Development of an Artificial Intelligence Web Application for Predicting Chemotherapy-Induced Neutropenia in Patients With Non-Small Cell Lung Cancer: A Prospective Study.

objectiveThis study aimed to develop an artificial intelligence (AI) web application for predicting chemotherapy-induced neutropenia (CIN) in patients with non-small cell lung cancer (NSCLC).

methodsWe conducted a prospective study including 310 patients who underwent 1047 chemotherapy cycles with NSCLC patients between 2019 and 2024 from Tianjin Medical University General Hospital. Detailed clinical information and laboratory data were collected. The dataset was randomly split into a training set (80%) and a test set (20%) at the patient level. Machine learning was employed to develop the model. After completing training and hyperparameter optimization (HPO) through cross-validation on the training set, the performance was compared through the test set. Meanwhile, a stacking ensemble model was developed by integrating these optimized base learners described above. Standard evaluation metrics, such as area under the receiver operating curve (AUC), Accuracy, Recall, Precision, and F1 score, and Brier score, were used for discrimination of the model. To enhance the transparency of the optimal model, the shapley additive explanations (SHAP) were used together with the Local interpretable model-agnostic explanations (LIME) and the partial dependence plot (PDP) techniques. Based on the best machine learning-based model, an AI application was developed on the Internet.

resultsThe categorical boosting (CatBoost) model with an AUC of 0.849 (95% Cl: 0.788-0.905), an accuracy of 0.813, a recall of 0.814, a specificity of 0.813, a F1 score of 0.642, and a Brier score of 0.141 had a higher discriminatory capability than other models. The five most significant features in the model were identified as body surface area, lymphocyte count, body mass index (BMI), absolute neutrophil count, and chemotherapy regimen. Notably, body surface area emerged as the most influential factor for predicting outcomes. Specifically, higher body surface area level was associated with an increased risk of CIN. The chemotherapy regimen of Paclitaxel + Carboplatin and Paclitaxel + Cisplatin showed a positive correlation with the risk of CIN. Lower lymphocyte count, BMI, and absolute neutrophil count were indicative of a higher risk of CIN. The AI application has been deployed online at http://39.96.172.15/, based on the CatBoost model.

conclusionsThe CatBoost model exhibited strong discriminatory ability in predicting CIN risk in NSCLC patients. The developed AI model serves as a valuable tool to enhance clinical decision-making.