Detection of Colorectal Cancer in Patients With a Positive Fecal Immunochemical Test Using an Electronic Nose Device (AeoNoseTM)

Study rationale:

Colonoscopy plays a key role in all colorectal cancer (CRC) screening modalities. An important limitation of colonoscopy remains that the diagnostic accuracy and therapeutic safety of colonoscopy both depend on several physician and patient-related factors. Significant amounts of polyps are missed during colonoscopy and severe untoward effects still occur in 0.1%-0.3% of all colonoscopy procedures, even among expert examiners. Moreover, the burdensome nature of the procedure and the necessity of full bowel preparation negatively influences participation rates and the use of colonoscopy as a CRC screening test is limited by healthcare costs and the expertise needed to perform the screening. These disadvantages of colonoscopy warrant the development of a new, non-invasive, accurate CRC screening test.

An electronic nose device is an artificial olfactory system that analyses volate organic compounds (VOCs) in exhaled breath. The use of electronic noses has already been assessed as a potential non-invasive diagnostic biomarker test for lung cancer, breast cancer and malignant melanomas. The present study aims to investigate the diagnostic accuracy of exhaled breath analysis with the AeonoseTM (The eNose Company, Zutphen, the Netherlands) to distinguish the breath of patients suspected for CRC (based on a positive fecal immunochemical test (FIT)), who are truly diagnosed with CRC, from patients suspected for CRC in whom this diagnosis is rejected after colonoscopy.

Objectives:

• To investigate the diagnostic accuracy of exhaled breath analysis with the AeonoseTM to distinguish the breath of patients suspected for CRC based on a positive FIT test, who are truly diagnosed with CRC, from patients suspected for CRC in whom this diagnosis is rejected after colonoscopy.
• To investigate the diagnostic accuracy of exhaled breath analysis with the AeonoseTM for other (pre-)malignant colon lesions (e.g. advanced adenomas, non-advanced adenomas, and sessile serrated lesions).
• To assess the acceptance rate of using the AeonoseTM device.

Study design: The present multicenter study will include 1950 FIT-positive patients. The study will be conducted in seven hospitals in the Netherlands. First a database of breath prints will be developed to detect CRC. In this study phase the AeonoseTM will be trained and the database of breath prints will be verified using "leave 10% out" cross validation. After the calibration phase the diagnostic accuracy of AeonoseTM will be assessed in new study patients (external validation).

Study population: Patients between the ages of 55 and 75 years old with a positive FIT-test referred for a colonoscopy procedure.

Main study parameters/endpoints:

• The ability of AeonoseTM to distinguish between patients with and without CRC in a FIT-positive population in terms of sensitivity and specificity.
• The diagnostic accuracy of AeonoseTM for other (pre-)malignant colon lesions.
• Acceptance rate of AeonoseTM

Statistical analysis:

Data on demographic and baseline characteristics will be summarized for continuous variables, in case of normal distribution by mean and standard deviation, and in case of non-normal distribution by median and interquartile range. For discrete variables (eg, race and sex) data will be summarized by proportions (percentages). Differences in baseline characteristics will be determined using independent sample t test, Fisher's exact test or Chi-square test, when appropriate. A 2-sided p-value <0.05 will be considered statistically significant.

The data will be analyzed by an artificial neural network (ANN) to identify data classifiers to extract differences between the patterns in patients by the presence or absence of CRC in biopsies. By analyzing a group of breath prints from patients with and without CRC, individual differences of diet, medications, co-morbidities, and other factors can be filtered out to identify features that distinguish the compounds of patients with CRC. This process continues until a selected level of discrimination is met. Data compression and ANN have been integrated in a proprietary software package (Aethena) of the eNose Company (Zutphen, the Netherlands).

The binary results will be presented in a scatterplot and a receiver operating characteristics curve (ROC-curve). Matthews correlation coefficients will be calculated to measure the quality of binary classifications. Cross-validation of the data will be performed using a leave-10%-out method. A portion of the data is left out to predict the most optimal model.