Background: This study aimed to examine the relationship between microbiological features and mortality in hematological malignancy patients who develop febrile neutropenia using machine learning algorithms. Methods: Patients with hematological malignancies who developed febrile neutropenia between 2011 and 2015 in a training and research hospital were included. The PyCaret low-code Python library was used to streamline the machine-learning workflow. Two separate models were developed to predict early and late mortality. The following machine learning algorithms were evaluated during the modeling process: Ridge Classifier, Random Forest Classifier, Linear Discriminant Analysis, Light Gradient Boosting Machine, Logistic Regression, Gradient Boosting Classifier, and Extra Trees Classifier. Accuracy and area under the receiver operating characteristic curve (AUC-ROC) metrics were calculated to evaluate the models’ predictive capability for both early and late mortality predictions. All analyses were performed using Python 3.12 and the PyCaret 3.0 library. Results: The dataset used in this study consisted of 159 patients. For early mortality prediction, the Ridge Classifier demonstrated the best performance with a test set accuracy of 0.92 and an AUC of 0.94. For late mortality prediction, the Random Forest Classifier achieved the highest accuracy of 0.94 and an AUC of 0.98. For both models, ICU admission was identified as the most important feature, with a relative importance of 23.6% for early mortality prediction and 25.3% for late mortality prediction. Other key variables included pneumonia, renal function, and the duration of neutropenia. Conclusion: Machine learning models can be applied and improved on more patient data, helping traditional statistical methods in medical research.
Çalık et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: