Journal: Nature medicine
This study developed and implemented an AI-based risk prediction model for patients undergoing elective colorectal cancer surgery, using real-world data from over 18,000 patients.
The model predicts 1-year mortality risk and supports personalized perioperative treatment pathways tailored to individual risk profiles.
Validation showed good predictive accuracy with an AUC of 0.79.
In a before-and-after cohort study, personalized treatment significantly reduced complications compared to standard care:
- Major complications: 19.1% vs. 28.0% (adjusted OR 0.63, P=0.02)
- Any medical complications: 23.7% vs. 37.3% (OR 0.53, P<0.001)
Economic modeling suggested this approach is cost-effective.
This scalable, registry-based AI decision-support tool demonstrates potential to improve surgical outcomes and optimize perioperative care in colorectal cancer patients.