, and mechanisms of resistance. For AMB and FC, only ECVs have been reported (22, 23). To determine the ECVs, Turnidge et al. and Kronvall each proposed a statistical method, and recently Meletiadis et al. used another one to calculate these values for Aspergillus fumigatus (12, 14, 27). Of these three proposed methods, that of Turnidge et al. is the most commonly used one. We have used another statistical method (the clustering method) which considers the MIC distribution of an agent for a determined species as a mixture of different subpopulations, each with its own normal distribution. As the goal of any susceptibility test is to identify resistant isolates, in the absence of CBP, ECVs are useful for separating isolates without mechanisms of resistance, known as wild-type (WT) isolates, from non-WT isolates (those that can harbor some resistance mechanisms). The CBP given by CLSI must be applied when tests are performed using the CLSI methodology. However, this methodology is cumbersome for clinical laboratories, which gave rise to the use of commercial methods such as the Sensititre YeastOne (SYO) test, a method that is widely used to determine the susceptibility of Candida spp. in many European laboratories. Since no CBP are available for this method, the purposes of our study were (i) to define the WT MIC distributions of the three echinocandins, AMB, and FC by the SYO method for four of the most common and two less common Candida species causing bloodstream infections; (ii) to propose the ECV for each species-drug combination for the SYO method; (iii) to assay another statistical method (clustering method) to determine the ECVs; and (iv) to compare the ECVs obtained by different methods with each other and also with those obtained by the CLSI methodology.

MATERIALS AND METHODS

Isolates.

A total of 2,119 bloodstream isolates were tested, including 857 isolates belonging to just one center (334 C. albicans, 315 C. parapsilosis, 56 C. glabrata, 62 C. tropicalis, 63 C. orthopsilosis, and 27 C. krusei isolates) and isolated from January 1995 to December 2010 and 1,262 isolates (610 C. albicans, 327 C. parapsilosis, 156 C. glabrata, 113 C. tropicalis, 30 Candida orthopsilosis, and 26 C. krusei isolates) obtained from 43 public tertiary care hospital centers representing all Spanish geographical areas (FUNGEMYCA epidemiological study) and isolated from January 2009 to February 2010. Each isolate represented one infectious episode per patient and was identified by its metabolic properties via Vitek-2 (bioMérieux, Marcy-l'Etoile, France) or Auxacolor (Bio-Rad, Madrid, Spain) methods in each center, stored in a water suspension, and sent to the reference center (Hospital Universitario La Fe, Valencia, Spain) for posterior studies. Four C. albicans strains with reduced CAS susceptibility (2 heterozygous and 2 homozygous for FKS1 gene mutation) were included to assess the ability of the methods to identify non-WT isolates (3, 5, 13).

All isolates of the C. parapsilosis complex were identified by molecular methods as described elsewhere (4, 10).

Antifungal susceptibility testing.

The susceptibility test was performed at the participating hospitals on the first isolate from each fungemia episode by the microdilution colorimetric SYO method, using the SYO-09 panel (Trek Diagnostic Systems, Cleveland, OH) as instructed in the commercial guidelines. The quality control strains, C. krusei ATCC 6258 and C. parapsilosis ATCC 22019, were first tested in each participating laboratory, and results were sent to the reference center. All MIC values were within the expected range.

Definitions.

The definitions of the WT population and ECV were those reported previously by other authors (6, 11, 25). A WT population is the subpopulation of isolates/MICs for a species-drug combination with no acquired phenotypically detectable resistance mechanisms.

The ECV is the highest MIC value of the WT population. It is calculated by taking into account the MIC distribution, the modal MIC of each distribution, and the inherent variability of the test (usually within one doubling dilution) and should encompass ≥95% of isolates. The number of isolates needed to calculate a representative ECV is not established, but there is a consensus among experts that recommends at least 50 strains from at least three to five different laboratories.

Statistical analysis.

