Machine learning models predict survival in unresectable stage III non-small cell lung cancer: Surveillance,...

Patients with unresectable stage III non-small cell lung cancer (NSCLC) have heterogeneous survival outcomes, making accurate prognostic prediction challenging. This study aimed to develop and validate machine learning (ML) models for predicting overall survival (OS) after diagnosis in this patient population.

Data from 11,675 patients retrieved from the Surveillance, Epidemiology, and End Results (SEER) database were used for model development and internal testing. An independent external cohort (n=162) from the Affiliated Hospital of Jiaxing University was used for validation. We constructed models to predict 6-month, 1-year and 2-year OS using five ML algorithms, with model performance evaluated via the area under the receiver operating characteristic curve (AUC), accuracy and calibration. The optimal model was interpreted using SHapley Additive exPlanations (SHAP).

The XGBoost model achieved the best performance for 6-month OS prediction in the test set (AUC =0.742, accuracy =0.708, sensitivity =0.746, specificity =0.637). It also showed predictive efficacy for 1-year (AUC =0.696) and 2-year (AUC =0.684) OS, with the highest discriminative ability at the 6-month time point. The model had good calibration and favorable net clinical benefit, while its performance decreased in external validation (AUC =0.647, accuracy =0.586). SHAP analysis revealed chemotherapy and radiotherapy as the most important predictive factors.

By integrating multiple clinical and treatment variables, the XGBoost model provides supplementary prognostic information and complements the conventional tumor-node-metastasis (TNM) staging system for 6-month, 1-year and 2-year OS prediction. This data-driven framework can assist individual risk stratification and improve the anatomical assessment of the TNM system.