Su1352 Detection of Clostridium difficile-Infected Stool by Electronic-Nose Analysis of Fecal Headspace Volatile Organic Compounds

Abstract

Introduction Electronic-noses (e-noses) have the ability of converting smells to digital signatures. By interacting with volatile organic compounds (VOC's) emitted by metabolic end-products of disease, e-noses create distinct patterns of aggregate VOC's to generate unique signatures that can be used to identify disease. The detection of Clostridium difficile (C. diff) by VOC's is prototypic as it has been successfully identified by scent using canines. We seek to pilot a novel portable e-nose device to profile fecal headspace to diagnose C. diff infection. Methods We are prospectively enrolling patients undergoing stool testing by C. diff PCR for infectious diarrhea for fecal headspace analysis using the Aetholab e-nose (eNose Company, Zutphen, NL). Stool specimens were obtained from the clinical microbiology lab of our institution within 7 days of ambient collection and aliquots of 10-ml of homogenized stool were transferred to disposable analysis bottles. The e-nose circulates fecal headspace gas within each analysis bottle to allow interaction with its respective sensor array. Each sensor array comprises of three metal-oxide sensors that undergo reversible redox reactions with VOC's at an electrochemical interface. Electrical resistance changes are measured by applying a 32-point thermal cycle to the sensor array generating a digital signature (Figure 1). This process has been shown to interact with a wide range of VOC's. Combined data from the 3 sensors in 7 combinations (A, B, C, AB, AC, AC, ABC) is permuted across 3 scaling factors generating a total of 21 data sets per analysis. These are then compressed through a multi-way analysis to reduce the large amount of data to avoid spurious associations. Finally each sample's compressed dataset is introduced into an artificial neural network (ANN) using a supervised approach. The 5 best models generated using a leave-one-out approach are used to obtain the reported results. Results Separation was observed using all 20 C. diff-PCR positive and 53 C. diff-PCR negative stools. The ROC curve for this model is shown in Figure 2 with an AUC = 0.85. This demonstrated 80% sensitivity, 85% specificity with 84% accuracy. The positive predictive value was 0.67 and the negative predictive value 0.92. Matthews correlation coefficient was 0.64. Conclusion Through the use of a pattern-recognition e-nose device it is possible to distinguish C. diffinfected stools. This is the first time that this type of device has been used for this diagnosis, and also the first time this e-nose has been applied for fecal headspace analysis of infectious disease. This indicates that C. diff testing can translate into a point-of-care evaluation.