Related Experiment Video
Updated: Oct 9, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Machine learning-based model for predicting post-operative complications and evaluating long-term survival in elderly
Qichang Xie1, Cheng Huang1, Junpeng Zhan1
1Department of Thoracic Surgery, Fujian Medical University Union Hospital, Fuzhou, 350001, China; Key Laboratory of Cardio-Thoracic Surgery (Fujian Medical University), Fuzhou, 350001, China; Clinical Research Center for Thoracic Tumors of Fujian Province, Fuzhou, 350001, China.
Background:
Postoperative complications remain a major challenge in elderly patients undergoing thoracoscopic resection for early-stage lung adenocarcinoma. Existing risk models insufficiently capture the multidimensional interplay between pulmonary functional reserve and systemic inflammatory response.
Methods:
We conducted a retrospective cohort study including patients aged ≥70 years with stage I-II early-stage lung adenocarcinoma undergoing video-assisted thoracoscopic surgery between 2016 and 2022. Patients with chronic obstructive pulmonary disease (COPD) were excluded, and the study cohort included individuals with both PRISm and normal spirometry. Preoperative spirometry and inflammatory biomarkers obtained within 1 week before surgery were collected as candidate features for the machine learning models. PRISm was defined as preserved FEV1/FVC with reduced FEV1%pred and/or FVC%pred. The primary outcome was post-operative complications (Clavien-Dindo grade ≥2), and secondary outcomes included overall survival and disease-free survival. Seven machine learning models were developed using cross-validation and evaluated for discrimination, calibration, and clinical utility. Model interpretability was assessed using SHAP.
Results:
Of 319 patients, 18.5% developed postoperative complications (Clavien-Dindo grade ≥2). PRISm emerged as a significant predictor of both short- and long-term outcomes, being associated with postoperative complications as well as all-cause mortality and disease-free survival. Among the machine-learning algorithms evaluated, XGBoost showed the best predictive performance for postoperative complications (test AUC, 0.786) and provided consistent clinical net benefit. Feature-importance and SHAP analyses further identified PRISm, predicted FEV1, and systemic inflammatory indices as key contributors to model prediction, with nonlinear interactions between impaired pulmonary function and systemic inflammation.
Conclusions:
PRISm was associated with both postoperative complications and long-term survival in elderly patients undergoing VATS for early-stage lung adenocarcinoma. Incorporating pulmonary function and systemic inflammatory indices into machine-learning models may improve perioperative risk stratification, while PRISm may provide additional prognostic information for both short- and long-term outcomes.