Related Experiment Video
Updated: Jun 22, 2026
08:50
Predictive Immune Modeling of Solid Tumors
Published on: February 25, 2020
Predicting recurrence within 5 years in Early-Stage lung adenocarcinoma with micropapillary and solid patterns
Zhongjie Wang1, Jie Chen1, Yuanyuan Xu2
1Department of Thoracic Surgery, Fujian Medical University Union Hospital, Fuzhou, Fujian Province, China; Key Laboratory of Cardio-Thoracic Surgery (Fujian Medical University), Fuzhou, China.
International Journal of Medical Informatics
|June 12, 2026
Summary
Machine learning models can predict 5-year recurrence in early-stage lung adenocarcinoma with high-risk patterns. The best model identified surgical procedure and CTR as key predictors, aiding clinical decisions.
Area of Science:
- Oncology
- Machine Learning in Medicine
- Pathology
Background:
- Micropapillary (MP) and solid (S) patterns in invasive adenocarcinoma (IAC) indicate high recurrence risk, even in early stages.
- Patients with early-stage IAC (≤3 cm, pN0M0) and MP/S components face significant postoperative recurrence risk.
Purpose of the Study:
- To develop and validate machine learning (ML) models for predicting 5-year recurrence in specific early-stage IAC patients.
- To identify key predictors of recurrence using ML and statistical analyses.
Main Methods:
- Retrospective analysis of 974 patients from two centers.
- Selection of independent predictive variables via logistic regression.
- Evaluation of eight ML models using AUC, calibration curves, and decision curve analysis (DCA).
- Interpretation of variable importance using SHapley Additive exPlanations (SHAP).
Main Results:
- Five independent predictors identified: surgical procedure, consolidation-to-tumor ratio (CTR), visceral pleural invasion, EGFR mutation status, and smoking history.
- The Neural Network model achieved the best performance (AUC 0.794 training, 0.764 test, 0.775 validation).
- SHAP analysis highlighted CTR and surgical procedure as the most influential predictors of recurrence.
Conclusions:
- A validated ML model integrating clinicopathological and imaging features predicts 5-year recurrence risk in pN0M0 IAC patients (≤3 cm) with MP/S components.
- The model demonstrates good predictive accuracy and clinical utility for assessing recurrence risk.
