Comparative study on predicting postoperative distant metastasis of lung cancer based on machine learning models

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Lung cancer is a leading cause of cancer mortality globally.
• Postoperative distant metastasis significantly impacts patient prognosis and survival.
• Timely prediction of metastatic potential is crucial for effective treatment strategies.

Purpose of the Study:

• To compare the predictive performance of nine machine learning (ML) models for postoperative lung cancer metastasis.
• To enhance model interpretability using SHAP (Shapley Additive Explanations).
• To develop a transparent risk stratification tool for postoperative lung cancer management.

Main Methods:

• Retrospective analysis of clinical data from 3,120 patients with stage I-III lung cancer.
• Development and evaluation of nine ML models including XGBoost, RF, LightGBM, AdaBoost, DT, GBDT, GNB, CNB, and MLP.
• Performance assessment using accuracy, precision, recall, F1 score, ROC-AUC, PR-AUC, calibration, and decision curve analysis (DCA) with nested cross-validation.

Main Results:

• Gradient Boosting Decision Tree (GBDT) achieved the highest predictive performance with an AUC of 0.810.
• Key predictors identified by SHAP analysis include adjuvant chemotherapy, adjuvant radiotherapy, pathological N stage, age, BMI, and preoperative neutrophil count.
• The developed models showed potential for clinical integration to aid real-time decision-making.

Conclusions:

• GBDT demonstrated superior predictive capability for postoperative lung cancer distant metastasis.
• SHAP analysis provided valuable insights into influential predictors, supporting personalized treatment strategies.
• The study offers a validated framework for precision management, warranting external validation before clinical deployment.