Data were analyzed with R software (version 2.14.2). Both on-scale and off-scale MICs were included, with the values left unchanged. In order to approach a normal distribution, the MICs were converted to log₂ values. Statistical ECVs were calculated following the methods described by Turnidge et al. and Kronvall (12, 27). Furthermore, a clustering method was performed using the Mclust library for R (version 3.4.11) (8, 9). To assess the ECVs of the different species for the different antifungals, instead of each species being considered a homogeneous group with a normal distribution, it was treated as a mixture of different subpopulations, each with its own normal distribution. A Gaussian mixture model is given by the following equation:

$$p\left(x/\lambda \right)-{\displaystyle \sum _{t-1}^M{w}_{1}g}\left(x|u_i,{\Sigma }_i\right)$$

where x is a continuous variable, λ stands for the mean vectors, covariance matrices, and mixture weights from all components, w_i is the probability that an observation belongs to the ith subpopulation, and g(x|μ_i, Σ_i) shows the component Gaussian densities. In order to approach a normal mixture distribution, the log₂ MIC values were smoothed by a kernel density algorithm (26). The clustering technique selects the number of normal components in the mixture by estimation of the most parsimonious model by use of the Bayesian information criterion (BIC). It is then able to discern the different normal distributions in the mixture, producing a probabilistic clustering that quantifies the uncertainty of observations belonging to the different subpopulations. Consequently, the ECV was set at the point of maximal uncertainty between the most resistant subpopulation and the others (Fig. 1). Since the ECVs estimated by the clustering method are on a continuous scale, values were rounded to the nearest higher dilution after reconversion to concentration units.

Fig 1.

Example of a normal mixture distribution (A) and distribution of normal components of the mixture (B). (C) Distribution of Candida tropicalis (CAT) with anidulafungin (AND). (D) Kernel density estimation and determination of epidemiological cutoff value by the clustering method.

RESULTS AND DISCUSSION

Several methods have been reported for determination of WT ECVs. For instance, Arendrup et al. (2) estimated EVCs as 2-fold dilution steps higher than the MIC₅₀, Rodríguez-Tudela et al. estimated them as 2-fold dilutions above the modal MIC (25), and Kronvall and Turnidge et al. calculated ECVs by statistical methods; the latter authors concluded that the values thus obtained fit with those determined through the MIC₅₀ or modal MIC (6, 12, 27). We used the clustering statistical method to determine ECVs and compared them with those obtained by other methods and with those reported for the CLSI methodology. Finally, the ECVs proposed are median values for the five methods analyzed.

For each species and antifungal agent, the MIC data, even for the species with few isolates, were obtained from more than 5 laboratories, and the modal MIC of the antifungal was within one dilution. Table 1 shows the MIC values of echinocandins for C. albicans strains with fks mutations. The in vitro activities of the antifungal agents tested, which have been reported elsewhere (16), were similar to those observed by other authors using the SYO method (1, 3, 7, 21). Table 2 depicts the WT MIC distributions of the five antifungal agents tested. All MIC distributions are typical for WT organisms and cover three to five 2-fold dilution steps surrounding the modal MIC, except those of AND (n = 33) for C. orthopsilosis, AMB for C. albicans (n = 923) and C. krusei (n = 51), and MCF for C. krusei (n = 33), with only one dilution above the modal MIC. The MIC distribution of AND for C. tropicalis (Fig. 1C) is not a typical Gaussian distribution, which may suggest that there is a complex of different genotypes with different susceptibility patterns, as occurs with the C. parapsilosis and C. glabrata complex. Table 3 shows the ECVs obtained for the SYO method, determined by the different above-mentioned methods (three statistical methods, those using two dilutions above the MIC₅₀ and modal MIC, and the median ECV of the five methods) and those published by Pfaller et al. for the CLSI. In general, the median ECVs for the SYO method were very similar to those for the CLSI method: 82.1% of ECVs were equal to or within one 2-fold dilution of those reported for the CLSI method, in agreement with the MIC values for contrasting these two methods (3, 21). The greatest difference was for C. tropicalis, for which the ECVs of AND were 3 dilutions higher for the SYO method. The statistical method proposed by Turnidge et al. gave lower ECVs than those proposed by Kronvall, although 83.3% were within one dilution. By the clustering method, the ECVs were lower (in 50% of cases) than those obtained by the method used by Turnidge et al.; nevertheless, 83.3% of ECVs were within one 2-fold dilution. For the echinocandin agents, the SYO median ECVs were one 2-fold dilution higher than those for the CLSI method, except for those for CAS (C. krusei and C. orthopsilosis) and MCF (C. krusei), which were two 2-fold dilutions higher, and that of AND for C. tropicalis, which was three dilutions higher. In the latter case, the ECVs obtained by use of the modal MIC and MIC₅₀ were one and two dilutions higher, respectively, than the ECVs for the CLSI method. The ECVs obtained by the MIC₅₀ and the modal MIC were lower, in general, than those obtained by applying statistical methods; however, 100 and 94.4% of the cases, respectively, were within one dilution of the median ECV (Table 3). All methods used to determine the ECV for echinocandins classified the C. albicans strains with fks mutations as non-WT strains.

