부산대학교 도서관

Neutropenia is one of the common side effects of chemotherapy in cancer patients. Chemotherapy-induced neutropenia can lead to complications, dose reduction or treatment delay. Namely, early diagnosis and management of neutropenia are needed to maximize the treatment. In the medical field, machine learning methods have been utilized to predict diseases. In this study, we predicted neutropenia 48 hours in advance of adult cancer patients who were prescribed cytotoxic drugs. Data of 5,308 breast cancer patients and 5,409 lung cancer patients were obtained from the National Cancer Center clinical data warehouse (CDW). Patient data from the two cancers were separately preprocessed and presented as time-series datasets of clinical events. Two neural network models were employed for prediction: Bi-LSTM and RTAIN. Bi-LSTM showed the best performance with the area under the receiver operating characteristic curve (AUROC) of 0.902 and 0.788 in breast and lung cancer patients, respectively. To identify important features and time points for predicting neutropenia, we employed RTAIN. It provided an interpretation of the prediction with AUROC values of 0.899 for breast and 0.736 for lung cancer patients. This study allows for the classification of patients at high risk for neutropenia and may assist in medical decisions before neutropenia occurs.