Machine learning insights into early mortality risks for small cell lung cancer patients post-chemotherapy

Introduction Small cell lung cancer (SCLC) is a highly aggressive form of lung cancer, and chemotherapy remains a cornerstone of its management. However, the treatment is associated with significant risks, including heightened toxicity and early mortality. This study aimed to quantify the 90-day mortality rate post-chemotherapy in SCLC patients, identify associated features, and develop a predictive machine learning model. Methods This study utilized data from the Surveillance, Epidemiology, and End Results (SEER) database (2000–2018) to identify prognostic features influencing early mortality in SCLC patients. Prognostic features were selected through univariate logistic regression and Lasso analyses. Predictive modeling was performed using advanced machine learning algorithms, including XGBoost, Multilayer Perceptron, K-Nearest Neighbor, and Random Forest. Additionally, traditional models, such as logistic regression and AJCC staging, were employed for comparison. Model performance was evaluated using key metrics, including the Area Under the Receiver Operating Characteristic Curve (AUC), calibration plots, the Kolmogorov–Smirnov (KS) statistic, and Decision Curve Analysis (DCA). Results Analysis of 12,500 eligible patients revealed 10 clinical features significantly impacting outcomes. The XGBoost model demonstrated superior discriminatory capability, achieving AUC scores of 0.95 in the training set and 0.78 in the validation set. It outperformed comparative models across all datasets, as evidenced by its AUC, KS score, calibration, and DCA results. Additionally, the model was integrated into a web-based platform to improve accessibility. Conclusion This study introduces a machine learning model alongside a web-based support system as critical resources for healthcare professionals, facilitating personalized clinical decision-making and enhancing treatment strategies for SCLC patients post-chemotherapy.