TABLE 1.

MICs of echinocandins for Candida albicans strains with reduced caspofungin susceptibility^a

Strain	MIC (μg/ml)
	Anidulafungin	Caspofungin	Micafungin
CAI4 (wild type)	0.016	0.06	0.008
CAI4RI (heterozygote)	1	4	0.5
NR2 (homozygote)	8	8	8
NR3 (homozygote)	0.5	4	1
NR4 (heterozygote)	0.5	2	0.5

^aMICs were determined by the Sensititre YeastOne method. Descriptions of these strains can be found in references 5 and 13.

TABLE 2.

Wild-type MIC distributions of five antifungal agents for six species of Candida by Sensititre YeastOne method

Agent	Species	No. of isolates with MIC (μg/ml)														Total no. of isolates
		0.008	0.015	0.03	0.06	0.125	0.25	0.5	1	2	4	8	16	32	64
Anidulafungin	C. albicans		289	140	125	86	6	3	5	1		1				656
	C. parapsilosis		11	4	5	4	10	6	149	157	5	1				352
	C. tropicalis		27	13	24	39	16	3				1				123
	C. glabrata		56	80	28	4		2	3		1					174
	C. krusei		5	7	12	6	2					1				33
	C. orthopsilosis			1		2	6	8	15	1						33
Caspofungin	C. albicans	19	75	242	248	114	39	7	2	1						747
	C. parapsilosis	5	7	7	8	26	112	178	105	38	3	1				490
	C. tropicalis	2	11	39	49	25	11	1								138
	C. glabrata	7	5	41	95	28	9		3							188
	C. krusei			1	4	1	20	6	2	1	1	1				37
	C. orthopsilosis	1		1	1	8	11	18	10	4						54
Micafungin	C. albicans	260	277	79	11	8	8	8	4	2		2				659
	C. parapsilosis	2	9	4	2	1	17	80	154	112	8	3				392
	C. tropicalis	6	28	57	26	3						1				121
	C. glabrata	38	99	28	2		2	2	3			2				176
	C. krusei			2	4	25			1			1				33
	C. orthopsilosis		1	1	0	1	11	15	3	1						33
Amphotericin B	C. albicans			11	23	177	334	352	26							923
	C. parapsilosis		7	2	22	140	205	183	43	1						603
	C. tropicalis				2	13	42	87	25	2						171
	C. glabrata					28	56	87	37	1						209
	C. krusei					4	8	16	21	2						51
	C. orthopsilosis			3	3	30	33	19	2							90
Flucytosine	C. albicans				568	174	84	56	19	3	3	3		1	4	915
	C. parapsilosis				336	169	90	27	10		1	1			1	635
	C. tropicalis				134	27	8	1	1	2	1			1		175
	C. glabrata				193	6	4			1	3	1				208
	C. krusei				1			1	3	2	9	19	17	1		53
	C. orthopsilosis				56	22	7	3	1		1				1	91

TABLE 3.

