Machine learning to predict lymph node metastasis in T1 esophageal squamous cell carcinoma: a multicenter study

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning in Medicine

Background:

• Accurate quantification of lymph node metastases (LNM) risk is crucial for T1 esophageal squamous cell carcinoma (ESCC) staging and treatment.
• Existing models demonstrate limitations in precisely assessing LNM risk in early-stage ESCC.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting LNM in patients with T1 ESCC.
• To compare the performance of ML models against established guidelines and traditional statistical methods.

Main Methods:

• A multicenter, population-based study was conducted.
• Several ML algorithms were developed, including Elastic Net Regression (ELR), Random Forest (RF), and Extreme Gradient Boosting (XGB), along with an ensemble model.
• Model performance was assessed using externally validated Area Under the Curve (AUC).

Main Results:

• All developed ML models demonstrated strong discriminating power for LNM prediction.
• Elastic Net Regression (ELR) achieved the highest externally validated AUC of 0.803.
• Key predictive features identified were lymphatic invasion, vascular invasion, and depth of tumor invasion.
• ELR performance significantly surpassed the NCCN guidelines (AUC=0.576) and a standard logistic model (AUC=0.670).

Conclusions:

• Machine learning approaches offer a highly effective strategy for estimating LNM risk in T1 ESCC.
• The developed ML models provide a superior tool for risk stratification compared to current clinical guidelines.