Clostridium difficile PCR Cycle Threshold Predicts Free Toxin

ABSTRACT

There is no stand-alone Clostridium difficile diagnostic that can sensitively and rapidly detect fecal free toxins. We investigated the performance of the C. difficile PCR cycle threshold (C_T) for predicting free toxin status. Consecutive stool samples (n = 312) positive for toxigenic C. difficile by the GeneXpert C. difficile/Epi tcdB PCR assay were tested with the rapid membrane C. Diff Quik Chek Complete immunoassay (RMEIA). RMEIA toxin-negative samples were tested with the cell cytotoxicity neutralization assay (CCNA) and tgcBIOMICS enzyme-linked immunosorbent assay (ELISA). Using RMEIA alone or in combination with CCNA and/or ELISA as the reference method, the accuracy of C_T was measured at different C_T cutoffs. Using RMEIA as the reference method, a C_T cutoff of 26.35 detected toxin-positive samples with a sensitivity, specificity, positive predictive value, and negative predictive value of 96.0% (95% confidence interval [CI], 90.2% to 98.9%), 65.9% (95% CI, 59.0% to 72.2%), 57.4% (95% CI, 52.7% to 62%), and 97.1% (95% CI, 92.8% to 98.9), respectively. Inclusion of CCNA in the reference method improved C_T specificity to 78.0% (95% CI, 70.7% to 84.2%). Intercartridge lot C_T variability measured as the average coefficient of variation was 2.8% (95% CI, 1.2% to 3.2%). Standardizing the input stool volume did not improve C_T toxin specificity. The median C_T values were not significantly different between stool samples with Bristol scores of 5, 6, and 7, between pediatric and adult samples, or between presumptive 027 and non-027 strains. In addition to sensitively detecting toxigenic C. difficile in stool, on-demand PCR may also be used to accurately predict toxin-negative stool samples, thus providing additional results in PCR-positive stool samples to guide therapy.

INTRODUCTION

Clostridium difficile is a cause of antibiotic-associated diarrhea (1). Laboratory assays employed to evaluate patients with suspected C. difficile infection (CDI) include enzyme immunoassay (EIA), which detects fecal free toxins (TcdA and TcdB), and PCR, which targets sequences encoding tcdA and/or tcdB (2). EIA glutamate dehydrogenase (GDH) is also used to detect C. difficile, but follow-up EIA toxin or PCR is needed to confirm toxigenicity. The best laboratory algorithm for diagnosing C. difficile infection (CDI) is debated (3). Compared to EIA toxin, PCR has a higher sensitivity for detection of toxigenic C. difficile in stool (4). However, studies suggest toxin positivity more accurately correlates with clinical outcomes compared with PCR (5–8). Polage and colleagues showed that patients with EIA toxin-negative (toxin⁻)/PCR-positive (PCR⁺) results had no CDI-related complications compared with EIA toxin⁺/PCR⁺ patients (7). These studies indicate that PCR is linked to overdiagnosis of patients colonized with C. difficile, triggering unnecessary antibiotic therapy. However, not all studies support this view (9–12). Some experts recommend using highly sensitive PCR to avoid missing toxin⁻/PCR⁺ CDI cases because EIA toxin is less sensitive compared with cell cytotoxicity neutralization assay (CCNA), which is the gold standard for fecal free toxin but takes several days to perform (3). To address these diagnostic challenges, European guidelines recommend a multistep testing algorithm starting with PCR or EIA GDH and followed by EIA toxin to rapidly identify toxin-positive and toxin⁻/PCR⁺ patients to facilitate appropriate clinical decision-making (4). Given the requirement for multistep testing, having a single stand-alone assay that can rapidly and sensitively detect toxigenic C. difficile and simultaneously predict free toxins would be valuable for guiding therapy and infection prevention practices.

