Beyond predictive performance: A systematic review and critical methodological appraisal of AI/ML and conventional modelling strategies in breast, colorectal, and pancreatic Cancer.
Nurdiana Zainol Abidin, Noorsuzana Mohd Shariff, Eva Nabiha Zamri et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review evaluated the methodological performance of AI/ML models versus conventional logistic regression and Cox proportional hazards models in predicting outcomes for breast, colorectal, and pancreatic cancer. Analyzing 65 studies with 907,567 participants, the review found that AI/ML methods showed superiority in specific tasks, particularly in imaging, while conventional models remained effective in structured datasets. Notably, in pancreatic cancer, CoxPH outperformed XGBoost with a C-index of 0.802 versus 0.723.
Key findings
- AI/ML models excelled in imaging-derived feature extraction and nonlinear perioperative prediction.
- CoxPH showed superior performance to XGBoost in pancreatic cancer risk assessment (C-index 0.802 vs 0.723).
- 83% of studies had moderate risk of bias due to limited external validation and sparse calibration reporting.
- Tree-based ensembles consistently outperformed logistic regression for predicting perioperative complications.
Why it matters
Understanding the comparative effectiveness of AI/ML versus conventional models is crucial for clinicians aiming to implement precision oncology strategies. This review highlights the contexts in which AI/ML can enhance predictive accuracy, guiding future research and clinical applications.