Keywords: AI, colorectal cancer, early detection, Empowerment of family doctors

Background:

Colorectal cancer (CRC) is a leading cause of cancer-related deaths worldwide. Despite the proven benefits of screening, participation remains low, especially among asymptomatic people. New tools are needed to help identify at-risk individuals proactively. Artificial intelligence (AI) provides promising opportunities for personalized screening strategies in primary care.

Research questions:

Can an AI-driven model using electronic health records (EHRs) enhance early detection of CRC by prioritizing patients for colonoscopy referrals within a primary care–based national health system?

Method:

We developed the LAPHA model using de-identified EHR data from Leumit Health Services in Israel. The model analyzes demographic data, diagnoses, and longitudinal laboratory test trajectories to estimate CRC risk within 2.5 years. High-risk individuals are proactively referred to colonoscopy through a nurse-led navigation center. We evaluated CRC detection and polyp yield among individuals undergoing colonoscopy, comparing these rates to the age-matched national incidence.

Results:

By July 2025, 685 patients were identified as high risk and contacted. Of these, 322 underwent colonoscopy, revealing 25 cases of CRC (7.8%). In comparison, the CRC incidence in the general age-matched population was 0.08%, resulting in an odds ratio of 108 (95% CI: 40–366). Among the 137 colonoscopy reports retrieved, polyps were found in 76% of cases, and polypectomy was performed in 44% of these cases. These are promising preliminary results from an ongoing implementation.

Conclusions:

The LAPHA model significantly boosted the detection rate of CRC and the removal of polyps in a real-world primary care setting. These findings support the incorporation of AI tools into family medicine workflows, providing clinicians with data-driven decision support for preventive care. The approach is scalable, appears resource-efficient, and aligns with the evolving role of primary care in population health management.

Points for discussion:

Proposals for other outcome measures to evaluate the model

Obstacles of the model and its implementation