PREDICTION OF NEUTROPENIA IN CANCER PATIENTS RECEIVING CHEMOTHERAPY USING MACHINE LEARNING TECHNIQUE

Abstract

Neutropenia is a common complication in patients with cancer receiving chemotherapy. Without appropriate prevention or management, this condition can lead to severe complications and may be life-threatening. Previous studies revealed that machine learning techniques can be effectively predicted chemotherapy-related complications. The objective of this study was to develop and compare the optimal machine learning models for predicting neutropenia in cancer patients receiving chemotherapy. A retrospective dataset of 511 patients was collected from the hospital information system of Ratchaphiphat Hospital between January 1, 2022, and March 31, 2024. Data were derived from medical records, physical examination, and laboratory information, and were further screened to identify relevant variables prior to model development. The dataset was split into two groups; a training set (80%) and a testing set (20%). The training dataset was balanced using the SMOTE-NC method. Five machine learning techniques: Logistic Regression (LR), Random Forest (RF), Adaptive Boosting (AdaBoost), Support Vector Machine (SVM), and Artificial Neural Network (ANN) were developed using optimal hyperparameters. Model selection was based on accuracy, precision, recall, and F1-measure. Model performance was evaluated using the testing dataset, based on its sensitivity, specificity, and AUC. The results demonstrated that the best-performing models were those developed using RF and AdaBoost, with sensitivity, specificity, and AUC values of 100%, 98.21%, and 1.000, respectively. Therefore, RF and AdaBoost, which are ensemble learning techniques, were the most suitable models for predicting neutropenia.