Several studies have shown a correlation between C. difficile fecal free toxins and bacterial/genomic burden or PCR cycle threshold (C_T) in diarrheal stools (11, 13–15). Leslie and colleagues showed that at a cutoff of ≥5.1 log₁₀ DNA copies/ml, they could correctly classify ≥95% of toxin-positive and 70% of toxin-negative stool samples (15). However, performance characteristics of PCR C_T cutoffs for discriminating toxin-positive and toxin-negative stool samples have not been reported. The aim of this study was to measure the accuracy of C. difficile PCR C_T toxin for predicting toxin-positive stool samples using EIA toxin and other toxin assays as the reference methods.

RESULTS

Study samples.

A total of 312 GeneXpert tcdB PCR-positive unformed stools from 282 unique patients were included in this study. Stool volume was insufficient to perform the enzyme-linked immunosorbent assay (ELISA) on 18 stool samples. Twenty-one patients provided two samples, 3 patients provided three samples, and 1 patient provided four samples. Of the patients, 48.9% (138) were female and 77.7% (219) were aged 18 years or older. Overall, 55.8% (174/312) of samples originated from inpatients, 17.2% (30/174) of which were from intensive care unit (ICU) patients.

Accuracy of C_T toxin.

The median GeneXpert tcdB PCR C_T value was significantly lower in rapid membrane C. Diff Quik Chek Complete immunoassay (RMEIA) toxin-positive samples (23.3 [interquartile range (IQR), 21.6 to 24.3]) compared with toxin-negative samples (29.2 [IQR, 24.5 to 32.7]; P < 0.001) (Fig. 1A). Using RMEIA toxin as the reference method for free toxins and assigning equal weight to C_T toxin sensitivity and specificity, the C_T cutoff 26.35 yielded a sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of 96.0% (95% CI, 90.2% to 98.9%), 65.9% (95% CI, 59.0% to 72.2%), 57.4% (95% CI, 52.7% to 62.0%), and 97.1% (95% CI, 92.8% to 98.9%), respectively (Fig. 1A and Table 1). To determine whether the specificity could be further improved by standardizing the input stool volume, PCR was repeated on the same stool samples using a scoop to transfer a standardized volume of stool. With standardized stool volume, the C_T cutoff 26.35 yielded a sensitivity, specificity, PPV, and NPV of 93.1% (95% CI, 86.2% to 97.2%), 60.7% (95% CI, 53.7% to 67.3%), 53.1% (95% CI, 48.7% to 57.4%), and 94.8% (95% CI, 89.9% to 97.4%), respectively (Fig. 1B). The sensitivity and specificity of C_T toxin were not significantly different using nonstandardized (swab) and standardized (scoop) stool volume (P = 0.5 and 0.6, respectively).

FIG 1.

C. difficile tcdB PCR cycle threshold in PCR-positive toxin-positive and toxin-negative stool samples. GeneXpert tcdB PCR was performed using nonstandardized (A) and standardized (B) stool volume. Toxin result was determined using the C. Diff Quik Chek Complete RMEIA. The horizontal lines represent median cycle thresholds. Cycle threshold cutoffs—derived based on equal weight assigned to sensitivity and specificity—are marked with dashed lines.

TABLE 1.

Performance of C. difficile tcdB PCR C_T toxin using different reference methods^a

Reference method	CT cutoffb	Sensitivity (% [n/N]) (95% CI)	Specificity (% [n/N]) (95% CI)	PPV (% [n/N]) (95% CI)	NPV (% [n/N]) (95% CI)
RMEIA	26.35	96.0 (97/101) (90.2–98.9)	65.9 (139/211) (59–72.2)	57.4 (97/169) (52.7–62.0)	97.1 (139/143) (92.8–98.9)
RMEIA or CCNA	26.35	87.6 (134/153) (81.3–92.4)	78.0 (124/159) (70.7–84.2)	79.3 (134/169) (73.9–83.8)	86.7 (124/143) (81–90.9)
RMEIA or ELISA	26.35	94.3 (116/123) (88.6–97.7)	70.2 (120/171) (62.7–76.9)	69.5 (116/167) (64.3–74.2)	94.5 (120/127) (89.2–97.3)
RMEIA, CCNA, or ELISA	26.85	89.4 (143/160) (83.5–93.7)	76.3 (103/135) (68.2–83.2)	81.7 (143/175) (76.7–85.9)	85.8 (103/120) (79.3–90.6)

^aStool volume was insufficient in 18 samples to perform ELISA.

