Abstract / Summary
INTRODUCTION: Colorectal cancer (CRC) is the third most diagnosed cancer worldwide. Only invasive screening methods have been shown to be effective in reducing CRC incidence through the detection of advanced adenomas (AA). This work aimed to develop an Artificial Intelligence (AI)-based blood test for CRC screening and evaluate its diagnostic accuracy. METHODS: The test is based on the detection of a protein and miRNA signature and on an AI-derived predictive algorithm. The training/test and validation of the model were conducted analyzing plasma samples from 3,569 participants recruited in two independent prospective clinical studies: a case-control cohort and a screening cohort. Colonoscopy was used as reference. The specificity, sensitivity, predictive value, and likelihood ratio of the test were validated in 2,856 patients from the screening cohort. RESULTS: The test showed 83.3% sensitivity for CRC (95% CI, 58.6 to 96.4) and 65.6% for AA (95% CI, 57.4 to 73.31) at 87.2% specificity (95% CI, 84.9 to 90.1) for no colonoscopy findings, non-neoplastic findings and non-advanced precancerous lesions. The sensitivity to detect CRC and AA located in the proximal colon was 80.0% (95% CI, 28.35 to 99) and 64.5% (95% CI, 52.7 to 75.1), respectively. DISCUSSION: In an average-risk screening population, the test appears to be highly sensitive for detecting CRC and its precursor lesions such as AA and has received Breakthrough Device Designation from the US Food and Drug Administration. These exploratory results will be validated in a pivotal clinical study conducted in a larger US screening cohort. ClinicalTrial.gov registry number: NCT06738511.