Artificial Intelligence Successfully Identifies Subjects Who are at High Risk for Colorectal Cancer and Will Not Show Up for Colonoscopy within Six Months Following a Positive Fecal Occult Blood Test

Background: We aimed to identify subjects that harbor colorectal cancer and will not complete colonoscopy within six months following a positive fecal occult blood test (target population) by artificial intelligence (A.I.) based tools. Methods: We trained and validated a machine learning model based on a dataset of the Clalit Health Services. The entire population included 25,219 subjects aged 50 to 74 years with a positive FOBT result who participated in the screening program between 2011 and 2013. The target population included 202 patients (0.8% of the total cohort; 26.5% of all cancer cases). Multiple socioeconomic, administrative, and laboratory data were collected. The portion of the total population singled out by the model was termed as subjects needed to engage (SNE).   Findings: Using two threshold levels the model reduced the number of SNE to 3.85% (of the entire validation cohort), identifying 25.8% of the target population [Positve Predictive Value Value (PPV) 5.1%, 95% Confidence Intrval (CI) 4.4-5.8; Negative Predictive Value (NPV) 99.4%, 95%CI 99.4-99.4; Area Under Receiver Operative Curve (AUROC) 0.74 95%CI 0.73-0.76] or reduced the number of SNE to 15.9%, identifying 55% of the target population (PPV 2.7%, 95%CI 2.4-3.0; NPV 99.5%, 95%CI 99.5-99.6; AUROC 0.74 95%CI 0.73-0.76). Interpretation: An AI-derived tool can successfully identify subjects with a positive FOBT that are at high risk for both harboring CRC and not completing colonoscopy within six months. Implementing this approach may improve FOBT-based CRC screening programs from the first day of a positive FOBT. Funding: None to declare. Declaration of Interest: None to declare. Ethical Approval: The CHS ethical committee authorized the study.