^bC_T cutoff was based on equal weight assigned to sensitivity and specificity; 26.85 is the correct cutoff for RMEIA, CCNA, or ELISA.

Given that RMEIA toxin can be falsely negative (4), we further tested RMEIA toxin-negative stool samples using the cell cytotoxicity neutralization assay (CCNA), which detects TcdB, and the tgcBIOMICS ELISA, which detects TcdA and TcdB. Inclusion of CCNA in the reference method compared with RMEIA alone improved C_T toxin specificity (at a C_T cutoff of 26.35) from 65.9% to 78.0% (95% CI, 70.7% to 84.2%) and decreased sensitivity from 96.0% to 87.6% (95% CI, 81.3 to 92.4) (Fig. 2B and Table 1). C_T toxin was positive in 71.2% (37/52) of RMEIA toxin-negative/CCNA toxin-positive samples. Inclusion of tgcBIOMICS ELISA in the reference method compared with RMEIA alone improved C_T specificity (at a C_T cutoff of 26.35) to 70.2% (95% CI, 62.7% to 76.9%) and decreased sensitivity to 94.3% (95% CI, 88.6% to 97.7%) (Fig. 2C and Table 1). C_T toxin was positive in 90.9% (20/22) of RMEIA toxin-negative/ELISA toxin-positive samples. The performance characteristics of C_T toxin using various reference methods with equal and unequal (i.e., sensitivity fixed at ≥99%) weights assigned to sensitivity and specificity are shown in Tables 1 and 2, respectively.

FIG 2.

C. difficile tcdB PCR cycle threshold in PCR-positive toxin-positive and toxin-negative stool samples using various toxin reference methods. Free toxin was detected using RMEIA alone (A), RMEIA or CCNA (B), RMEIA or ELISA (C), and RMEIA, CCNA, or ELISA (D). Red dots show samples that were toxin negative by RMEIA but positive by CCNA and/or ELISA. The horizontal lines represent median cycle thresholds. Cycle threshold cutoffs—derived based on equal weight assigned to sensitivity and specificity—are marked with dashed lines.

TABLE 2.

Performance of C. difficile tcdB PCR C_T toxin using different reference methods with sensitivity fixed at ≥99%^a

Reference method	CT cutoff	Sensitivity (% [n/N]) (95% CI)	Specificity (% [n/N]) (95% CI)	PPV (% [n/N]) (95% CI)	NPV (% [n/N]) (95% CI)
RMEIA	27.55	99.0 (100/101) (94.6–100)	58.8 (124/211) (51.8–65.5)	53.5 (100/187) (49.4–57.5)	99.2 (124/125) (94.6–99.9)
RMEIA or CCNA	30.85	99.3 (152/153) (96.4–100)	49.1 (78/159) (41.0–57.1)	65.2 (152/233) (61.7–68.6)	98.7 (78/79) (91.7–99.8)
RMEIA or ELISA	29.15	99.2 (122/123) (95.5–100)	53.8 (92/171) (46.0–61.4)	60.7 (122/201) (56.8–64.5)	98.9 (92/93) (92.9–99.8)
RMEIA, CCNA or ELISA	30.85	99.4 (159/160) (96.6–100)	50.4 (68/135) (41.6–59.1)	70.3 (159/226) (66.7–73.8)	98.5 (68/69) (90.5–99.8)

^aStool volume was insufficient in 18 samples to perform ELISA.

Exploratory investigation of false-positive C_T toxin results.

