Abstract
OBJECTIVE: This study aimed to develop models for predicting prolonged postoperative length of stay (PPOLOS) in lung cancer patients undergoing video-assisted thoracoscopic surgery (VATS) by utilizing machine-learning techniques. These models aim to offer valuable insights for clinical decision-making. METHODS: This retrospective cohort study analyzed a dataset of lung cancer patients who underwent VATS, identifying 25 numerical features and 45 textual features. Three classification machine-learning models were developed: XGBoost, random forest, and neural network. The performance of these models was evaluated based on accuracy (ACC) and area under the receiver operating characteristic curve, whereas the importance of variables was assessed using the feature importance parameter from the random forest model. RESULTS: Of the 6767 lung cancer patients, 1481 patients (21.9%) experienced a postoperative length of stay of > 4 days. The majority were male (4111, 60.8%), married (6246, 92.3%), and diagnosed with adenocarcinoma (4145, 61.3%). The Random Forest classifier exhibited superior prediction performance with an area under the curve (AUC) of 0.792 and ACC of 0.804. The calibration plot revealed that all three classifiers were in close alignment with the ideal calibration line, indicating high calibration reliability. The five most critical features identified were the following: surgical duration (0.116), age (0.066), creatinine (0.062), hemoglobin (0.058), and total protein (0.054). CONCLUSIONS: This study developed and evaluated three machine-learning models for predicting PPOLOS in lung cancer patients undergoing VATS. The findings revealed that the Random Forest model is most accurately predicting the PPOLOS. Findings of this study enable the identification of crucial determinants and the formulation of targeted interventions to shorten the length of stay among lung cancer patients after VATS, which contribute to optimize the allocation of healthcare resources.