Predicting pathologic complete response in locally advanced rectal cancer patients after neoadjuvant therapy: a...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning in Medicine

Background:

• The watch and wait strategy is a safe alternative for locally advanced rectal cancer (LARC) patients achieving pathological complete response (pCR) post-neoadjuvant therapy (NAT).
• Current restaging methods lack the precision needed for optimal clinical decision-making in these patients.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting pCR in LARC patients before treatment.
• To provide a more accurate tool for identifying candidates for the watch and wait strategy.

Main Methods:

• An extreme gradient boosting ML model was developed using data from LARC patients who underwent NAT.
• Patient data was split into training and tuning sets (7:3 ratio).
• Feature importance was assessed using SHapley Additive exPlanations (SHAP) values, and the ML model was compared against a conventional nomogram.

Main Results:

• The ML model demonstrated superior performance, significantly improving the area under the receiver operating characteristic curve (0.95 vs. 0.72), sensitivity (82.2% vs. 43%), and specificity (91.6% vs. 87.1%) compared to the nomogram.
• Key predictors for pCR included neoadjuvant radiotherapy, preoperative levels of carbohydrate antigen 125 (CA125), CA199, carcinoembryonic antigen, and tumor invasion depth.

Conclusions:

• The developed ML model shows potential as an effective alternative to existing assessment tools for LARC.
• This model can aid clinicians in tailoring treatment strategies, facilitating the watch and wait approach for pCR patients and ensuring timely surgery for others.