Although our study was not powered for subanalyses, we performed several exploratory analyses to look for potential explanations for the false-positive C_T toxin results. With the reference method defined as toxin positivity with RMEIA, CCNA, or ELISA, there were 29 stool samples from 29 patients who were falsely categorized as toxin positive with C_T toxin. First, we determined whether false-positive C_T toxin results were due to longer time to stool processing (i.e., testing with RMEIA and freezing samples for CCNA and ELISA). The median time to stool processing was numerically longer in the 29 false-positive C_T toxin samples versus the 138 correctly C_T toxin-positive samples, but the difference was not statistically significant (32.9 h [IQR, 17.6 to 56.0] versus 23.3 h [IQR, 17.6 to 48.5]; P = 0.4]). Two out of 29 samples with false-positive results were processed after 72 h of collection. Second, we determined whether anti-C. difficile antibiotic exposure (i.e., resulting in dead organisms not producing toxin but detected by PCR) or laxative therapy (i.e., causing dilution of toxin to below level of detection) could explain the false-positive C_T toxin results. After matching patients for age, sex, and inpatient status, antibiotic exposure (treatment for C. difficile or other bacteria) and laxative therapy were not significantly different in patients with true-positive and those with false-positive C_T toxin results (see Table S1 in the supplemental material). Third, we determined whether stool quality (i.e., Bristol score) was different in stool samples with false-positive C_T toxin results. The proportion of Bristol 5, 6, and 7 stools was not significantly different in samples with true-positive and those with false-positive C_T toxin results (Bristol 5, 6.5% [9/138] versus 10.3% [3/29]; Bristol 6, 57.2% [79/138] versus 65.5% [19/29]; Bristol 7, 36.2% [50/138] versus 24.1% [7/29]; P = 0.37 for all comparisons). Lastly, we determined whether presumptive 027 and non-027 strains were differentially represented in samples with false-positive C_T toxin results. The proportion of those with presumptive 027 infection was smaller in the 29 false-positive C_T toxin samples compared with the 138 correctly C_T toxin-positive samples, but the difference was not statistically significant (3.5% [1/29] versus 17.4% [24/138]; P = 0.08).

Intercartridge C_T reproducibility.

The average coefficient of variation of GeneXpert tcdB PCR C_T values for 20 randomly selected positive stool samples tested with four different cartridge lots was 2.8% (95% CI 1.19% to 3.16%). At a C_T cutoff of 26.35, one (1.3%) out of 80 PCR runs had a discordant C_T toxin result. The C_T values and standard deviation for quadruplicate runs are shown in Table 3.

TABLE 3.

Intercartridge C_T reproducibility with GeneXpert tcdB PCR using four cartridge lots^a

Sample no.	CT value for:				SD
	Lot 1	Lot 2	Lot 3	Lot 4
1	18.4	18.5	18.5	19.1	0.3
2	19.5	22.1	19.6	19.8	1.2
3	20.4	20.4	20.4	20.6	0.1
4	20.8	21.5	22.2	21.4	0.6
5	21.4	21.2	20.3	20.9	0.5
6	22.2	21.1	21.4	21.4	0.5
7	22.4	22	21.7	21.7	0.3
8	22.6	23.3	22.7	22.6	0.3
9	23.3	23.4	23.1	23.4	0.1
10	23.6	24.1	23.6	24	0.3
11	24.2	23.3	24.3	23.6	0.5
12	24.3	24.3	24.1	24.2	0.1
13	26.3	24.5	24.2	29.3	2.3
14	30.5	30.3	30.5	30.4	0.1
15	30.8	31.3	30.9	30.9	0.2
16	30.9	29.8	29.6	29.9	0.6
17	31.6	32.2	32.5	31.8	0.4
18	31.9	31.9	31.5	32.3	0.3
19	35.4	34.2	34.2	36.5	1.1
20	38.8	36.7	36.4	38.9	1.3

^aRandomly selected PCR-positive stool samples were tested. Cartridge lot numbers included 1000037261, 1000037262, 1000037264, and 1000037265.

Impact of stool quality on C_T.