Comparison of ECVs obtained for five antifungal agents by use of different methods

Species	No. of isolates tested	Agenta	ECV obtained by indicated method (%)b
			MIC50 + 2 dilutions	Modal MIC + 2 dilutions	Method of Turnidge et al.	Method of Kronvallc	Clustering methodc	Median for all 5 studied methods	CLSI methodd
C. albicans	656	AND	0.12 (99.7)	0.06 (84.45)	0.25 (98.5)	0.25 (95.65)	0.25 (98.63)	0.25 (98.5)	0.12 (99.7)
	659	MCF	0.06 (95.1)	0.06 (95.1)	0.06 (95.1)	0.12 (94.95)	0.06 (94.99)	0.06 (95.1)	0.03 (97.7)
	747	CAS	0.25 (98.7)	0.25 (98.7)	0.25 (98.7)	0.5 (98.99)	0.25 (98.66)	0.25 (98.7)	0.12 (99.8)
	923	AMB	1 (100)	2 00)	2 (100)	2 (100)	1 (100)	2 (100)	2 (99.8)
	915	FLC	0.25 (90.27)	0.25 (90.27)	1 (96.4)	1 (96.99)	2 (98.8)	1 (98.5)	0.5 (94.2)
C. parapsilosis	352	AND	4 (99.71)	8 (100)	8 (100)	8 (99.87)	4 (99.75)	8 (100)	4 (100)e
	392	MCF	4 (99.23)	4 (99.23)	8 (100)	8 (99.36)	4 (99.24)	4 (99.2)	4 (100)e
	490	CAS	2 (99.2)	2 (99.2)	4 (99.8)	4 (99.39)	2 (99.19)	2 (99.2)	1(98.6)e
	603	AMB	1 (99.83)	1 (99.83)	2 (100)	2 (99.92)	1 (100)	1 (99.8)	2 (99.7)
	635	FLC	0.25 (93.7)	0.25 (93.7)	0.5 (97.9)	1 (98.48)	2 (99.52)	0.5 (97.9)	0.5 (98.7)
C. tropicalis	123	AND	0.25 (96.75)	0.5 (99.2)	1 (99.2)	1 (99.59)	1 (99.18)	1 (99.2)	0.12 (98.9)
	121	MCF	0.12 (99.2)	0.12 (99.2)	0.12 (99.2)	0.25 (99.58)	0.5 (99.17)	0.12 (99.2)	0.12 (99.1)
	138	CAS	0.25 (99.3)	0.25 (99.3)	0.5 (100)	0.5 (99.64)	0.25 (99.28)	0.25 (99.2)	0.12 (99.4)
	171	AMB	2 (100)	2 (100)	2 (100)	2 (99.11)	1 (98.82)	2 (100)	2 (99.8)
	175	FLC	0.25 (96.57)	0.25 (96.57)	0.5 (97.1)	0.5 (97.31)	0.5 (97.65)	0.5 (97.2)	0.5 (93.0)
C. glabrata	174	AND	0.12 (96.55)	0.12 (96.55)	0.12 (96.55)	0.25 (97.1)	0.25 (96.55)	0.12 (96.6)	0.25 (99.4)
	174	MCF	0.06 (95.97)	0.06 (95.97)	0.06 (95.97)	0.25 (96.33)	0.06 (95.98)	0.06 (96.0)	0.03 (98.2)
	188	CAS	0.25 (98.4)	0.25 (98.4)	0.25 (98.4)	0.5 (98.65)	0.12 (93.62)	0.25 (98.4)	0.12 (98.5)
	209	AMB	2 (100)	2 (100)	2 (100)	2 (99.76)	1 (99.52)	2 (100)	2 (99.6
	208	FLC	0.25 (97.6)	0.25 (97.6)	0.25 (97.6)	0.5 (97.83)	0.12 (95.67)	0.25 (97.6)	0.5 (9800)
C. krusei	33	AND	0.25 (96.97)	0.25 (96.97)	0.25 (97.0)	0.5 (98.44)	1 (96.97)	0.25 (97.0)	0.12 (99.3)
	33	MCF	0.5 (93.94)	0.5 (93.94)	0.25 (94.0)	1 (96.83)	0.25 (93.94)	0.5 (93.9)	0.12 (97.8)
	37	CAS	1 (91.89)	1 (91.89)	2 (94.6)	4 (95.77)	1 (91.89)	1 (91.9)	0.25 (96.3)
	51	AMB	2 (100)	4 (100)	4 (100)	4 (100)	2 (100)	4 (100)	2 (99.3)
	53	FLC	32 (100)	32 (100)	64 (100)	64 (100)	16 (98.11)	32 (100)	32 (94.4)
C. orthopsilosis	33	AND	2 (100)	4 (100)	4 (100)	4 (98.46)	2 (97.06)	4 (100)	2 (100)
	33	MCF	2 (100)	2 (100)	4 (100)	4 (100)	2 (100)	2 (100)	1 (100)
	54	CAS	2 (100)	2 (100)	4 (100)	4 (100)	2 (100)	2 (100)	0.5 (100)
	90	AMB	1 (100)	1 (100)	1 (100)	1 (100)	1 (100)	1 (100)	ND
	91	FLC	0.25 (93.41)	0.25 (93.41)	1 (97.8)	0.5 (96.07)	0.25 (94.44)	0.25 (93.4)	ND

