Machine learning to predict early recurrence after oesophageal cancer surgery

Area of Science:

• Oncology
• Surgical Oncology
• Machine Learning in Medicine

Background:

• Early recurrence of esophageal adenocarcinoma after surgery (oesophagectomy) affects 20-30% of patients despite neoadjuvant treatment.
• Existing models for predicting this risk have poor performance.
• Accurate quantification of early recurrence risk is challenging.

Purpose of the Study:

• To develop a predictive model for early recurrence of esophageal adenocarcinoma post-oesophagectomy.
• To utilize a large multinational cohort and advanced machine learning techniques.
• To improve risk stratification for patients undergoing oesophagectomy.

Main Methods:

• Analysis of consecutive patients who underwent oesophagectomy for adenocarcinoma with neoadjuvant treatment.
• Development of predictive models using elastic net regression (ELR), random forest (RF), and extreme gradient boosting (XGB).
• Generation of a combined (ensemble) model incorporating ELR, RF, and XGB.

Main Results:

• A total of 812 patients were included, with a 29.1% recurrence rate within one year.
• All models showed good discrimination, with the ensemble model achieving the highest area under the receiver operating characteristic curve (AUC) of 0.805.
• Key predictors identified were the number of positive lymph nodes (25.7%) and lymphovascular invasion (16.9%).

Conclusions:

• Machine learning models, particularly the ensemble approach, demonstrate excellent performance in predicting early recurrence after oesophagectomy.
• The developed model provides a valuable tool for prognostication for clinicians and patients.
• The model's reliance on factors like lymph node status and lymphovascular invasion offers clinically relevant insights.