To determine whether the C_T cutoff may be impacted by stool quality, we compared the median C_T values in stool samples with Bristol scores 5, 6, and 7. As shown in Fig. 3, the median C_T values were not significantly different in RMEIA toxin-positive (Bristol 7 versus 5, 23.7 [IQR, 21.7 to 25.2] versus 23.6 [IQR, 22.1 to 24.1]; P = 0.7; Bristol 7 versus 6, 23.7 [IQR, 21.7 to 25.2] versus 22.6 [IQR, 21.3 to 24]; P = 0.19) and RMEIA toxin-negative (Bristol 7 versus 5, 29.2 [IQR, 23.8 to 32.8] versus 30.1 [IQR, 24.9 to 32.5]; P = 0.71; Bristol 7 versus 6, 29.2 [IQR, 23.8 to 32.8] versus 29.1 [IQR, 24.5 to 32.7]; P = 0.8) stool samples.

FIG 3.

C. difficile tcdB PCR cycle threshold in PCR-positive stool samples with different Bristol scores. C. Diff Quik Chek Complete RMEIA toxin-positive stool samples with Bristol scores 5 (n = 8), 6 (n = 58), and 7 (n = 35), and RMEIA toxin-negative stool samples with Bristol scores 5 (n = 13), 6 (n = 124), and 7 (n = 74) were included. The horizontal lines represent the median cycle thresholds. Median C_T values for Bristol 5 and 6 were compared to that for Bristol 7.

Impact of patient's age on C_T.

To determine whether the C_T cutoff may be impacted by patient's age, we compared the median C_T values in adult and pediatric patients. As show in Fig. 4, the median C_T values were not significantly different between adult and pediatric patients with RMEIA toxin-positive (23.1 [IQR, 21.6 to 24.3] versus 23.7 [IQR, 21.5 to 24.8]; P = 0.74) and RMEIA toxin-negative (29.2 [IQR, 24.3 to 33.0] versus 29 [IQR, 24.8 to 31.8]; P = 0.4) stool samples.

FIG 4.

C. difficile tcdB PCR cycle threshold in PCR-positive pediatric and adult stool samples. C. Diff Quik Chek Complete RMEIA toxin-positive stool samples from pediatric (n = 23) and adult (n = 78) patients and from toxin-negative pediatric (n = 44) and adult (n = 167) patients were included. The horizontal lines represent the median cycle thresholds.

Impact of immune status on C_T.

To determine whether the C_T cutoff may be different in immunocompromised patients, we compared the median C_T values in adult patients staying in bone marrow transplant and oncology wards with adult patients staying in nonimmunocompromised wards. The median C_T value was not significantly different between immunocompromised and nonimmunocompromised patients with RMEIA toxin-positive (22.8 [IQR, 21.3 to 24.2] versus 22.9 [IQR, 21.7 to 24.3]; P = 0.84) and RMEIA toxin-negative (30.6 [IQR, 24.9 to 35.5] versus 30.3 [IQR, 24.8 to 33.8]; P = 0.38) stool samples (see Fig. S1 in the supplemental material).

Impact of 027 strain on C_T.

To determine whether the C_T cutoff may be impacted by the presumptive 027 strain, we compared the median C_T values in stool samples positive for presumptive 027 to those with non-027 infection. As shown in Fig. 5, the median C_T value was not significantly different between presumptive 027 and non-027 strains for RMEIA toxin-positive (22.8 [IQR, 21.8 to 25.3] versus 23.4 [IQR, 21.6 to 24.2]; P = 0.47) and RMEIA toxin-negative (29.1 [IQR, 24.8 to 32.3] versus 29.2 [IQR, 24.4 to 32.7]; P = 0.96) stool samples. Presumptive 027 C. difficile comprised 17.8% (18/101) of RMEIA toxin-positive stool samples and 10.0% (21/211) of RMEIA toxin-negative stool samples (P = 0.07).

FIG 5.

C. difficile tcdB PCR cycle threshold in PCR-positive stool samples with 027 and non-027 strains. C. Diff Quik Chek Complete RMEIA toxin-positive stool samples with 027 (n = 18) and non-027 (n = 83) C. difficile and RMEIA toxin-negative stool samples with 027 (n = 21) and non-027 (n = 190) C. difficile were included. The horizontal lines represent the median cycle thresholds.