^aAND, anidulafungin; MCF, micafungin; CAS, caspofungin; AMB, amphotericin B; FLC, flucytosine; ND, not determined.

^bThe percentage shows the percentage of isolates for which the MIC is less than or equal to the ECV (μg/ml).

^cThe percentage shows the percentage of isolates for which the MIC is less than or equal to the ECV for the modeled population.

^dData are from references 18 and 20.

^eData for the C. parapsilosis complex.

The ECVs of AMB were very similar by all methods: the methods of Turnidge et al. and Kronvall gave the same values, and by the clustering method, the ECVs were one 2-fold dilution lower, except for C. orthopsilosis, for which the ECVs were the same, independent of the method used.

Pearson's correlation between the median ECVs for the SYO method and the values for each of the other methods, and also with the CLSI method, was very good (≥0.98 and 0.99, respectively). In comparing species, the correlation was ≥0.94, except for C. parapsilosis and C. tropicalis (0.84), and in comparing antifungal agents, the correlation was ≥0.9, except for CAS (0.88).

The ECVs were also calculated for the distribution of MICs of the two sources of isolates individually, i.e., 1,262 isolates from 43 laboratories and 857 isolates from one center. The results obtained, in general, were the same as those found with all 2,119 isolates or fell within one 2-fold dilution, which is in agreement with the work of Turnidge et al. and Kronvall, who reported that their methods are also valid for data sets generated by a single laboratory and method (12, 27). It is difficult to determine which statistical method defines the ECV best, due to the scarcity of resistant isolates. Perhaps the optimal method depends on the heterogeneity of the MIC distribution, since different species may have different numbers of subpopulations and, consequently, different MIC distribution shapes (unimodal, bimodal, or multimodal). In fact, the majority of differences, in general, were within only one 2-fold dilution, and furthermore, Pearson's correlation between methods was very good, so we have reported the median value for the five methods evaluated as the tentative ECV for the SYO method. Moreover, independent of the method used, a 5% risk of misidentification is always assumed (in fact, it is 2.5%, because the distribution has two tails). For a collection of 229 WT isolates and 50 isolates with fks mutations, Pfaller et al. showed that the ECV of anidulafungin for the CLSI broth microdilution method correctly classified 90% of mutant strains and 98% of those without fks mutations (22).

All ECVs included ≥95% of isolates, except for those of CAS and MCF for C. krusei, which included ≥92% and 94% of isolates, respectively. This may have been due to the small number of isolates, but other authors who studied larger numbers of isolates also reported ECVs that did not include 95% of isolates (19, 23).

The advantage of this study is that all isolates used were obtained from patients without prior treatment with any of the antifungal agents tested; thus, the outlier isolates probably represent mutated strains. The data came from 43 laboratories, and all isolates were obtained from blood cultures. Among the limitations of the study is the small number of C. krusei and C. orthopsilosis isolates tested with echinocandins, but we have kept the tentative ECV for these species just for comparison among methods. Moreover, all isolates came from just one country, so these values may not be completely representative but could be shared with other data sets in the future to establish future ECVs for the SYO method. The small number of strains with known mechanisms of resistance is another limitation. We propose, for the first time, tentative ECVs for the SYO method to help in monitoring the development of resistance to drugs by use of this method, which is the aim of susceptibility tests as well as ECVs.