Queen Elizabeth Hospital Birmingham (QEHB), part of University Hospitals Birmingham (UHB) NHS Foundation Trust is a tertiary referral National Health Service teaching hospital in Birmingham, UK that provides clinical services to over one million patients every year.
C. difficile Testing
In line with national guidance, an algorithmic approach to identifying CDI is undertaken at QEHB [4, 20, 21]. A three-stage algorithm is employed. Briefly, any patient with ≥1 episode of unexplained diarrhoea had their faecal specimen tested for CDI. The CDI testing algorithm consists of an initial screening step using a Premier GDH EIA (Meridian Bioscience, Cincinnati, Ohio), followed by a NAAT (Cepheid, Xpert™ C. difficile, US) for GDH positive samples only. The premier GDH involved undertaking an enzyme immunoassay looking for the presence of GDH as previously descrbed [4, 21]. The Cepheid NAAT is a real-time PCR assay targeting the toxin genes (B toxin gene and tcdB) [12, 14]. The Cepheid PCR targets the toxin B gene (tcdB) as it is independently capable of causing CDI [12, 14]. The cycle threshold (CT) value of the Cepheid real time PCR describes the number of cycles needed until DNA amplification occurs exponentially in a real-time PCR assay and they are correlated with the amount of target sequence in the sample . All samples which were GDH and NAAT positive have a Premier Toxins A and B EIA (Meridian Bioscience, Cincinnati, Ohio) which is undertaken once a week [20, 22]. The Premier Toxins A and B EIA was undertaken as previously described, with a sensitivity of 94.7% and specificity 97.3% quoted by the manufacturers when compared to the reference cytotoxin method [4, 21].
Study design and definitions
All diarrhoeal samples (Bristol stool type 5–7)  from patients between Jan 2012 to Dec 2016 at QEHB positive by GDH and NAAT were included in the study which equated to 1346 patients.
At QEHB a CDI episode was defined as the presence of a positive test result for toxigenic C. difficile by GDH and NAAT and the presence of unexplained diarrhoea (≥1 episode of unexplained diarrhoea). Severity of CDI was based on the Public Health England (PHE) toolkit for management of C. difficile . A severe C. difficile case was defined as: WCC >15 × 10  g/dL; acutely rising blood creatinine (e.g. >50% increase above baseline); temperature > 38.5 °C; or evidence of severe colitis (abdominal signs, radiology) .
Recurrent CDI and treatment failure
Recurrent CDI was defined as the return of diarrhoea (≥1 episode of unexplained diarrhoea) within 30 days of a previous CDI episode and the presence of a positive test result for toxigenic C. difficile by GDH and NAAT . Treatment failure was defined as cases where failure to respond to treatment resulted in a change of management of the patient .
Clinical data collection
Patient data collected at the time of a positive result included: patient demographics (age, sex), markers for CDI severity (white cell count, C-reactive protein, serum creatinine, serum albumin, temperature, stool frequency) and mortality (one month and 3-month all-cause mortality). Clinical severity data were obtained from 80 patients with the lowest and 138 patients with the highest CT values; mortality data was collected from all 1346 patients in the study.
Cepheid NAAT CT values were compared to Premier Toxins A and B EIA positivity and patient mortality (one and three month). All analyses including receiver operating characteristic curve (ROC), Youden’s index, error rates and univariate logistic regression models were performed using the hmeasure, plotROC and base packages in the statistical programming language R [24,25,26].
The performance of NAAT CT value as a classifier of toxin EIA positivity was assessed using a ROC curve and the area under the ROC curve (AUC). The sensitivity (TPF; True Positive Fraction), specificity, False Positive Fraction (FPF), Positive Predictive Value (PPV) and negative predictive value (NPV) of the prediction rule of low CT values corresponding to EIA positivity were calculated using the hmeasure and plotROC packages in R [24, 25]. To determine a NAAT CT cut off value, two commonly-used strategies were used in the study picking a threshold that minimises the error rate and picking one that maximises Youden’s index [27, 28].
Linear regression mortality
Logistic regression models were used to assess NAAT CT values as an explanatory variable for mortality and EIA toxin positivity. These models were fitted using R .