DISCUSSION

Given the lack of a stand-alone C. difficile diagnostic that can sensitively and rapidly detect free toxins in stool, some experts recommend a multistep testing algorithm in which a nucleic acid amplification test (NAAT) or EIA GDH is performed in tandem with EIA toxin to rapidly identify toxin-positive and toxin⁻/NAAT⁺ patients to facilitate appropriate clinical decision-making (3, 4). Proponents of toxin testing recommend that only toxin-positive patients be treated for CDI and toxin⁻/NAAT⁺ patients be evaluated clinically to determine if they have CDI or are colonized with C. difficile (3, 4). In this study, we showed that by defining a C_T cutoff for GeneXpert C. difficile/Epi tcdB PCR, a sample-to-answer real-time PCR assay, we could sensitively predict 96.0% of toxin⁺/PCR⁺ stool samples with an NPV of 97.1% and with a specificity of 78.0% using RMEIA toxin and CCNA as the reference method. PPV was based on PCR-positive samples only. If we included PCR-negative samples, which made up 85.0% of total samples and are assumed to be RMEIA negative and CCNA negative (4), the NPV of C_T toxin would be 99.8% and 99.0% when using RMEIA toxin alone or RMEIA and CCNA, respectively, as the reference method. Furthermore, C_T toxin was positive in 71% (37/52) of CCNA⁺/NAAT⁺ samples that were negative by RMEIA toxin, although the clinical significance of this is unclear (7). Overall, with the approach undertaken in this study, PCR could sensitively detect presence of toxigenic C. difficile (4), sensitively predict fecal free toxin, and predict toxin-negativity with a high NPV. Compared to current practices at our institution and other U.S. hospitals where nearly all patients with positive C. difficile PCR results are treated for CDI (16, 17), reporting C_T toxin result in addition to PCR result has the potential to reduce anti-C. difficile therapy by 45.8% based on results of this study (143 of 312 PCR-positive samples were C_T toxin negative), if only toxin-positive patients are treated. This approach has the potential to have far-reaching impact as stand-alone C. difficile NAAT has been widely adopted for CDI diagnosis in the United States (18). Further studies are needed to implement C_T toxin reporting under routine clinical practice and measure its impact on provider behavior, patient outcomes, and antibiotic stewardship.

Although Leslie and colleagues showed that at a cutoff of ≥5.1 log₁₀ DNA copies/ml, they could correctly classify ≥95% of toxin-positive and 70% of toxin-negative stool samples (15), this is the first study to comprehensively investigate the analytical performance of C_T cutoffs for prediction of free fecal toxin status and to investigate the potential impact of preanalytical and analytical factors and strain type on its accuracy. We showed that stool quality (i.e., Bristol score), patient's age (i.e., pediatric versus adult), immune status (i.e., immunocompromised versus nonimmunocompromised), and strain type (i.e., presumptive 027 versus non-027) did not significantly change the median C_T values, which suggests that a single C_T cutoff could be applied across different Bristol grades (of unformed stool), different age groups, patients with different immune status, and different strains of C. difficile. We also showed that time to stool processing, although greater for false-positive samples, did not vary significantly between stool samples with true-positive and those with false-positive C_T toxin results. Similarly, anti-C. difficile therapy was not more common in the latter group. However, there was a trend toward higher non-027 strain representation in stool samples with false-positive C_T toxin results. This finding suggests that some of the negative reference method results may be due to the presence of low-toxin-producing C. difficile strain types. Other preanalytical factors that could potentially play a role but which were not investigated in this study include the patient's temperature, dietary intake, anti-toxin antibodies, and inflammation. Studies have shown that TcdA and TcdB are unstable at body temperature and are degraded by digestive enzymes (19, 20). The type of food consumed may influence the types of digestive enzymes present in stool. Importantly, we showed that transfer swabs, which are recommended by the manufacturer for simple transfer of stool to sample reagent, yield C_T toxin sensitivity and specificity that are not significantly different from those obtained using a scoop to transfer a standardized stool volume. This finding is reassuring and is consistent with a prior study which estimated the transfer swabs to hold approximately 100 μl of unformed stool (15). Lastly, we showed that intercartridge lot C_T variability is relatively low. This finding indicates the reproducibility of GeneXpert C. difficile/Epi tcdB PCR C_T is sufficiently high for using C_T at the defined C_T cutoff to determine free toxin status in patients with positive PCR results.

