Developing and external validating a prediction model using machine learning and logistic regression: informing the...

Area of Science:

• Colorectal Surgery
• Surgical Oncology
• Medical Imaging

Background:

• Preoperative prediction of surgical difficulty in robotic-assisted total mesorectal excision for rectal cancer is challenging.
• Pelvic anatomical parameters from MRI are underutilized in robotic surgery prediction.
• Developing accurate predictive models is essential for optimizing patient outcomes.

Purpose of the Study:

• To develop and validate a predictive model for adverse surgical outcomes in rectal cancer surgery.
• To integrate machine learning and logistic regression with preoperative MRI pelvimetry and clinical data.
• To identify key predictors of surgical complexity in robotic and laparoscopic rectal cancer resections.

Main Methods:

• Retrospective multi-center study of 1,367 patients undergoing radical resection for mid-to-low rectal cancer.
• Analysis of eleven MRI-based pelvic parameters and clinical characteristics.
• Development and comparison of logistic regression and machine learning models (Random Forest, XGBoost, LightGBM).

Main Results:

• Logistic regression outperformed machine learning models, with an AUC of 0.857 (training), 0.820 (internal), and 0.810 (external validation).
• Key predictors identified include BMI > 25 kg/m², neoadjuvant chemoradiotherapy, tumor distance < 5 cm, laparoscopic approach, interspinous distance < 9.94 cm, and intertuberous diameter < 11.98 cm.
• The developed nomogram demonstrated excellent discrimination and clinical utility.

Conclusions:

• A robust prediction model integrating MRI pelvimetry and clinical factors was successfully developed and validated.
• The model aids in identifying patients at high risk for adverse outcomes in rectal cancer surgery.
• The findings support using this model to guide surgical approach selection, potentially favoring robotic surgery in complex cases.