Predicting Treatment Response After Total Neoadjuvant Therapy for Locally Advanced Rectal Cancer.

Area of Science:

• Oncology
• Clinical Decision Support
• Artificial Intelligence in Medicine

Background:

• Selecting patients for watch-and-wait (W/W) after total neoadjuvant therapy (TNT) for locally advanced rectal cancer is complex.
• Accurate prediction of pathological complete response (pCR) is crucial for optimizing W/W selection.

Purpose of the Study:

• To develop and validate a predictive model for pCR following TNT.
• To inform patient selection for the W/W strategy in rectal cancer management.

Main Methods:

• An ensemble of tabular foundation models was fine-tuned to predict pCR using pre-treatment, post-treatment, and pre-surgery variables.
• The model was externally validated on patients undergoing TNT with either total mesorectal excision (TNT+TME) or W/W (TNT+W/W).
• Performance was assessed using AUROC, AUPRC, and Brier scores.

Main Results:

• The model predicted pCR in the TNT+TME cohort with an AUROC of 0.71 and AUPRC of 0.44.
• External validation in the TNT+W/W cohort showed an AUROC of 0.69 for predicting persistent clinical complete response (pcCR).
• The model demonstrated good calibration and improved performance after recalibration.

Conclusions:

• A novel predictive model shows good discrimination and calibration for pCR after TNT+TME.
• The model has utility in predicting pcCR for patients managed with TNT+W/W, supporting its use in W/W patient selection.
• This tool can aid clinicians in making informed decisions regarding W/W strategies for rectal cancer.