Development and Validation of an Artificial Intelligence–Driven Predictive Triage System for Colonoscopy Referral in Australian Primary Care

Colorectal cancer (CRC) is Australia’s second leading cause of cancer-related deaths, and cases in people under 50 are on the rise. Each year, nearly one million colonoscopies are performed, yet referral practices for patients with symptoms remain inconsistent. Current triage methods—based on symptoms, faecal immunochemical tests (FIT), and clinical judgement—are often inaccurate. This leads to low-risk patients being over-referred, high-risk patients being missed, long waiting lists, and inefficient use of limited endoscopy resources. These delays can result in late-stage diagnoses, poorer outcomes, and inequities in care.
Internationally, predictive models that combine clinical and laboratory data have improved CRC risk assessment, but few have been tested in Australian primary care, where patient populations and health systems differ. Artificial intelligence (AI) and machine learning offer an opportunity to integrate demographic, clinical, pathology, FIT, and referral text data to provide personalised, explainable risk estimates. Such tools could help prioritise colonoscopy for those who need it most, reduce delays, and improve patient experience.
This project aims to develop and validate an AI-driven triage tool for colonoscopy referrals in Australian primary care. Using retrospective general practice referral data, we will compare multiple machine learning models to identify predictors of significant findings at colonoscopy. The best-performing model will underpin a transparent, clinician-support tool for future prospective testing. Co-designed with clinicians and patients, this tool seeks to improve equity, efficiency, and early detection, while informing national referral guidelines.