The application of PCR for detection of toxigenic C. difficile and prediction of free toxin result has certain advantages and disadvantages over multistep testing algorithms employing EIA and CCNA (4). First, DNA is likely more stable than TcdA and TcdB, therefore PCR is presumably less affected by preanalytical factors such as temperature, pH, and digestive enzymes that may degrade toxins (19, 20). Second, PCR can be used as a stand-alone test, obviating need for multistep testing, which saves time and resources. However, PCR may be potentially more expensive if EIA GDH is used to screen samples. Third, PCR C_T toxin bypasses potential for false-negative toxin results due to interribotype toxin divergence (21). Fourth, if a C_T cutoff is selected to maximize prediction of toxin-positive samples using EIA as the reference method, PCR would miss a fraction of samples that are EIA negative but CCNA positive. However, a three-step algorithm that includes CCNA is neither practical nor actionable. Lastly, the major disadvantage of using the C_T toxin approach is that 22% to 34% of free toxin-negative strains are misclassified as toxin positive, which may result in some overdiagnosis relative to a two-step approach incorporating a direct free fecal toxin test.

The findings from this study are promising and are consistent with prior investigations (11, 13–15). However, this study has several limitations. First, further studies are needed to validate the C_T cutoff and to confirm the analytical accuracy of C_T toxin reported here. We determined the C_T cutoff according to the Youden maximum index value, which assigns equal weight to sensitivity and specificity (22), and also by fixing the sensitivity at 99%. Alternatively, C_T cutoff may be determined by considering the costs associated with false-positive and false-negative results, as well as the prevalence of CDI (23). Second, we did not evaluate the clinical performance of the C_T toxin cutoff. Some studies have demonstrated that lower C_T is associated with CDI severity (24) and poor outcome (24, 25), although these associations were not observed in another study (26). While we could consider a clinical gold standard, that too has drawbacks, in that clinical decision making is driven so strongly by the laboratory PCR results, which each patient suspected of having CDI receives. Further studies are needed to evaluate the clinical safety of C_T toxin results and determine the clinical significance of false-positive C_T toxin results (i.e., whether false-positive C_T toxin patients have CDI or not). Third, although we showed no statistically significant difference in the median C_T values between different Bristol stool scores, age groups, immune statuses, presumptive 027 and non-027 strain type, and preanalytical factors (i.e., time to stool processing and antibiotic treatment), our study was not powered to evaluate each of these factors and thus lack of statistical significance does not exclude the possibility that there may be an effect on C_T. Furthermore, the impact of strain type beyond that of 027 has to be investigated. Fourth, we investigated the C_T toxin performance using GeneXpert C. difficile/Epi tcdB PCR; however, other real-time PCR assays are likely to be equally accurate in predicting fecal free toxin. Further studies are needed to define assay-specific C_T cutoffs and investigate the performance of C_T toxin with other PCR assays. Fifth, although sample selection bias could have influenced our results, we included only PCR-positive stool samples because GeneXpert C. difficile tcdB PCR has been shown to have a sensitivity of 96% to 100% in CCNA-positive stool samples (4). Sixth, we did not test RMEIA toxin-positive samples with CCNA because RMEIA has been shown to have a specificity of 99% to 100% in CCNA-negative stool samples (4). Therefore, the influence of sample selection bias was minimized. Lastly, in addition to RMEIA toxin, we also included CCNA and ELISA toxin in our reference method to maximize free toxin detection. It is possible that we still missed toxin-positive stool samples even with the combination of these methods. This may explain some of the false-positive C_T toxin results. Although not commercially available, inclusion of a recently described ultrasensitive toxin test in the reference standard might have detected more toxin-positive samples (27). Further studies are needed to test this hypothesis.

In summary, C. difficile tcdB PCR C_T may be used to predict free toxin results with high sensitivity and NPV, providing additional results to guide therapy. Further studies are needed to implement C_T toxin reporting and measure its impact on patient care and antibiotic stewardship.

MATERIALS AND METHODS

Ethics.

This study was approved by the Stanford University Internal Review Board. A waiver of the informed-consent requirement was obtained for the use of discarded stool samples.

Study design.

Between March and November 2016, consecutive unformed stool specimens (n = 312) collected from adult and pediatric patients and sent to the Stanford Health Care clinical microbiology laboratory for C. difficile testing with the GeneXpert C. difficile tcdB PCR assay (Cepheid, Sunnyvale, CA) were included in this study if the PCR result was positive and if there was sufficient stool quantity for further testing. PCR-positive stool samples were prospectively tested for fecal free toxins with the rapid membrane EIA (RMEIA). EIA-negative samples were further tested for free TcdB using the cell cytotoxicity neutralization assay (CCNA) and the enzyme linked immunosorbent assay (ELISA).

Clinical data.

An electronic report generated from the laboratory information system was used to obtain patient age, sex, and patient location. Chart review was performed in a subset of patients with discordant C_T toxin results to obtain antibiotic and laxative exposure in the 60 days prior to specimen collection. The stool softener docusate was not considered a laxative.

PCR.

Fresh stool samples were tested with the GeneXpert C. difficile/Epi tcdB PCR assay per the package instructions. A swab was used to transfer a nonstandardized volume of stool to the sample reagent. For the purpose of this study, fresh stools were also tested with a standardized volume of stool using a disposable scoop (Health Natura, Tuscaloosa, AL) to transfer 110 μl of stool to the sample reagent. The qualitative result, C_T for tcdB PCR, and presumptive 027/NAP1/BI or non-027 strain type result were recorded from the assay software. The intercartridge lot C_T reproducibility was measured by testing 20 randomly selected PCR-positive stool samples with four different cartridge lot numbers. Cartridge lots tested included 1000037261, 1000037262, 1000037264, and 1000037265.

RMEIA.

C. Diff Quik Chek Complete EIA (TechLab, Blacksburg, VA) was performed on fresh stool specimens per the package insert. Samples were refrigerated at 4°C until processing was performed. All but 22 samples were in compliance with the package insert and tested within 72 h of collection. Leftover aliquots of stool were stored at −80°C for testing with CCNA and ELISA as described below.

CCNA.

Stool samples frozen for 3 to 5 months were tested for TcdB using the C. Difficile Tox-B test (TechLab). A frozen 30 mg stool aliquot was thawed, transferred to 170 μl of diluent, and tested per the package insert. MRC-5 human lung tissue embryonic cells (Quidel, Santa Clara, CA) were used as indicator cells. Cells were incubated in a CO₂ incubator at 37°C for 48 h. Cytotoxic effect was considered positive if at least 50% of cells in a well were rounded in 48 h.

ELISA.

A frozen 30 mg stool aliquot was tested for TcdA and TcdB with the tgcBIOMICS ELISA kit (tgcBIOMICS, Bingen, Germany) according to the manufacturer's instruction. An assay cutoff of 0.91 ng/ml was determined by testing 20 PCR-negative stool samples (data not shown).

Statistical analysis.

The Mann Whitney U test was used to compare median tcdB C_T values. Fisher's exact test was used to analyze differences between proportions. The receiver operating characteristic (ROC) curve was used to measure C_T performance for predicting fecal free toxins. The C_T cutoff was determined using the Youden maximum index value, which assigns equal weight to sensitivity and specificity (22), and also by fixing sensitivity to ≥99%. Average coefficient of variation was used to measure intercartridge lot C_T variability. Statistical analysis was done with GraphPad Prism 5.0 software (San Diego, CA).